Ask any Head of Content or Editor who’s ever assigned an article to a freelance writer, and they’ll tell you that the trick is in the brief. A strong article brief is one that tells the writer exactly what the article’s purpose is, who’s in the target audience, what you hope to accomplish, what type of resources to use, and how to structure the content.
Good content leaders know that strong briefs are the best way to set themselves up for a smooth editorial process and publication success. Simply giving a writer a keyword or a blog post title to run with is akin to a dice roll. You might get the result you’re hoping for, but you’re more likely to get something completely different.
Perhaps for this reason, many content leads expressed strong interest in AI writing tools like Jasper, Copy AI, and Byword when they were first released. Maybe, they thought, these tools would speed things up!
But as it turns out, giving an AI tool a keyword and a tone of voice and saying “go” is even worse than doing that with a writer. Most of these AI tools can generate strong outlines for articles, and they can generate as much text as you ask them to, but depending on the tool’s training, most can’t research or synthesize well enough to produce a publishable article.
As Google has reminded us over and over again in the past few years, the content we produce must be helpful to readers in order to have success on channels like organic search. While that does mean we need to choose carefully what type of content is best for AI and programmatic SEO, it doesn’t mean we need to rule out AI entirely. We just have to be thoughtful about how we brief.
Whether you decide to use ChatGPT, Google’s Bard, or Bing, crafting strong AI prompts is just like writing a detailed brief for a writer — except you have to do it a little differently. You’re working with a robot, not a person, after all. Let’s dive in to how it’s done.
A single prompt for many articles
To effectively write a prompt for AI article generation, you must begin with an article concept that’s a good fit for AI. In our view, all AI content should be reproducible.
It’s a mistake to use AI to generate one-off content or taste-making content. AI is great, but it can’t conduct a detailed interview and it can’t form an opinion that a real, reasonable person would trust. So it’s not a good fit for content your brand might want to produce, like interviews with your team, deep dives into your product, or case studies of work you’ve done for your clients. Save that type of content for human writers.
Instead, AI is a great fit for content that follows a template. In other words — it’s a great fit for programmatic SEO, or the generation of multiple articles targeting a single head keyword and a modifier. Think,
best food for [PET]
hiking trails in [TOWN]
[CUISINE] recipes for beginners
In the above examples, the modifier is in brackets and the head keyword is in lowercase. Imagine that if you’re an outdoor supply retailer, you’d produce an article for each of the towns you have stores in, so outdoorsy folks in your target demographic who are looking for good hiking trails in those towns land on your website. You might start with writing a prompt for an AI article about hiking trails in San Francisco, but you could easily reuse that prompt for articles about hiking trails all over the world.
The good news is that this way of thinking makes prompt generation multi-purpose. Once you land on a solid prompt structure, you can reuse prompts for each article you generate and sub out keywords and data sources to keep them relevant.
The process of crafting a strong AI article prompt takes time, but if done well, it can be used several times over — saving you time in the end.
Research first, then prompt
It’s tempting to begin working with an AI tool with a desire to produce … something! Anything! Content!
Any content is better than no content for a company blog or knowledge base, right? That’s true to an extent, but when you approach an AI tool with uncertainty, you’re more likely to end up with thin, unhelpful content.
The best way to develop a productive prompt for AI article generation is to do a lot of pre-work before even turning to an AI tool. Your mantra must be “Research first, then prompt.”
Think about the outdoor supply retailer example from the above section — they want to produce articles about the best hiking trails in each town they have a store in. Why? Well, the theory is that outdoorsy folks in their target demographic who are looking for good hiking trails in those towns land on your website.
The content strategist in that example has to have done a lot of thinking before turning to an AI tool to produce the content. To land on their theory, they’ve likely done:
keyword research to confirm that people are searching for hiking trails in the towns where they have stores
competitor research to see what types of headers and content people are looking for in articles about the best hiking trails
outline crafting with their readers and customers’ particular pain points in mind, and a tie in to their outdoor supply store
source identification to make sure the AI tool has sources to read so the hiking trails their articles recommend aren’t just made up nonsense
Traditionally, content marketers do this work with a lot of thought and the help of a whole suite of tools. They use SEO-friendly tools like Ahrefs or SEMrush to do a bit of keyword and competitive research, identify top results in the SERPs for their most promising keywords to help with outline generation, and conduct interviews or use other, more traditional research strategies to identify promising sources. Then, they pass a detailed brief along to a writer, wait to get a first draft, and begin the editing process.
So far, AI tools haven’t done away with the need to do that work. In order to produce high-quality articles with the help of AI, you still need to do keyword and competitor research, craft an outline, and identify sources. Essentially, you need to brief your AI tool with as much information as you’d use to brief a writer. The work of keyword and competitor research, outline crafting, and sourcing is what content professionals learn to do from years of experience. And if you want a publishable article generated by AI, that work is necessary.
But AI tools can help with the two earliest stages of prompting for AI article generation: article ideation and outline generation.
1. Use GPT to help with keyword ideation, then use a keyword research tool to validate.
The old way: Use SEO-friendly tools like Ahrefs or SEMrush to do keyword research and determine exactly which keywords you should target and the user intent they serve.
The new way: Tell GPT about your business and ask it for ideas about what types of articles to write, then validate those ideas using a keyword research tool to make sure they have some search traffic (but your bar can be much lower if you’re using AI to generate the articles)
The example: Say you’re working on content marketing for a direct-to-consumer juice company that helps customers understand what kind of juice is most supportive for their health and wellness concerns and sends them a set of juices every week to drink with breakfast. You’d use the Open AI Playground to ask for article inspiration that meets your goal of teaching customers about the benefits of juice and attract people who are already searching for health and wellness ideas.
If I were the person responsible for marketing this juice company, I’d select the second option for a series of articles: “Top 5 Juice Ingredients for Energy Boost and Mental Clarity” Why? Because I can easily see how I’d make it into a series.
I could produce a single article on the top five juice ingredients for energy and mental clarity, but I could also produce articles on ingredients for focus, love, clear skin, muscle tone, and any number of ideas — and I’d easily be able to guide readers toward purchasing a juice subscription as an easy way to incorporate these ingredients into their routines.
Plus, I’d probably already have a bunch of information about juice ingredients, so source identification would be easy.
The next steps: Using GPT to identify an article idea took ideation from something that could take a few hours, to just a few minutes. The next step would be to list out all of the possible keywords you could target, like “ingredients for focus,” “ingredients for digestion,” “ingredients for clear skin,” etc. Then, plug those keywords into a keyword research tool to confirm they have search volume, make phrasing tweaks, and start looking at competitors.
If you don’t have a keyword research tool at your disposal, just Google your target keywords and see what comes up. Do the results resemble the type of article you’d like to produce? If so, you’re on the right track.
2. Ask GPT to analyze competitors and help you craft an outline.
The old way: Use SEO tools to analyze SERPs and do competitor research, then manually synthesize existing articles to establish the best possible outline for your article
The new way: Analyze SERPs, then feed those articles to GPT and ask it to put a competitive article outline together
The example: Let’s continue with the juice company example. I ran a quick Google search for “juice ingredients for focus” and found some promising results. The top four results were very simple articles that are ranking for this keyword, and Google’s inclusion of a “People also ask” box indicates that people are asking similar questions about other results juice can deliver.
If I were the content strategist marketing this DTC juice company, I’d do a little happy dance. We’ve landed on something people are searching for, and the competition is pretty thin.
Some people might toss this idea to the side, thinking the readers of these articles are likely looking to make juice, not buy juice. But think about what we can establish by exposing new readers to our product, showing them we know a lot about how to make juice, and introducing a simpler, faster option than making juice at home. Producing AI articles to target these keywords is low-effort and potentially high-value.
So, I plugged the competitor articles’ outlines into my Open AI Playground chat, using the following prompt:
Okay, now I want help producing an outline for an article titled, "Top 5 Juice Ingredients for Focus." I found the following header ideas. Please read them, analyze them, and write me an outline for my article.
HEADER IDEAS: What Causes Mental and Physical Fatigue?, Juice for Energy and Focus, How Can this Juice Recipe Boost Energy & Focus?, Apple Lemon Kale, Celery, and Spinach, Wrapping It Up, Ingredients:, How to Make:, Benefits of Green Juice for Brain Function
The result I got gave me a good starting place for my outline — with not only a few headers, but also a discussion of what can happen under each of those headers.
The next steps: This outline is a good start, but it needs some revision. We don’t want to copy the header from the first article we found, so I’d want to rewrite the header “What causes mental and physical fatigue?” To make the article reproducible, I’d probably write it as “What makes people lack focus?” That way, I could repurpose the header for articles about juices for other concerns, without needing to do much rewriting.
Revising your generated outline ensures that the article you produce will do the work of content marketing for your business — lead customers toward your product or service. You know your customers’ pain points and your product’s differentiators best, and working this into the outline isn’t something GPT can do for you.
The same is true for source identification. Because AI models like GPT can’t read the internet, you’ll need to find sources yourself — though some tools, like Bard and Bing, are getting closer to reading the internet and providing sources, so you could try using these tools to find good sources.
(In the future, daydream will be able to help with article ideation. For now, if you have questions about these steps, you can reach out to our team for support with the early stages of outline crafting and source identification.)
Write multiple prompts for each article, not just one
The best structure for an AI article prompt isn’t just one prompt — it’s several prompts.
That may sound counterintuitive. Most people think it’s smart to give an AI tool an article’s target keyword, tone of voice, outline, and talking points in one go. It’s like drafting a brief for a writer and hitting “send.” You give a writer a full assignment at once, right? It would be silly to ask a writer to draft a blog post header by header.
That’s the way most AI writing tools structure their workflows, too. You type in your keyword and tone of voice, get a generated article outline, and hit “write.”
The problem with this approach is that it takes too long a view of an article. It’s simply not detailed enough.
When a reader comes in contact with an article, they’re typically looking to learn something. (Duh!) Yes, they might read headers like an outline, but they also read each section of an article, looking for an answer.
Writers know this, and they know their job is to communicate their research and learning clearly, so they research the content that goes under each heading separately. Yes, they consider the article as a whole, but they also consider the information under each heading separately. And they might use a different source to tell the reader what they need to know in each section.
That means they give more thought to each section than a series of talking points. They also do research and compose each section like a mini-article.
The best approach to writing an article with AI as a helper is to give the AI tool both the long view of the article and the detailed view — and take each heading one-by-one. That means writing a bunch of linked prompts, not just one.
AI article prompting, step-by-step
Here’s a standard flow you can use to generate an article with multiple prompts:
Introduce the article’s title, purpose, and publication. Tell the tool what the article’s general structure should be — including headers — and ask it to write an introduction to the article.
Ask the tool to write the content that will go under each header, one at a time. Explain what the reader should know after reading the section and how long it should be. Give the AI tool any source information it should have to write each section. Don’t move on to the next section until you’re satisfied with the output. Tweak the prompt by giving the tool more information or providing it with an example.
When you’ve reached the end of your outline, tell the tool to summarize all of the above content in a few paragraphs. Give it a CTA based on the action you’d like readers to take. This will be your article’s conclusion.
When you write multiple prompts to generate a single article with an all-purpose tool like ChatGPT or Bing, you need to write the article in a linear order — each section one at a time.
If you use a tool like daydream, though, you can work on your prompts in any order and still generate the article all at once. And you can still go back to tweak each prompt depending on the output you get.
Simple, right? Just three steps and you’ve got an article!
We tease. Obviously, you’ll need to do a lot of pre-work before generating an article following this workflow.
It’s important to remember that the increased ease and speed of AI article writing comes when you replicate the process for multiple articles. (That’s programmatic SEO.) But AI isn’t a mind reader, and it needs a lot of structure and context to produce a usable result.
How long should an AI article prompt be?
When you use a multiple-prompt approach to producing a single article, each prompt can be pretty short. In fact, writing really long prompts for each section can confuse AI tools and give them too much information to work with.
Remember that as you generate content with AI, you can always clarify your instructions and see if that helps.
The exception to the “shorter is better” rule is context. We have no qualms about giving AI tools access to our own research by copying and pasting it and telling the tool to cite it as a source.
(Think, “You’ve already read this article from the New York Times and you can quote it in this section, as long as you put the information in quotes and cite your source as an article titled, ‘[TITLE]’ by [WRITER].”)
Because most AI models are not able to browse the internet (models like GPT cut off at 2021), you can’t just drop a link to an article and tell your model to read it. You can do this if you’re a customer of a purpose-built tool like daydream, but if you’re using a general-purpose tool, you’ll need to copy and paste all source information into your prompt, which makes it significantly longer.
How should you phrase AI article prompts?
Briefing AI tools is different from briefing real writers in that AI tools do not understand jokes or tone. It doesn’t help to use “please” and “thank you” with AI. AI needs direct and simple instructions (another reason taking article generation header-by-header is useful — it simplifies each request to one action item).
At daydream, when we first started experimenting with using AI to generate articles for programmatic SEO, we realized we were giving too many instructions per prompt and our outputs were suffering. Early outputs were long — and they sometimes included the context we’d provided verbatim. Whoops!
We learned that our AI model worked best with just one action verb per prompt. Even if we also asked the tool to “read” or “understand” in order to write an article section, we realized we needed to keep commands to “write” to a minimum to avoid confusion.
The formula for AI article prompts
While writing AI article prompts takes a lot of pre-work, the prompts themselves can actually be quite simple. Remember that you should write a separate prompt for each header.
If you’re using a tool like daydream, you can easily write a draft prompt for each header, review prompts one at a time or all at once, drop links to the sources you’d like to reference in each prompt, generate an article, and then tweak each prompt one-by-one — adding sources and context to each section.
Based on our experience, the prompt for each section of your article should include:
a description of the task at hand: “Now we’re going to write the content that goes under the header, [HEADER]” or “Let’s write an introductory paragraph for an article titled, [TITLE]”
references to what the tool should know: “As you write this section, refer to [SOURCE]” or “Use this [SOURCE] as an example”
questions the section should answer: “This section should answer the following questions” or “Answer each of these questions under a separate header within this section”
required formatting: “Include a bulleted list of examples” or “This section should have two sub-sections, titled [TITLE] and [TITLE]”
As long as each prompt includes these four items, most AI tools will follow the instructions and generate something you can actually use — not something you’ll need to rewrite later.
Write smarter and faster with AI and research
Using AI to write a publishable article doesn’t take the process of content briefing and editing from hours to minutes. But it can significantly speed up the content production process — especially when you want to produce multiple similar articles. You can re-use prompts and research to generate lots of articles much faster and more cheaply than you’d be able to otherwise.
Writing a strong prompt for the first article in a series requires thorough research and outline generation, a distinct prompt for each section, and iteration on phrasing and sourcing. But the prompts themselves can be simple, straightforward, and pretty short.
To learn more about using AI for programmatic SEO, subscribe to our newsletter. Or, to get started with programmatic SEO for your website, sign up for our waitlist.
Subscribe to our waitlist
Ask any Head of Content or Editor who’s ever assigned an article to a freelance writer, and they’ll tell you that the trick is in the brief. A strong article brief is one that tells the writer exactly what the article’s purpose is, who’s in the target audience, what you hope to accomplish, what type of resources to use, and how to structure the content.
Good content leaders know that strong briefs are the best way to set themselves up for a smooth editorial process and publication success. Simply giving a writer a keyword or a blog post title to run with is akin to a dice roll. You might get the result you’re hoping for, but you’re more likely to get something completely different.
Perhaps for this reason, many content leads expressed strong interest in AI writing tools like Jasper, Copy AI, and Byword when they were first released. Maybe, they thought, these tools would speed things up!
But as it turns out, giving an AI tool a keyword and a tone of voice and saying “go” is even worse than doing that with a writer. Most of these AI tools can generate strong outlines for articles, and they can generate as much text as you ask them to, but depending on the tool’s training, most can’t research or synthesize well enough to produce a publishable article.
As Google has reminded us over and over again in the past few years, the content we produce must be helpful to readers in order to have success on channels like organic search. While that does mean we need to choose carefully what type of content is best for AI and programmatic SEO, it doesn’t mean we need to rule out AI entirely. We just have to be thoughtful about how we brief.
Whether you decide to use ChatGPT, Google’s Bard, or Bing, crafting strong AI prompts is just like writing a detailed brief for a writer — except you have to do it a little differently. You’re working with a robot, not a person, after all. Let’s dive in to how it’s done.
A single prompt for many articles
To effectively write a prompt for AI article generation, you must begin with an article concept that’s a good fit for AI. In our view, all AI content should be reproducible.
It’s a mistake to use AI to generate one-off content or taste-making content. AI is great, but it can’t conduct a detailed interview and it can’t form an opinion that a real, reasonable person would trust. So it’s not a good fit for content your brand might want to produce, like interviews with your team, deep dives into your product, or case studies of work you’ve done for your clients. Save that type of content for human writers.
Instead, AI is a great fit for content that follows a template. In other words — it’s a great fit for programmatic SEO, or the generation of multiple articles targeting a single head keyword and a modifier. Think,
best food for [PET]
hiking trails in [TOWN]
[CUISINE] recipes for beginners
In the above examples, the modifier is in brackets and the head keyword is in lowercase. Imagine that if you’re an outdoor supply retailer, you’d produce an article for each of the towns you have stores in, so outdoorsy folks in your target demographic who are looking for good hiking trails in those towns land on your website. You might start with writing a prompt for an AI article about hiking trails in San Francisco, but you could easily reuse that prompt for articles about hiking trails all over the world.
The good news is that this way of thinking makes prompt generation multi-purpose. Once you land on a solid prompt structure, you can reuse prompts for each article you generate and sub out keywords and data sources to keep them relevant.
The process of crafting a strong AI article prompt takes time, but if done well, it can be used several times over — saving you time in the end.
Research first, then prompt
It’s tempting to begin working with an AI tool with a desire to produce … something! Anything! Content!
Any content is better than no content for a company blog or knowledge base, right? That’s true to an extent, but when you approach an AI tool with uncertainty, you’re more likely to end up with thin, unhelpful content.
The best way to develop a productive prompt for AI article generation is to do a lot of pre-work before even turning to an AI tool. Your mantra must be “Research first, then prompt.”
Think about the outdoor supply retailer example from the above section — they want to produce articles about the best hiking trails in each town they have a store in. Why? Well, the theory is that outdoorsy folks in their target demographic who are looking for good hiking trails in those towns land on your website.
The content strategist in that example has to have done a lot of thinking before turning to an AI tool to produce the content. To land on their theory, they’ve likely done:
keyword research to confirm that people are searching for hiking trails in the towns where they have stores
competitor research to see what types of headers and content people are looking for in articles about the best hiking trails
outline crafting with their readers and customers’ particular pain points in mind, and a tie in to their outdoor supply store
source identification to make sure the AI tool has sources to read so the hiking trails their articles recommend aren’t just made up nonsense
Traditionally, content marketers do this work with a lot of thought and the help of a whole suite of tools. They use SEO-friendly tools like Ahrefs or SEMrush to do a bit of keyword and competitive research, identify top results in the SERPs for their most promising keywords to help with outline generation, and conduct interviews or use other, more traditional research strategies to identify promising sources. Then, they pass a detailed brief along to a writer, wait to get a first draft, and begin the editing process.
So far, AI tools haven’t done away with the need to do that work. In order to produce high-quality articles with the help of AI, you still need to do keyword and competitor research, craft an outline, and identify sources. Essentially, you need to brief your AI tool with as much information as you’d use to brief a writer. The work of keyword and competitor research, outline crafting, and sourcing is what content professionals learn to do from years of experience. And if you want a publishable article generated by AI, that work is necessary.
But AI tools can help with the two earliest stages of prompting for AI article generation: article ideation and outline generation.
1. Use GPT to help with keyword ideation, then use a keyword research tool to validate.
The old way: Use SEO-friendly tools like Ahrefs or SEMrush to do keyword research and determine exactly which keywords you should target and the user intent they serve.
The new way: Tell GPT about your business and ask it for ideas about what types of articles to write, then validate those ideas using a keyword research tool to make sure they have some search traffic (but your bar can be much lower if you’re using AI to generate the articles)
The example: Say you’re working on content marketing for a direct-to-consumer juice company that helps customers understand what kind of juice is most supportive for their health and wellness concerns and sends them a set of juices every week to drink with breakfast. You’d use the Open AI Playground to ask for article inspiration that meets your goal of teaching customers about the benefits of juice and attract people who are already searching for health and wellness ideas.
If I were the person responsible for marketing this juice company, I’d select the second option for a series of articles: “Top 5 Juice Ingredients for Energy Boost and Mental Clarity” Why? Because I can easily see how I’d make it into a series.
I could produce a single article on the top five juice ingredients for energy and mental clarity, but I could also produce articles on ingredients for focus, love, clear skin, muscle tone, and any number of ideas — and I’d easily be able to guide readers toward purchasing a juice subscription as an easy way to incorporate these ingredients into their routines.
Plus, I’d probably already have a bunch of information about juice ingredients, so source identification would be easy.
The next steps: Using GPT to identify an article idea took ideation from something that could take a few hours, to just a few minutes. The next step would be to list out all of the possible keywords you could target, like “ingredients for focus,” “ingredients for digestion,” “ingredients for clear skin,” etc. Then, plug those keywords into a keyword research tool to confirm they have search volume, make phrasing tweaks, and start looking at competitors.
If you don’t have a keyword research tool at your disposal, just Google your target keywords and see what comes up. Do the results resemble the type of article you’d like to produce? If so, you’re on the right track.
2. Ask GPT to analyze competitors and help you craft an outline.
The old way: Use SEO tools to analyze SERPs and do competitor research, then manually synthesize existing articles to establish the best possible outline for your article
The new way: Analyze SERPs, then feed those articles to GPT and ask it to put a competitive article outline together
The example: Let’s continue with the juice company example. I ran a quick Google search for “juice ingredients for focus” and found some promising results. The top four results were very simple articles that are ranking for this keyword, and Google’s inclusion of a “People also ask” box indicates that people are asking similar questions about other results juice can deliver.
If I were the content strategist marketing this DTC juice company, I’d do a little happy dance. We’ve landed on something people are searching for, and the competition is pretty thin.
Some people might toss this idea to the side, thinking the readers of these articles are likely looking to make juice, not buy juice. But think about what we can establish by exposing new readers to our product, showing them we know a lot about how to make juice, and introducing a simpler, faster option than making juice at home. Producing AI articles to target these keywords is low-effort and potentially high-value.
So, I plugged the competitor articles’ outlines into my Open AI Playground chat, using the following prompt:
Okay, now I want help producing an outline for an article titled, "Top 5 Juice Ingredients for Focus." I found the following header ideas. Please read them, analyze them, and write me an outline for my article.
HEADER IDEAS: What Causes Mental and Physical Fatigue?, Juice for Energy and Focus, How Can this Juice Recipe Boost Energy & Focus?, Apple Lemon Kale, Celery, and Spinach, Wrapping It Up, Ingredients:, How to Make:, Benefits of Green Juice for Brain Function
The result I got gave me a good starting place for my outline — with not only a few headers, but also a discussion of what can happen under each of those headers.
The next steps: This outline is a good start, but it needs some revision. We don’t want to copy the header from the first article we found, so I’d want to rewrite the header “What causes mental and physical fatigue?” To make the article reproducible, I’d probably write it as “What makes people lack focus?” That way, I could repurpose the header for articles about juices for other concerns, without needing to do much rewriting.
Revising your generated outline ensures that the article you produce will do the work of content marketing for your business — lead customers toward your product or service. You know your customers’ pain points and your product’s differentiators best, and working this into the outline isn’t something GPT can do for you.
The same is true for source identification. Because AI models like GPT can’t read the internet, you’ll need to find sources yourself — though some tools, like Bard and Bing, are getting closer to reading the internet and providing sources, so you could try using these tools to find good sources.
(In the future, daydream will be able to help with article ideation. For now, if you have questions about these steps, you can reach out to our team for support with the early stages of outline crafting and source identification.)
Write multiple prompts for each article, not just one
The best structure for an AI article prompt isn’t just one prompt — it’s several prompts.
That may sound counterintuitive. Most people think it’s smart to give an AI tool an article’s target keyword, tone of voice, outline, and talking points in one go. It’s like drafting a brief for a writer and hitting “send.” You give a writer a full assignment at once, right? It would be silly to ask a writer to draft a blog post header by header.
That’s the way most AI writing tools structure their workflows, too. You type in your keyword and tone of voice, get a generated article outline, and hit “write.”
The problem with this approach is that it takes too long a view of an article. It’s simply not detailed enough.
When a reader comes in contact with an article, they’re typically looking to learn something. (Duh!) Yes, they might read headers like an outline, but they also read each section of an article, looking for an answer.
Writers know this, and they know their job is to communicate their research and learning clearly, so they research the content that goes under each heading separately. Yes, they consider the article as a whole, but they also consider the information under each heading separately. And they might use a different source to tell the reader what they need to know in each section.
That means they give more thought to each section than a series of talking points. They also do research and compose each section like a mini-article.
The best approach to writing an article with AI as a helper is to give the AI tool both the long view of the article and the detailed view — and take each heading one-by-one. That means writing a bunch of linked prompts, not just one.
AI article prompting, step-by-step
Here’s a standard flow you can use to generate an article with multiple prompts:
Introduce the article’s title, purpose, and publication. Tell the tool what the article’s general structure should be — including headers — and ask it to write an introduction to the article.
Ask the tool to write the content that will go under each header, one at a time. Explain what the reader should know after reading the section and how long it should be. Give the AI tool any source information it should have to write each section. Don’t move on to the next section until you’re satisfied with the output. Tweak the prompt by giving the tool more information or providing it with an example.
When you’ve reached the end of your outline, tell the tool to summarize all of the above content in a few paragraphs. Give it a CTA based on the action you’d like readers to take. This will be your article’s conclusion.
When you write multiple prompts to generate a single article with an all-purpose tool like ChatGPT or Bing, you need to write the article in a linear order — each section one at a time.
If you use a tool like daydream, though, you can work on your prompts in any order and still generate the article all at once. And you can still go back to tweak each prompt depending on the output you get.
Simple, right? Just three steps and you’ve got an article!
We tease. Obviously, you’ll need to do a lot of pre-work before generating an article following this workflow.
It’s important to remember that the increased ease and speed of AI article writing comes when you replicate the process for multiple articles. (That’s programmatic SEO.) But AI isn’t a mind reader, and it needs a lot of structure and context to produce a usable result.
How long should an AI article prompt be?
When you use a multiple-prompt approach to producing a single article, each prompt can be pretty short. In fact, writing really long prompts for each section can confuse AI tools and give them too much information to work with.
Remember that as you generate content with AI, you can always clarify your instructions and see if that helps.
The exception to the “shorter is better” rule is context. We have no qualms about giving AI tools access to our own research by copying and pasting it and telling the tool to cite it as a source.
(Think, “You’ve already read this article from the New York Times and you can quote it in this section, as long as you put the information in quotes and cite your source as an article titled, ‘[TITLE]’ by [WRITER].”)
Because most AI models are not able to browse the internet (models like GPT cut off at 2021), you can’t just drop a link to an article and tell your model to read it. You can do this if you’re a customer of a purpose-built tool like daydream, but if you’re using a general-purpose tool, you’ll need to copy and paste all source information into your prompt, which makes it significantly longer.
How should you phrase AI article prompts?
Briefing AI tools is different from briefing real writers in that AI tools do not understand jokes or tone. It doesn’t help to use “please” and “thank you” with AI. AI needs direct and simple instructions (another reason taking article generation header-by-header is useful — it simplifies each request to one action item).
At daydream, when we first started experimenting with using AI to generate articles for programmatic SEO, we realized we were giving too many instructions per prompt and our outputs were suffering. Early outputs were long — and they sometimes included the context we’d provided verbatim. Whoops!
We learned that our AI model worked best with just one action verb per prompt. Even if we also asked the tool to “read” or “understand” in order to write an article section, we realized we needed to keep commands to “write” to a minimum to avoid confusion.
The formula for AI article prompts
While writing AI article prompts takes a lot of pre-work, the prompts themselves can actually be quite simple. Remember that you should write a separate prompt for each header.
If you’re using a tool like daydream, you can easily write a draft prompt for each header, review prompts one at a time or all at once, drop links to the sources you’d like to reference in each prompt, generate an article, and then tweak each prompt one-by-one — adding sources and context to each section.
Based on our experience, the prompt for each section of your article should include:
a description of the task at hand: “Now we’re going to write the content that goes under the header, [HEADER]” or “Let’s write an introductory paragraph for an article titled, [TITLE]”
references to what the tool should know: “As you write this section, refer to [SOURCE]” or “Use this [SOURCE] as an example”
questions the section should answer: “This section should answer the following questions” or “Answer each of these questions under a separate header within this section”
required formatting: “Include a bulleted list of examples” or “This section should have two sub-sections, titled [TITLE] and [TITLE]”
As long as each prompt includes these four items, most AI tools will follow the instructions and generate something you can actually use — not something you’ll need to rewrite later.
Write smarter and faster with AI and research
Using AI to write a publishable article doesn’t take the process of content briefing and editing from hours to minutes. But it can significantly speed up the content production process — especially when you want to produce multiple similar articles. You can re-use prompts and research to generate lots of articles much faster and more cheaply than you’d be able to otherwise.
Writing a strong prompt for the first article in a series requires thorough research and outline generation, a distinct prompt for each section, and iteration on phrasing and sourcing. But the prompts themselves can be simple, straightforward, and pretty short.
To learn more about using AI for programmatic SEO, subscribe to our newsletter. Or, to get started with programmatic SEO for your website, sign up for our waitlist.
Subscribe to our waitlist
Ask any Head of Content or Editor who’s ever assigned an article to a freelance writer, and they’ll tell you that the trick is in the brief. A strong article brief is one that tells the writer exactly what the article’s purpose is, who’s in the target audience, what you hope to accomplish, what type of resources to use, and how to structure the content.
Good content leaders know that strong briefs are the best way to set themselves up for a smooth editorial process and publication success. Simply giving a writer a keyword or a blog post title to run with is akin to a dice roll. You might get the result you’re hoping for, but you’re more likely to get something completely different.
Perhaps for this reason, many content leads expressed strong interest in AI writing tools like Jasper, Copy AI, and Byword when they were first released. Maybe, they thought, these tools would speed things up!
But as it turns out, giving an AI tool a keyword and a tone of voice and saying “go” is even worse than doing that with a writer. Most of these AI tools can generate strong outlines for articles, and they can generate as much text as you ask them to, but depending on the tool’s training, most can’t research or synthesize well enough to produce a publishable article.
As Google has reminded us over and over again in the past few years, the content we produce must be helpful to readers in order to have success on channels like organic search. While that does mean we need to choose carefully what type of content is best for AI and programmatic SEO, it doesn’t mean we need to rule out AI entirely. We just have to be thoughtful about how we brief.
Whether you decide to use ChatGPT, Google’s Bard, or Bing, crafting strong AI prompts is just like writing a detailed brief for a writer — except you have to do it a little differently. You’re working with a robot, not a person, after all. Let’s dive in to how it’s done.
A single prompt for many articles
To effectively write a prompt for AI article generation, you must begin with an article concept that’s a good fit for AI. In our view, all AI content should be reproducible.
It’s a mistake to use AI to generate one-off content or taste-making content. AI is great, but it can’t conduct a detailed interview and it can’t form an opinion that a real, reasonable person would trust. So it’s not a good fit for content your brand might want to produce, like interviews with your team, deep dives into your product, or case studies of work you’ve done for your clients. Save that type of content for human writers.
Instead, AI is a great fit for content that follows a template. In other words — it’s a great fit for programmatic SEO, or the generation of multiple articles targeting a single head keyword and a modifier. Think,
best food for [PET]
hiking trails in [TOWN]
[CUISINE] recipes for beginners
In the above examples, the modifier is in brackets and the head keyword is in lowercase. Imagine that if you’re an outdoor supply retailer, you’d produce an article for each of the towns you have stores in, so outdoorsy folks in your target demographic who are looking for good hiking trails in those towns land on your website. You might start with writing a prompt for an AI article about hiking trails in San Francisco, but you could easily reuse that prompt for articles about hiking trails all over the world.
The good news is that this way of thinking makes prompt generation multi-purpose. Once you land on a solid prompt structure, you can reuse prompts for each article you generate and sub out keywords and data sources to keep them relevant.
The process of crafting a strong AI article prompt takes time, but if done well, it can be used several times over — saving you time in the end.
Research first, then prompt
It’s tempting to begin working with an AI tool with a desire to produce … something! Anything! Content!
Any content is better than no content for a company blog or knowledge base, right? That’s true to an extent, but when you approach an AI tool with uncertainty, you’re more likely to end up with thin, unhelpful content.
The best way to develop a productive prompt for AI article generation is to do a lot of pre-work before even turning to an AI tool. Your mantra must be “Research first, then prompt.”
Think about the outdoor supply retailer example from the above section — they want to produce articles about the best hiking trails in each town they have a store in. Why? Well, the theory is that outdoorsy folks in their target demographic who are looking for good hiking trails in those towns land on your website.
The content strategist in that example has to have done a lot of thinking before turning to an AI tool to produce the content. To land on their theory, they’ve likely done:
keyword research to confirm that people are searching for hiking trails in the towns where they have stores
competitor research to see what types of headers and content people are looking for in articles about the best hiking trails
outline crafting with their readers and customers’ particular pain points in mind, and a tie in to their outdoor supply store
source identification to make sure the AI tool has sources to read so the hiking trails their articles recommend aren’t just made up nonsense
Traditionally, content marketers do this work with a lot of thought and the help of a whole suite of tools. They use SEO-friendly tools like Ahrefs or SEMrush to do a bit of keyword and competitive research, identify top results in the SERPs for their most promising keywords to help with outline generation, and conduct interviews or use other, more traditional research strategies to identify promising sources. Then, they pass a detailed brief along to a writer, wait to get a first draft, and begin the editing process.
So far, AI tools haven’t done away with the need to do that work. In order to produce high-quality articles with the help of AI, you still need to do keyword and competitor research, craft an outline, and identify sources. Essentially, you need to brief your AI tool with as much information as you’d use to brief a writer. The work of keyword and competitor research, outline crafting, and sourcing is what content professionals learn to do from years of experience. And if you want a publishable article generated by AI, that work is necessary.
But AI tools can help with the two earliest stages of prompting for AI article generation: article ideation and outline generation.
1. Use GPT to help with keyword ideation, then use a keyword research tool to validate.
The old way: Use SEO-friendly tools like Ahrefs or SEMrush to do keyword research and determine exactly which keywords you should target and the user intent they serve.
The new way: Tell GPT about your business and ask it for ideas about what types of articles to write, then validate those ideas using a keyword research tool to make sure they have some search traffic (but your bar can be much lower if you’re using AI to generate the articles)
The example: Say you’re working on content marketing for a direct-to-consumer juice company that helps customers understand what kind of juice is most supportive for their health and wellness concerns and sends them a set of juices every week to drink with breakfast. You’d use the Open AI Playground to ask for article inspiration that meets your goal of teaching customers about the benefits of juice and attract people who are already searching for health and wellness ideas.
If I were the person responsible for marketing this juice company, I’d select the second option for a series of articles: “Top 5 Juice Ingredients for Energy Boost and Mental Clarity” Why? Because I can easily see how I’d make it into a series.
I could produce a single article on the top five juice ingredients for energy and mental clarity, but I could also produce articles on ingredients for focus, love, clear skin, muscle tone, and any number of ideas — and I’d easily be able to guide readers toward purchasing a juice subscription as an easy way to incorporate these ingredients into their routines.
Plus, I’d probably already have a bunch of information about juice ingredients, so source identification would be easy.
The next steps: Using GPT to identify an article idea took ideation from something that could take a few hours, to just a few minutes. The next step would be to list out all of the possible keywords you could target, like “ingredients for focus,” “ingredients for digestion,” “ingredients for clear skin,” etc. Then, plug those keywords into a keyword research tool to confirm they have search volume, make phrasing tweaks, and start looking at competitors.
If you don’t have a keyword research tool at your disposal, just Google your target keywords and see what comes up. Do the results resemble the type of article you’d like to produce? If so, you’re on the right track.
2. Ask GPT to analyze competitors and help you craft an outline.
The old way: Use SEO tools to analyze SERPs and do competitor research, then manually synthesize existing articles to establish the best possible outline for your article
The new way: Analyze SERPs, then feed those articles to GPT and ask it to put a competitive article outline together
The example: Let’s continue with the juice company example. I ran a quick Google search for “juice ingredients for focus” and found some promising results. The top four results were very simple articles that are ranking for this keyword, and Google’s inclusion of a “People also ask” box indicates that people are asking similar questions about other results juice can deliver.
If I were the content strategist marketing this DTC juice company, I’d do a little happy dance. We’ve landed on something people are searching for, and the competition is pretty thin.
Some people might toss this idea to the side, thinking the readers of these articles are likely looking to make juice, not buy juice. But think about what we can establish by exposing new readers to our product, showing them we know a lot about how to make juice, and introducing a simpler, faster option than making juice at home. Producing AI articles to target these keywords is low-effort and potentially high-value.
So, I plugged the competitor articles’ outlines into my Open AI Playground chat, using the following prompt:
Okay, now I want help producing an outline for an article titled, "Top 5 Juice Ingredients for Focus." I found the following header ideas. Please read them, analyze them, and write me an outline for my article.
HEADER IDEAS: What Causes Mental and Physical Fatigue?, Juice for Energy and Focus, How Can this Juice Recipe Boost Energy & Focus?, Apple Lemon Kale, Celery, and Spinach, Wrapping It Up, Ingredients:, How to Make:, Benefits of Green Juice for Brain Function
The result I got gave me a good starting place for my outline — with not only a few headers, but also a discussion of what can happen under each of those headers.
The next steps: This outline is a good start, but it needs some revision. We don’t want to copy the header from the first article we found, so I’d want to rewrite the header “What causes mental and physical fatigue?” To make the article reproducible, I’d probably write it as “What makes people lack focus?” That way, I could repurpose the header for articles about juices for other concerns, without needing to do much rewriting.
Revising your generated outline ensures that the article you produce will do the work of content marketing for your business — lead customers toward your product or service. You know your customers’ pain points and your product’s differentiators best, and working this into the outline isn’t something GPT can do for you.
The same is true for source identification. Because AI models like GPT can’t read the internet, you’ll need to find sources yourself — though some tools, like Bard and Bing, are getting closer to reading the internet and providing sources, so you could try using these tools to find good sources.
(In the future, daydream will be able to help with article ideation. For now, if you have questions about these steps, you can reach out to our team for support with the early stages of outline crafting and source identification.)
Write multiple prompts for each article, not just one
The best structure for an AI article prompt isn’t just one prompt — it’s several prompts.
That may sound counterintuitive. Most people think it’s smart to give an AI tool an article’s target keyword, tone of voice, outline, and talking points in one go. It’s like drafting a brief for a writer and hitting “send.” You give a writer a full assignment at once, right? It would be silly to ask a writer to draft a blog post header by header.
That’s the way most AI writing tools structure their workflows, too. You type in your keyword and tone of voice, get a generated article outline, and hit “write.”
The problem with this approach is that it takes too long a view of an article. It’s simply not detailed enough.
When a reader comes in contact with an article, they’re typically looking to learn something. (Duh!) Yes, they might read headers like an outline, but they also read each section of an article, looking for an answer.
Writers know this, and they know their job is to communicate their research and learning clearly, so they research the content that goes under each heading separately. Yes, they consider the article as a whole, but they also consider the information under each heading separately. And they might use a different source to tell the reader what they need to know in each section.
That means they give more thought to each section than a series of talking points. They also do research and compose each section like a mini-article.
The best approach to writing an article with AI as a helper is to give the AI tool both the long view of the article and the detailed view — and take each heading one-by-one. That means writing a bunch of linked prompts, not just one.
AI article prompting, step-by-step
Here’s a standard flow you can use to generate an article with multiple prompts:
Introduce the article’s title, purpose, and publication. Tell the tool what the article’s general structure should be — including headers — and ask it to write an introduction to the article.
Ask the tool to write the content that will go under each header, one at a time. Explain what the reader should know after reading the section and how long it should be. Give the AI tool any source information it should have to write each section. Don’t move on to the next section until you’re satisfied with the output. Tweak the prompt by giving the tool more information or providing it with an example.
When you’ve reached the end of your outline, tell the tool to summarize all of the above content in a few paragraphs. Give it a CTA based on the action you’d like readers to take. This will be your article’s conclusion.
When you write multiple prompts to generate a single article with an all-purpose tool like ChatGPT or Bing, you need to write the article in a linear order — each section one at a time.
If you use a tool like daydream, though, you can work on your prompts in any order and still generate the article all at once. And you can still go back to tweak each prompt depending on the output you get.
Simple, right? Just three steps and you’ve got an article!
We tease. Obviously, you’ll need to do a lot of pre-work before generating an article following this workflow.
It’s important to remember that the increased ease and speed of AI article writing comes when you replicate the process for multiple articles. (That’s programmatic SEO.) But AI isn’t a mind reader, and it needs a lot of structure and context to produce a usable result.
How long should an AI article prompt be?
When you use a multiple-prompt approach to producing a single article, each prompt can be pretty short. In fact, writing really long prompts for each section can confuse AI tools and give them too much information to work with.
Remember that as you generate content with AI, you can always clarify your instructions and see if that helps.
The exception to the “shorter is better” rule is context. We have no qualms about giving AI tools access to our own research by copying and pasting it and telling the tool to cite it as a source.
(Think, “You’ve already read this article from the New York Times and you can quote it in this section, as long as you put the information in quotes and cite your source as an article titled, ‘[TITLE]’ by [WRITER].”)
Because most AI models are not able to browse the internet (models like GPT cut off at 2021), you can’t just drop a link to an article and tell your model to read it. You can do this if you’re a customer of a purpose-built tool like daydream, but if you’re using a general-purpose tool, you’ll need to copy and paste all source information into your prompt, which makes it significantly longer.
How should you phrase AI article prompts?
Briefing AI tools is different from briefing real writers in that AI tools do not understand jokes or tone. It doesn’t help to use “please” and “thank you” with AI. AI needs direct and simple instructions (another reason taking article generation header-by-header is useful — it simplifies each request to one action item).
At daydream, when we first started experimenting with using AI to generate articles for programmatic SEO, we realized we were giving too many instructions per prompt and our outputs were suffering. Early outputs were long — and they sometimes included the context we’d provided verbatim. Whoops!
We learned that our AI model worked best with just one action verb per prompt. Even if we also asked the tool to “read” or “understand” in order to write an article section, we realized we needed to keep commands to “write” to a minimum to avoid confusion.
The formula for AI article prompts
While writing AI article prompts takes a lot of pre-work, the prompts themselves can actually be quite simple. Remember that you should write a separate prompt for each header.
If you’re using a tool like daydream, you can easily write a draft prompt for each header, review prompts one at a time or all at once, drop links to the sources you’d like to reference in each prompt, generate an article, and then tweak each prompt one-by-one — adding sources and context to each section.
Based on our experience, the prompt for each section of your article should include:
a description of the task at hand: “Now we’re going to write the content that goes under the header, [HEADER]” or “Let’s write an introductory paragraph for an article titled, [TITLE]”
references to what the tool should know: “As you write this section, refer to [SOURCE]” or “Use this [SOURCE] as an example”
questions the section should answer: “This section should answer the following questions” or “Answer each of these questions under a separate header within this section”
required formatting: “Include a bulleted list of examples” or “This section should have two sub-sections, titled [TITLE] and [TITLE]”
As long as each prompt includes these four items, most AI tools will follow the instructions and generate something you can actually use — not something you’ll need to rewrite later.
Write smarter and faster with AI and research
Using AI to write a publishable article doesn’t take the process of content briefing and editing from hours to minutes. But it can significantly speed up the content production process — especially when you want to produce multiple similar articles. You can re-use prompts and research to generate lots of articles much faster and more cheaply than you’d be able to otherwise.
Writing a strong prompt for the first article in a series requires thorough research and outline generation, a distinct prompt for each section, and iteration on phrasing and sourcing. But the prompts themselves can be simple, straightforward, and pretty short.
To learn more about using AI for programmatic SEO, subscribe to our newsletter. Or, to get started with programmatic SEO for your website, sign up for our waitlist.
Subscribe to our waitlist
Ask any Head of Content or Editor who’s ever assigned an article to a freelance writer, and they’ll tell you that the trick is in the brief. A strong article brief is one that tells the writer exactly what the article’s purpose is, who’s in the target audience, what you hope to accomplish, what type of resources to use, and how to structure the content.
Good content leaders know that strong briefs are the best way to set themselves up for a smooth editorial process and publication success. Simply giving a writer a keyword or a blog post title to run with is akin to a dice roll. You might get the result you’re hoping for, but you’re more likely to get something completely different.
Perhaps for this reason, many content leads expressed strong interest in AI writing tools like Jasper, Copy AI, and Byword when they were first released. Maybe, they thought, these tools would speed things up!
But as it turns out, giving an AI tool a keyword and a tone of voice and saying “go” is even worse than doing that with a writer. Most of these AI tools can generate strong outlines for articles, and they can generate as much text as you ask them to, but depending on the tool’s training, most can’t research or synthesize well enough to produce a publishable article.
As Google has reminded us over and over again in the past few years, the content we produce must be helpful to readers in order to have success on channels like organic search. While that does mean we need to choose carefully what type of content is best for AI and programmatic SEO, it doesn’t mean we need to rule out AI entirely. We just have to be thoughtful about how we brief.
Whether you decide to use ChatGPT, Google’s Bard, or Bing, crafting strong AI prompts is just like writing a detailed brief for a writer — except you have to do it a little differently. You’re working with a robot, not a person, after all. Let’s dive in to how it’s done.
A single prompt for many articles
To effectively write a prompt for AI article generation, you must begin with an article concept that’s a good fit for AI. In our view, all AI content should be reproducible.
It’s a mistake to use AI to generate one-off content or taste-making content. AI is great, but it can’t conduct a detailed interview and it can’t form an opinion that a real, reasonable person would trust. So it’s not a good fit for content your brand might want to produce, like interviews with your team, deep dives into your product, or case studies of work you’ve done for your clients. Save that type of content for human writers.
Instead, AI is a great fit for content that follows a template. In other words — it’s a great fit for programmatic SEO, or the generation of multiple articles targeting a single head keyword and a modifier. Think,
best food for [PET]
hiking trails in [TOWN]
[CUISINE] recipes for beginners
In the above examples, the modifier is in brackets and the head keyword is in lowercase. Imagine that if you’re an outdoor supply retailer, you’d produce an article for each of the towns you have stores in, so outdoorsy folks in your target demographic who are looking for good hiking trails in those towns land on your website. You might start with writing a prompt for an AI article about hiking trails in San Francisco, but you could easily reuse that prompt for articles about hiking trails all over the world.
The good news is that this way of thinking makes prompt generation multi-purpose. Once you land on a solid prompt structure, you can reuse prompts for each article you generate and sub out keywords and data sources to keep them relevant.
The process of crafting a strong AI article prompt takes time, but if done well, it can be used several times over — saving you time in the end.
Research first, then prompt
It’s tempting to begin working with an AI tool with a desire to produce … something! Anything! Content!
Any content is better than no content for a company blog or knowledge base, right? That’s true to an extent, but when you approach an AI tool with uncertainty, you’re more likely to end up with thin, unhelpful content.
The best way to develop a productive prompt for AI article generation is to do a lot of pre-work before even turning to an AI tool. Your mantra must be “Research first, then prompt.”
Think about the outdoor supply retailer example from the above section — they want to produce articles about the best hiking trails in each town they have a store in. Why? Well, the theory is that outdoorsy folks in their target demographic who are looking for good hiking trails in those towns land on your website.
The content strategist in that example has to have done a lot of thinking before turning to an AI tool to produce the content. To land on their theory, they’ve likely done:
keyword research to confirm that people are searching for hiking trails in the towns where they have stores
competitor research to see what types of headers and content people are looking for in articles about the best hiking trails
outline crafting with their readers and customers’ particular pain points in mind, and a tie in to their outdoor supply store
source identification to make sure the AI tool has sources to read so the hiking trails their articles recommend aren’t just made up nonsense
Traditionally, content marketers do this work with a lot of thought and the help of a whole suite of tools. They use SEO-friendly tools like Ahrefs or SEMrush to do a bit of keyword and competitive research, identify top results in the SERPs for their most promising keywords to help with outline generation, and conduct interviews or use other, more traditional research strategies to identify promising sources. Then, they pass a detailed brief along to a writer, wait to get a first draft, and begin the editing process.
So far, AI tools haven’t done away with the need to do that work. In order to produce high-quality articles with the help of AI, you still need to do keyword and competitor research, craft an outline, and identify sources. Essentially, you need to brief your AI tool with as much information as you’d use to brief a writer. The work of keyword and competitor research, outline crafting, and sourcing is what content professionals learn to do from years of experience. And if you want a publishable article generated by AI, that work is necessary.
But AI tools can help with the two earliest stages of prompting for AI article generation: article ideation and outline generation.
1. Use GPT to help with keyword ideation, then use a keyword research tool to validate.
The old way: Use SEO-friendly tools like Ahrefs or SEMrush to do keyword research and determine exactly which keywords you should target and the user intent they serve.
The new way: Tell GPT about your business and ask it for ideas about what types of articles to write, then validate those ideas using a keyword research tool to make sure they have some search traffic (but your bar can be much lower if you’re using AI to generate the articles)
The example: Say you’re working on content marketing for a direct-to-consumer juice company that helps customers understand what kind of juice is most supportive for their health and wellness concerns and sends them a set of juices every week to drink with breakfast. You’d use the Open AI Playground to ask for article inspiration that meets your goal of teaching customers about the benefits of juice and attract people who are already searching for health and wellness ideas.
If I were the person responsible for marketing this juice company, I’d select the second option for a series of articles: “Top 5 Juice Ingredients for Energy Boost and Mental Clarity” Why? Because I can easily see how I’d make it into a series.
I could produce a single article on the top five juice ingredients for energy and mental clarity, but I could also produce articles on ingredients for focus, love, clear skin, muscle tone, and any number of ideas — and I’d easily be able to guide readers toward purchasing a juice subscription as an easy way to incorporate these ingredients into their routines.
Plus, I’d probably already have a bunch of information about juice ingredients, so source identification would be easy.
The next steps: Using GPT to identify an article idea took ideation from something that could take a few hours, to just a few minutes. The next step would be to list out all of the possible keywords you could target, like “ingredients for focus,” “ingredients for digestion,” “ingredients for clear skin,” etc. Then, plug those keywords into a keyword research tool to confirm they have search volume, make phrasing tweaks, and start looking at competitors.
If you don’t have a keyword research tool at your disposal, just Google your target keywords and see what comes up. Do the results resemble the type of article you’d like to produce? If so, you’re on the right track.
2. Ask GPT to analyze competitors and help you craft an outline.
The old way: Use SEO tools to analyze SERPs and do competitor research, then manually synthesize existing articles to establish the best possible outline for your article
The new way: Analyze SERPs, then feed those articles to GPT and ask it to put a competitive article outline together
The example: Let’s continue with the juice company example. I ran a quick Google search for “juice ingredients for focus” and found some promising results. The top four results were very simple articles that are ranking for this keyword, and Google’s inclusion of a “People also ask” box indicates that people are asking similar questions about other results juice can deliver.
If I were the content strategist marketing this DTC juice company, I’d do a little happy dance. We’ve landed on something people are searching for, and the competition is pretty thin.
Some people might toss this idea to the side, thinking the readers of these articles are likely looking to make juice, not buy juice. But think about what we can establish by exposing new readers to our product, showing them we know a lot about how to make juice, and introducing a simpler, faster option than making juice at home. Producing AI articles to target these keywords is low-effort and potentially high-value.
So, I plugged the competitor articles’ outlines into my Open AI Playground chat, using the following prompt:
Okay, now I want help producing an outline for an article titled, "Top 5 Juice Ingredients for Focus." I found the following header ideas. Please read them, analyze them, and write me an outline for my article.
HEADER IDEAS: What Causes Mental and Physical Fatigue?, Juice for Energy and Focus, How Can this Juice Recipe Boost Energy & Focus?, Apple Lemon Kale, Celery, and Spinach, Wrapping It Up, Ingredients:, How to Make:, Benefits of Green Juice for Brain Function
The result I got gave me a good starting place for my outline — with not only a few headers, but also a discussion of what can happen under each of those headers.
The next steps: This outline is a good start, but it needs some revision. We don’t want to copy the header from the first article we found, so I’d want to rewrite the header “What causes mental and physical fatigue?” To make the article reproducible, I’d probably write it as “What makes people lack focus?” That way, I could repurpose the header for articles about juices for other concerns, without needing to do much rewriting.
Revising your generated outline ensures that the article you produce will do the work of content marketing for your business — lead customers toward your product or service. You know your customers’ pain points and your product’s differentiators best, and working this into the outline isn’t something GPT can do for you.
The same is true for source identification. Because AI models like GPT can’t read the internet, you’ll need to find sources yourself — though some tools, like Bard and Bing, are getting closer to reading the internet and providing sources, so you could try using these tools to find good sources.
(In the future, daydream will be able to help with article ideation. For now, if you have questions about these steps, you can reach out to our team for support with the early stages of outline crafting and source identification.)
Write multiple prompts for each article, not just one
The best structure for an AI article prompt isn’t just one prompt — it’s several prompts.
That may sound counterintuitive. Most people think it’s smart to give an AI tool an article’s target keyword, tone of voice, outline, and talking points in one go. It’s like drafting a brief for a writer and hitting “send.” You give a writer a full assignment at once, right? It would be silly to ask a writer to draft a blog post header by header.
That’s the way most AI writing tools structure their workflows, too. You type in your keyword and tone of voice, get a generated article outline, and hit “write.”
The problem with this approach is that it takes too long a view of an article. It’s simply not detailed enough.
When a reader comes in contact with an article, they’re typically looking to learn something. (Duh!) Yes, they might read headers like an outline, but they also read each section of an article, looking for an answer.
Writers know this, and they know their job is to communicate their research and learning clearly, so they research the content that goes under each heading separately. Yes, they consider the article as a whole, but they also consider the information under each heading separately. And they might use a different source to tell the reader what they need to know in each section.
That means they give more thought to each section than a series of talking points. They also do research and compose each section like a mini-article.
The best approach to writing an article with AI as a helper is to give the AI tool both the long view of the article and the detailed view — and take each heading one-by-one. That means writing a bunch of linked prompts, not just one.
AI article prompting, step-by-step
Here’s a standard flow you can use to generate an article with multiple prompts:
Introduce the article’s title, purpose, and publication. Tell the tool what the article’s general structure should be — including headers — and ask it to write an introduction to the article.
Ask the tool to write the content that will go under each header, one at a time. Explain what the reader should know after reading the section and how long it should be. Give the AI tool any source information it should have to write each section. Don’t move on to the next section until you’re satisfied with the output. Tweak the prompt by giving the tool more information or providing it with an example.
When you’ve reached the end of your outline, tell the tool to summarize all of the above content in a few paragraphs. Give it a CTA based on the action you’d like readers to take. This will be your article’s conclusion.
When you write multiple prompts to generate a single article with an all-purpose tool like ChatGPT or Bing, you need to write the article in a linear order — each section one at a time.
If you use a tool like daydream, though, you can work on your prompts in any order and still generate the article all at once. And you can still go back to tweak each prompt depending on the output you get.
Simple, right? Just three steps and you’ve got an article!
We tease. Obviously, you’ll need to do a lot of pre-work before generating an article following this workflow.
It’s important to remember that the increased ease and speed of AI article writing comes when you replicate the process for multiple articles. (That’s programmatic SEO.) But AI isn’t a mind reader, and it needs a lot of structure and context to produce a usable result.
How long should an AI article prompt be?
When you use a multiple-prompt approach to producing a single article, each prompt can be pretty short. In fact, writing really long prompts for each section can confuse AI tools and give them too much information to work with.
Remember that as you generate content with AI, you can always clarify your instructions and see if that helps.
The exception to the “shorter is better” rule is context. We have no qualms about giving AI tools access to our own research by copying and pasting it and telling the tool to cite it as a source.
(Think, “You’ve already read this article from the New York Times and you can quote it in this section, as long as you put the information in quotes and cite your source as an article titled, ‘[TITLE]’ by [WRITER].”)
Because most AI models are not able to browse the internet (models like GPT cut off at 2021), you can’t just drop a link to an article and tell your model to read it. You can do this if you’re a customer of a purpose-built tool like daydream, but if you’re using a general-purpose tool, you’ll need to copy and paste all source information into your prompt, which makes it significantly longer.
How should you phrase AI article prompts?
Briefing AI tools is different from briefing real writers in that AI tools do not understand jokes or tone. It doesn’t help to use “please” and “thank you” with AI. AI needs direct and simple instructions (another reason taking article generation header-by-header is useful — it simplifies each request to one action item).
At daydream, when we first started experimenting with using AI to generate articles for programmatic SEO, we realized we were giving too many instructions per prompt and our outputs were suffering. Early outputs were long — and they sometimes included the context we’d provided verbatim. Whoops!
We learned that our AI model worked best with just one action verb per prompt. Even if we also asked the tool to “read” or “understand” in order to write an article section, we realized we needed to keep commands to “write” to a minimum to avoid confusion.
The formula for AI article prompts
While writing AI article prompts takes a lot of pre-work, the prompts themselves can actually be quite simple. Remember that you should write a separate prompt for each header.
If you’re using a tool like daydream, you can easily write a draft prompt for each header, review prompts one at a time or all at once, drop links to the sources you’d like to reference in each prompt, generate an article, and then tweak each prompt one-by-one — adding sources and context to each section.
Based on our experience, the prompt for each section of your article should include:
a description of the task at hand: “Now we’re going to write the content that goes under the header, [HEADER]” or “Let’s write an introductory paragraph for an article titled, [TITLE]”
references to what the tool should know: “As you write this section, refer to [SOURCE]” or “Use this [SOURCE] as an example”
questions the section should answer: “This section should answer the following questions” or “Answer each of these questions under a separate header within this section”
required formatting: “Include a bulleted list of examples” or “This section should have two sub-sections, titled [TITLE] and [TITLE]”
As long as each prompt includes these four items, most AI tools will follow the instructions and generate something you can actually use — not something you’ll need to rewrite later.
Write smarter and faster with AI and research
Using AI to write a publishable article doesn’t take the process of content briefing and editing from hours to minutes. But it can significantly speed up the content production process — especially when you want to produce multiple similar articles. You can re-use prompts and research to generate lots of articles much faster and more cheaply than you’d be able to otherwise.
Writing a strong prompt for the first article in a series requires thorough research and outline generation, a distinct prompt for each section, and iteration on phrasing and sourcing. But the prompts themselves can be simple, straightforward, and pretty short.
To learn more about using AI for programmatic SEO, subscribe to our newsletter. Or, to get started with programmatic SEO for your website, sign up for our waitlist.