How to Do Keyword Research With AI (ChatGPT Prompts + Method)

Choosing the right keywords means choosing the questions your website will answer. For a blog, an online store or a local contractor’s site, it’s often the difference between a page nobody finds and a page that brings in visitors for years. It’s also a tedious task: coming up with ideas, sorting them, grouping them, deciding where to start.

The good news: doing keyword research with AI saves a lot of time on all the thinking and organizing. The bad news: ChatGPT isn’t a search database, and it can give you numbers that look precise but are based on nothing. In this guide, I’ll show you a complete method, copy-and-paste prompts and, above all, how to combine AI with reliable sources so you don’t build your strategy on sand.

Contents

1. What AI can do, and what it can't
2. The 6-step method
3. A real-world example
4. How to spot a good keyword for a small site
5. Bonus prompts to go further
6. Mistakes to avoid
7. My checklist
8. Frequently asked questions
9. The bottom line

What AI can do, and what it can’t

Before you start, you need to be clear on who does what. I see far too many keyword tables “with volumes” produced by ChatGPT and used as is to decide on a content strategy. That’s risky.

Task	ChatGPT	Where to check
Finding ideas and phrasing variations	Very good	Google suggestions, your customers
Guessing an audience’s questions	Good starting point	Forums, reviews, questions you receive
Sorting by intent (info, comparison, purchase)	Very good	Actual Google results
Grouping into topics and pages	Very good	Your judgment
Giving search volumes	Unreliable	Data tools, Search Console
Estimating competition	Unreliable	Looking at the results
Prioritizing for your business	Good with your context	You

In my experience, the rule is simple: ChatGPT to think and organize, data sources to measure. OpenAI’s terms of use say as much: answers aren’t always accurate, should be evaluated before use and shouldn’t be the only basis for important decisions.

The 6-step method

Here’s the process I suggest. It works with the free version of ChatGPT as well as other assistants (see my comparison to help you choose an AI).

1. Describe your business and audience to the AI.
2. Generate a broad list of topics and phrasings.
3. Add real-world sources: Google suggestions, customer questions, Search Console.
4. Sort by search intent.
5. Group into pages so two articles don’t compete with each other.
6. Prioritize and build a content calendar.

Step 1: Give it context

Without context, ChatGPT will suggest generic keywords everyone is already targeting. With precise context, it becomes much more relevant.

I run [type of site] about [topic]. My audience: [who, level, location, main problems]. I offer [products, services or content]. My main competitors are [names, if you know them]. Before suggesting anything, ask me 5 questions to better understand what my audience is searching for.

Answer its questions: that’s often where you clarify your own positioning. If you regularly work on the same topic, a ChatGPT project lets you keep that context from one conversation to the next.

Step 2: Generate a broad list

At this stage, you want quantity and variety, not perfection.

Based on this context, list 40 searches someone in my audience might type into Google. Vary the phrasing: full questions ("how to…", "why…", "what's the best…"), short searches, comparisons ("X vs Y"), searches with a location, a budget or a level ("for beginners"). Write them the way real people would type them, in plain words. Don't include any volume or difficulty.

That last sentence matters: it keeps ChatGPT from adding made-up numbers that would give a false sense of precision.

You can then dig into a specific topic:

For the topic "[topic]", suggest 15 phrasing variations and 10 follow-up questions someone would ask once they've gotten the answer to their first question.

Step 3: Check against real-world sources

This is the step that turns a list of guesses into a usable list. Here are the sources I use, with no paid tools:

• Google autocomplete: type the beginning of a search and note what the search bar suggests.
• “People also ask” questions shown in the search results.
• Your Search Console if your site already exists: it shows the queries that actually bring in visitors.
• Your conversations with customers: emails, messages, questions asked over the phone or in store.
• Forums and reviews where your audience speaks in its own words.

Gather all of this in a doc or spreadsheet, then give it back to ChatGPT:

Here are two lists. List A comes from your suggestions. List B comes from real-world sources (Google suggestions, Search Console, my customers' questions). Merge them, remove duplicates and near-identical phrasings, and flag the items in list A that have no equivalent in list B: those are assumptions to check.

If you’re working with an exported file, my guide to analyzing a spreadsheet with AI in Excel can also help.

Step 4: Sort by intent

Useful keyword research isn’t just about words: it describes what the person wants. That’s one of ChatGPT’s strong suits.

Sort each search in this list by its main intent: "learn," "compare," "buy or book," "find a local provider," "solve a specific problem." Present the result in a table with three columns: search, intent, best-suited page format (guide, list, comparison, product page, tutorial). Flag searches whose intent is ambiguous.

For ambiguous cases, run the search yourself: the pages that show up first reveal which intent Google favors today.

Step 5: Group into pages

A classic mistake: writing an article for every variation. “Clean oven,” “how to clean an oven” and “oven cleaning hacks” deserve the same page, not three. Grouping keeps your own pages from competing with each other.

Group these searches into clusters that can be covered by one single page, because they share the same intent and call for the same answer. For each cluster, suggest: a main topic, the most representative search, the variations to work naturally into the text and the questions to cover in subheadings or an FAQ.

Then organize the whole thing into topics:

Based on these clusters, suggest a structure for my site with 4 to 6 main topics. For each one, name a main page and the supporting pages that link to it. Present the result as a simple site tree.

Step 6: Prioritize and plan

Not everything needs to be written right away. Give ChatGPT your criteria, and keep the final decision.

Here are my topic clusters. Help me prioritize them using these criteria: direct connection to my business (important), ability to share personal experience (important), seasonal urgency, ease of writing. Show a table with a score from 1 to 3 per criterion and a final ranking. Don't use any volume data: if you think a topic is searched a lot or a little, write "to check."

Then turn that ranking into a calendar. My article on planning your schedule with ChatGPT has useful prompts for this step, and my methods for finding content ideas with AI will help you spin each topic into several angles.

A real-world example

Take Sam, who rents kayaks in a coastal town (an illustrative case). He describes his business to ChatGPT: hourly rentals, guided tours, families and tourists.

ChatGPT gives him a list of searches: “kayak rental [town],” “kayaking with kids minimum age,” “beginner kayak tour,” “what to wear kayaking,” and so on. Sam compares it with Google suggestions and the questions customers ask him at the counter. He discovers that people very often ask whether they can bring a dog, a topic that wasn’t on the AI’s list.

He then has everything sorted and grouped: a local booking page, a “family kayaking” guide, a practical page on gear and safety rules, and an article on the most beautiful paddling routes. He starts with the family guide, because that’s where he has the most experience to share.

The AI didn’t give him the “right” answer: it helped him organize thinking that only he could validate.

How to spot a good keyword for a small site

Once the list is organized, you still have to judge each topic. ChatGPT can help you ask the right questions, but the answer comes from your knowledge of the field. Here are the criteria I look at first:

• The search matches what you offer. Can a visitor who arrives through this search become a loyal reader, customer or subscriber?
• You can do better than the current results. Look at the top pages: if they’re thorough, recent and written by specialists, look for a narrower angle.
• The phrasing is natural. A search nobody would say out loud is often an AI invention.
• The topic stays useful over time, or comes back every year in the same season.
• You have something to show: a photo, an example, an experience, a well-argued opinion.

You can ask ChatGPT to play devil’s advocate:

Here's the topic I plan to tackle first: "[topic]". Give me the 3 best reasons not to start with this one, and what I should check in Google's results to decide.

Bonus prompts to go further

Here's a page from my site and the main search I'm targeting: [search]. Does the page clearly answer this search in the first few lines? Which important variations or questions aren't covered?

Here are the searches that bring visitors to my site (Search Console export): [list]. Identify the topics I don't have a dedicated page for and the ones where several of my pages seem to answer the same thing.

Translate this list of searches into Spanish the way Spanish speakers in the US would naturally phrase them, not word for word. Flag cases where usage differs by region.

For that last one, always check with someone who speaks the language; my guide to translating with ChatGPT covers the precautions.

Mistakes to avoid

• Believing ChatGPT’s numbers. If it gives you a search volume or a difficulty score, treat it as a guess. It isn’t an official source for that data.
• Only going after big keywords. For a small site, specific searches and long questions are often easier to rank for and bring more qualified visitors, in my opinion.
• Forgetting intent. A keyword that attracts the merely curious does nothing for you if you sell a local service.
• Creating one page per variation. You spread your efforts thin and your pages compete with each other.
• Ignoring your customers’ words. Real phrasings beat the “marketing” phrasings AI suggests.
• Sharing sensitive data. If you paste a Search Console export or customer data, remove personal and confidential information. Check your account’s data settings.
• Stopping at the list. Keyword research is only worth something if it leads to genuinely useful pages. For that step, follow my method for using ChatGPT for SEO without producing mediocre content.

My checklist

• My audience and business are described precisely in the prompt.
• The AI’s list has been checked against at least two real-world sources.
• Every search is sorted by intent.
• Variations are grouped into pages, with no duplicates.
• No made-up volumes appear in my final table.
• Priorities take into account what I can bring personally.
• Each cluster matches an existing page or one planned in my calendar.

Frequently asked questions

Can ChatGPT give you a keyword’s search volume?

It can suggest one, but I’d advise you not to rely on it. ChatGPT isn’t a reliable source of volumes; use data tools or your site’s Search Console to measure.

Can AI replace a keyword research tool?

Not completely. It does a great job with part of the thinking, sorting and organizing, but not with measured data. The two complement each other.

Which AI should you use for keyword research?

ChatGPT, Gemini, Claude or Copilot can all help generate and sort ideas. Pick the one you’re comfortable with; my roundup of the best AI tools can point you in the right direction.

How many keywords should you target per page?

Rather than a number, think in clusters: one main search, its close variations and related questions, as long as they share the same intent. If the intent changes, plan another page.

Is the free version of ChatGPT enough?

For the method described here, the free version already covers the essentials. Usage limits can vary; I go over the differences in my guide to ChatGPT Free.

How often should you redo your keyword research?

In my practice, I revisit it when I’m preparing a new content calendar, and I keep an eye on Search Console regularly to spot new queries. There’s no universal rule.

The bottom line

To do keyword research with AI: give precise context, generate a broad list, check it against real-world sources, sort by intent, group into pages and prioritize based on what you can bring. Use ChatGPT to think and organize, never as a source of volumes.

To turn your keywords into useful pages, read my guide to using ChatGPT for SEO.