AI keyword research: using ChatGPT and AI tools to find keywords
Short answer Use AI for the slow word work: listing the ways customers describe a job, sorting phrases by intent and grouping them into pages. Don't trust its numbers. A chatbot can invent search volumes, and every SEO tool's volume is an estimate. Check the phrases in Search Console, Keyword Planner and autocomplete before you build a page.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callWhat AI does well in keyword research, and what to check
AI is fast with words and weak with facts. Give it the word jobs. Take every number from Google.
| Job | Hand it to AI? | Check it with |
|---|---|---|
| List how customers might phrase a service ("furnace repair", "furnace not turning on", "furnace making noise") | Yes. This is where it saves the most time | Google autocomplete: do people actually type these? |
| Sort phrases into "ready to hire", "comparing" and "learning" | Yes, mostly right | Your own judgment: you know which calls turn into jobs |
| Group 200 phrases into a handful of pages | Yes | One group per service you sell, not per phrase |
| Turn customer emails or call notes into search phrases | Yes, a strong use | Search Console: does your site already show for them? |
| Monthly search volume | No | Keyword Planner, as a range |
| How hard a phrase is to rank for | No | Look at who ranks on page one today |
| Which phrases will rise next year | No | Google Trends, which shows past interest only |
Google itself says AI fits the research step:
"Generative AI can be particularly useful when researching a topic, and to add structure to original content." (Google Search Central: Guidance on using generative AI content)
Why you can't trust ChatGPT's keyword numbers
Ask a chatbot for "the top 20 keywords with search volume" and it will often give you a neat table with numbers. Unless it pulled them from a real data source and shows where, those numbers may be made up. OpenAI calls this a hallucination and lists what it looks like:
"Incorrect definitions, dates, or facts" and "Overconfident answers to ambiguous or complex questions" (OpenAI Help Center: Does ChatGPT tell the truth?)
Its advice on the same page: "Always verify quotes, data, technical information or references to external documents." Treat any volume a chatbot gives you as a guess until a Google tool confirms it.
SEO tools' "AI" volumes are estimates too
Paid tools such as Ahrefs and Semrush give better numbers than a chatbot, but they are still modeled, not counted. Ahrefs says so plainly:
"Search volume in any SEO tool, including Ahrefs, is always an estimation. Only Google has access to the exact data." (Ahrefs Help Center)
Semrush says it uses machine learning and overlays third-party data with "historical clickstream data" (Semrush Knowledge Base). Clickstream means browsing data from a sample of users, then scaled up. For a big national phrase, the estimate is usually close enough. For "drywall repair" in a town of 30,000, the sample can be tiny, and two tools can disagree by a lot.
Reading Google's own numbers correctly
Google's figures are the closest to real, but each has a catch an AI summary often skips.
| Tool | What the number is | The catch |
|---|---|---|
| Search Console → Performance | Real searches where your site appeared, with clicks and impressions (Search Console Help) | Only covers searches you already show for |
| Keyword Planner "Avg. monthly searches" | Searches for the phrase "and its close variants", averaged over 12 months (Google Ads Help) | Google: "Your search volume statistics are rounded." Very small phrases may not appear at all (Google Ads Help) |
| Keyword Planner "Competition" | How many advertisers showed ads on the phrase | It is about ads, not about ranking (below) |
| Google Trends | Interest over time, including seasonal peaks | Relative, not counts (Google Trends Help) |
The "Competition" column trips up many owners, and many AI tools repeat the mistake. Google defines it this way:
"Competition: The number of advertisers that showed on each keyword relative to all keywords across Google." (Google Ads Help: About Keyword Planner metrics)
"Low competition" there means few advertisers, not that ranking for free will be easy. To judge ranking difficulty, search the phrase and look at who holds page one: other local businesses you can match, or national sites with years of content.
A 30-minute AI keyword routine
- Feed it your real words. Paste your service list and a dozen customer emails or call notes. Ask: "List the phrases a customer would type into Google to find these services. No numbers."
- Ask for intent. "Label each phrase: ready to hire, comparing options, or learning. Group them by the service they lead to."
- Check each group in autocomplete. Type the main phrase into Google and note the suggestions. Drop phrases nobody seems to type.
- Get ranges from Keyword Planner. Paste the surviving phrases. Keep low-volume buyer phrases anyway: a phrase too small to measure can still bring a paying customer.
- Compare with Search Console. Which phrases already show your site at positions 8–30? Those pages are the quickest wins.
- One page per service. Build or fix the page for each group. Google matches the variants on its own, so you don't need a page per phrase. For the page itself, see on-page SEO for service pages.
For the free tools in more depth, see keyword research for a local business.
Does AI search change keyword research?
The method stays the same; the questions get longer. Google says AI Mode helps "for queries where further exploration, reasoning, or complex comparisons are needed" (Google Search Central: AI features and your website). So add the full questions customers ask ("is it worth repairing a 15-year-old furnace?") next to the short phrases, and answer them on your service pages.
You can measure it in the same place. Google says sites shown in AI Overviews and AI Mode "are included in the overall search traffic in Search Console" and appear under the "Web" search type (same page). There is no separate AI keyword report.
Should you pay for an AI keyword tool?
Most small service businesses don't need one to start. The free Google tools cover a list of 10–30 services. A paid tool earns its cost when you want to see which phrases competitors rank for, or track hundreds of phrases across many cities. Even then, its volumes are estimates, so confirm the phrases that matter before you build pages around them.
Using AI to write the pages afterward is allowed, within limits. Google warns that generating "many pages without adding value for users" may break its scaled content abuse policy (Google Search Central). See does Google penalize AI content?.
Example
Illustrative cases with invented numbers.
A roofer (US dollars)
ChatGPT lists 60 phrases and claims "roof replacement cost" gets 12,000 searches a month in his city of 80,000. Keyword Planner shows a range of 100–1K. He keeps the phrase but sizes his expectations to the range, and builds one replacement page with his own price range.
A Toronto bookkeeper (Canadian dollars)
She pastes 15 client emails into an AI chatbot. It pulls out "bookkeeping for Airbnb hosts", a phrase she never had a page for. Keyword Planner shows no data. Search Console shows 40 impressions a month already. She writes the page and mentions her C$250-a-month starting fee.
A dentist comparing two tools
Two paid tools give 900 and 2,400 monthly searches for "emergency dentist" plus his city. He stops comparing and checks Search Console: his page gets 600 impressions a month at position 14. He works on that page first.
FAQ
Can ChatGPT do keyword research for free?
It can brainstorm phrases and group them for free. It can't tell you reliably how many people search them; use Keyword Planner and Search Console for that.
Which AI keyword tool is most accurate?
None is exact. Ahrefs says every SEO tool's volume is an estimate and only Google has the exact data. Google's own tools are the closest, and Search Console shows real searches for your site.
Is "low competition" in Keyword Planner good for SEO?
Not necessarily. Google defines it as the number of advertisers showing ads on the phrase, so it tells you about ads, not how hard ranking will be.
Can AI predict which keywords will trend?
Not reliably. Google Trends shows rising searches from the past, and Google's other tools average past months. Use them to spot growth that has already started.
Do I need different keywords for AI Overviews and ChatGPT?
No separate list. Add the full questions customers ask to your pages, and track Google's AI results inside the normal Search Console report.
Different for you?
- You're starting from zero with free tools: keyword research for a local business.
- You want to read your own search data: Google Search Console for a small business.
- You're wondering how much of SEO AI can take over: can AI do SEO?.
- You plan to write the pages with AI: does Google penalize AI content?.
- You'd rather have the research and pages done for you: SEO & GEO Agent.