Reference · 5 min read

AI

ChatGPT for Keyword Research: Use the Ideas, Check the Numbers

In this article 8
  1. What ChatGPT is good at in keyword research
  2. Where the numbers have to come from
  3. A routine that takes an afternoon
  4. What to ask and what not to ask
  5. Buyers now ask the assistant too
  6. When not to hire us
  7. Questions
  8. Sources

TL;DR the short answer

  1. ChatGPT is good at the part of keyword research that needs language: the words buyers use, the questions they ask and how those questions group into pages.
  2. On its own it has no keyword database, so a search volume in its answer is an estimate; the real numbers are in Google Search Console and Keyword Planner.
  3. Use ChatGPT first for ideas and last for sorting, and put a real data source in the middle.

Use ChatGPT for keyword research to find the words your buyers use, and never for the numbers. It writes a list of fifty phrases in a minute, and if you ask, it attaches a monthly search volume to each. On its own it has no keyword database, so those volumes are estimates written like facts, and the mistake that costs money is planning a year of articles on them.

Two cards joined by an arrow. Left: keyword ideas, from the chat. Right: search volume, from a data source such as Search Console or Keyword Planner. The idea and the number come from different places.

What ChatGPT is good at in keyword research

Keyword research has a language half and a data half. ChatGPT is strong in the first.

It knows how people phrase things. Give it your service page and ask how a plant manager would describe the same problem, and it returns the buyer’s words in place of yours. Owners write “process cooling solutions”; buyers type “chiller keeps tripping on high pressure”. Ask what someone wants to know before hiring an accountant in Toronto and you get the questions in the order of their worries, price first.

It also sorts. Paste two hundred queries from your own Search Console and ask it to group them by the question behind them. The groups are a first draft of your page list, and by hand that job takes hours.

Where the numbers have to come from

Google publishes how often a phrase is searched through Keyword Planner (Google Ads Help, 2026). The queries your own site already appears for are in the Search Console performance report (Search Console Help, 2026). Paid tools such as Ahrefs and Semrush measure the same things for any site.

ChatGPT is none of these. Unless you have connected a keyword tool to it, a volume in its answer comes from the language it learned, not from a count of searches.

TaskAsk ChatGPTCheck it in
The words buyers useYesYour own sales emails and call notes
Questions around a topicYesSearch Console queries
Monthly search volumeNoKeyword Planner or your keyword tool
Who ranks for the phrase nowNoThe search results page itself
Grouping queries into pagesYesYour judgment: one page per buyer question

The table has one rule in it. Anything that is a count gets checked against a source that counts.

A routine that takes an afternoon

  1. Export the queries your site already appears for from the Search Console performance report, for the longest period it offers.
  2. Paste them into ChatGPT with a short description of what you sell and to whom. Ask it to group the queries by the buyer’s question and to list the questions a buyer would ask that are missing.
  3. Check the missing ones in Keyword Planner or your keyword tool, with the location set to Canada. Drop the ones nobody searches.
  4. Choose by buyer, not by volume. A phrase searched rarely by purchasing managers is worth more than a phrase searched often by students.

The fourth step is where owners go wrong after the numbers are right. A big volume flatters. The phrase that brings a quote request is usually longer and plainer.

Keep the result in one sheet, the buyer’s question next to the page that answers it. That sheet is the content plan, and it is worth more than the keyword list it came from.

Four numbered steps in a row: export your queries, group them in the chat, check volume in a data source, choose by buyer. Only the second step happens in ChatGPT.

What to ask and what not to ask

A good prompt names the buyer and the moment. “Keywords for HVAC” returns a list anyone could write. “Questions a property manager of a 40-unit building asks before replacing a boiler” returns the phrases your page should answer.

Ask for the buyer’s wording, in the order a person would ask, and forbid marketing terms. Then ask which of your existing pages answers each question, and paste the page text so it can tell.

Run the same prompt twice. If the two lists differ a lot, the model is guessing at your market, and it needs more of your own material: a proposal or the questions customers email.

Do not ask it how hard a phrase is to rank for or how many clicks a page will get. Those answers need measurement, and you will get a confident paragraph in its place.

Buyers now ask the assistant too

Part of your market no longer types two words into Google. A buyer asks an AI assistant a full question: which firms in Toronto do this and what it should cost.

Those questions are longer than search queries and they describe a situation. You can use ChatGPT to draft them: ask what a buyer in your trade would ask an assistant before choosing a supplier. Then put the questions to the assistant yourself and read who gets named.

If your competitors are named and you are not, more keywords will not fix it. That is the question our AI Search Visibility audit is built to answer.

Questions we get

Can ChatGPT do keyword research?

It can do the language half: suggest the phrases a buyer would type, list the questions around a topic and group a long list of queries into pages. It cannot do the data half unless a keyword tool is connected to it.

Does ChatGPT know search volume?

Not on its own. ChatGPT has no keyword database, so a monthly volume in its answer is an estimate written like a fact. Check volumes in Keyword Planner (Google Ads Help, 2026) or in the keyword tool you already pay for.

What is a good ChatGPT prompt for keyword research?

Give it your service page and one buyer: 'Here is our page on industrial chillers. List 30 questions a plant manager in Ontario would type into Google before requesting a quote, in his words, no marketing terms.' Then check the list against real data.

Does ChatGPT replace Ahrefs or Semrush?

No. Those tools hold measured data on searches and rankings, and ChatGPT holds language. For a small site, Search Console and Keyword Planner are enough. Search Console is free, and Keyword Planner comes free with a Google Ads account.

Sources

  1. Google Ads Help, Use Keyword Planner 2026
  2. Search Console Help, Performance report (Search results) 2026
Related page AI Search Visibility See the service page

More on AI

More on
AI

One question per article, with a date and an author.

All articlesEvery article on AI, and the other topics.

All on AI →

Contact

Ask the question the article did not answer.

Or send a note.

We reply within one business day. No newsletters, no spam.