Keyword research with AI: do it in an hour, without an expensive tool

You do keyword research with AI in five steps. You collect your customers’ questions, let AI expand and group them, get the numbers from Google’s free tools and choose one main question per page. AI takes over the thinking and sorting. The numbers come from somewhere else. Mix those up and you build your site on made-up figures.

Classic keyword research by an agency easily costs a few hundred to a thousand dollars and delivers a spreadsheet with hundreds of rows. For a large online store that is useful. For a business with thirty pages it is too much. You need twenty to fifty good queries, and you need to know which page belongs to which question.

What AI can and cannot do with keywords

Task	Can AI do it?	Notes
Coming up with questions customers ask	Yes, well	Especially if you describe your audience
Grouping keywords by topic	Yes, well	Faster and tidier than by hand
Explaining the intent behind a query	Yes	Does someone want to buy, compare or learn?
Saying how often something is searched	No	A language model does not have those numbers
Saying how heavy the competition is	No	For that you have to look at the results
Choosing what pays off for your business	No	You know what a customer is worth to you

The fourth row is the trap: search volume. Ask ChatGPT for the search volume of a term and you get a number. It looks convincing. It is a guess. Which assistant is best for this work is in SEO with ChatGPT, Claude or Gemini.

Step 1: collect your customers’ questions (15 minutes)

Do not start with a tool. Start with your own business.

Write down:

• the five questions you get most often on the phone
• the questions from your last ten quote requests
• what customers say when you ask how they found you
• the services you earn the most on

A stair remodeling company in Columbus comes up with this list:

• how much does a stair remodel cost
• can it be done in one day
• carpet to hardwood stairs
• enclose open stair risers
• stair remodel columbus

Those are five queries. They come straight from customers, so people search for them.

Step 2: let AI expand the list (10 minutes)

Now comes the AI. Give context, or you get generic words.

I run a stair remodeling company in Columbus, Ohio. We remodel staircases in one day with hardwood, vinyl or laminate treads. Our customers are homeowners aged 35 to 65. Below are five questions customers ask us. Give me thirty other questions these people would type into Google before getting a stair remodel. Split them into: researching, comparing and buying. Do not give search volumes.

The last sentence matters. It prevents made-up numbers.

You get questions you did not think of. “DIY stair remodel or hire a pro”, “stair treads with radiant floor heat”, “make open stairs safe for toddlers”. Cross out what you do not do or do not want.

Step 3: add the numbers (20 minutes)

Now you test the list against real data. Three free sources.

Google Search Console. Open the Performance report and look at Queries. Here are the queries for which Google already shows your site, with impressions and position. This is the most valuable source there is: real numbers, from your site.

Google Keyword Planner. Free, but you need a Google Ads account with a campaign. You can pause that campaign right away. Paste your list in and you see a range per term, such as 100 to 1K searches a month. Rough, but enough to tell big from small.

Google itself. Type a query and look at the suggestions in the search bar and the “People also ask” box. Those are queries people really search.

Put the outcome in a spreadsheet. Three columns are enough: the query, the range and your current position if you have one.

Step 4: let AI group them (5 minutes)

Paste the list back into the AI.

Below is a list of search queries. Group them into topics I can make one page for. Give the main question per group and say whether it should be a service page or a blog post. Put queries with the same intent together, even if the words differ.

This is where AI saves the most time. “How much does a stair remodel cost”, “stair remodel price” and “cost to replace stair carpet with wood” belong on one page. Make three pages of them and they compete with each other.

The stair company ends up with nine groups. Four service pages, one pricing page and four blog posts.

Step 5: choose and assign (10 minutes)

Now you decide yourself. Per group, ask three questions.

Do I already have a page for this? If so, improve it. If not, create one.

Can I win here? Type the main question into Google. Only national chains and big comparison sites? Then it will be hard. Local competitors with mediocre pages? Then you have a chance.

Does it pay? “DIY stair remodel” is searched often, but those people do not call you. “Stair remodel Columbus” is searched less and brings customers.

Start with the groups where you already have a page at position 5 to 20. That is where you see results fastest. How fast is in how long SEO takes.

One page, one main question

The outcome of your research is a short table.

Page	Main question	Related questions
/stair-remodel	stair remodel columbus	staircase renovation, one-day stair remodel
/pricing	how much does a stair remodel cost	stair remodel price, cost to replace stair carpet
/hardwood-stairs	hardwood stair treads	carpet to hardwood stairs, replace stair treads
/blog/enclose-open-stairs	enclose open stair risers	make open stairs safe for toddlers, add stair risers

The main question goes in the title and the first paragraph. The related questions become subheadings. How to write the title is in writing an SEO title. How to connect the pages to each other is in topic clusters.

Keywords for AI assistants

There is a second list you now need too. What do people ask ChatGPT?

Those questions are longer and more personal. Nobody types “stair remodel columbus” into ChatGPT. They do type: “I have old open stairs and two toddlers. Which company near Columbus can make them safe and look good?”

Write down five to ten of those questions. Ask them in ChatGPT and see who gets named. Which questions are best to track is in which prompts to track.

Research that updates itself

Keyword research is a snapshot. In six months people search for other things, and your site gets shown for new questions.

Software connected to Search Console sees that continuously. GreatCompany reads your queries and looks for the question where you are just short of page 1. That is where the improvement starts. The questions to AI are measured too: every day GreatCompany asks your customers’ questions to ChatGPT, Gemini and Perplexity.

Want to know where your opportunities are? Ask your SEO assistant: “Where do I gain the most?” You get your opportunities in order, based on your own numbers.

What else you can leave to software is in automating SEO with AI.

See which questions you are already found for

Join the waitlist. When we open in the US, your free report shows which errors we find on your site, and for which customer questions ChatGPT names you or your competitor.