How to Do Keyword Research With AI for a Small Business Website

Use AI to brainstorm and organise, and real search data to decide. List the words customers actually use, have AI expand them into 100-200 phrases grouped by intent, check demand in Google's free tools (Keyword Planner, Search Console, autocomplete, Trends), then give each important page one main phrase. Treat any search volume an assistant quotes as a guess.

That last point matters because chat assistants don't have live access to search volume data. Ask one for "monthly searches" and it will produce tidy, plausible numbers that nobody measured. The other common mistake is chasing broad phrases. A small business site won't rank for "books" or "cakes", but it can win specific searches with clear intent, such as "signed first editions" or "order a book in", which are often the ones that turn into sales.

Follow me on Instagram@sagnikteaches

Start with the words your customers use

Everything below is worked through for a made-up but typical independent bookshop with a website, a small online shop, a programme of author events and a book club. The owner's first instinct was to brainstorm keywords from the industry's vocabulary ("literary fiction", "independent bookseller"). Customers rarely talk like that, so she started with what they actually say. Good sources:

Connect on LinkedInSagnik Bhattacharya

• Enquiry emails and messages: "Can you order in a book that's out of print?"
• Reviews, yours and competitors': "great for signed copies", "lovely kids' section".
• Questions at the till, noted for a fortnight on a pad by the counter.
• Your website search box, if it has one, and chat logs.
• Google Search Console, the free Google tool that shows which searches your site already appears for. More on this below.

Paste the raw material into an assistant and ask it to extract the phrases:

Subscribe on YouTube@codingliquids

Below are customer emails, reviews and till questions for my independent
bookshop. Extract every phrase a customer might type into Google when
looking for what they asked about. Keep their wording. Group them by
what the customer wanted (e.g. buy, find out, visit, book an event).
Don't invent phrases that aren't implied by the text.

Part of the result (illustrative): "order a book in", "out of print book search", "signed copies", "book club [town]", "children's books for reluctant readers", "book tokens", "author event tickets", "second-hand books [town]", "gift ideas for book lovers". Every one of those came from a real customer, which makes them far better seeds than a brainstorm.

Expand the list with AI, then cut the fantasy

Now use AI for what it's good at: generating variations you hadn't thought of. Give it the seed list and ask for expansion by intent, the reason behind a search:

Here are seed search phrases from real customers of my independent
bookshop in [town]: [list]. Expand to about 150 phrases real people
might search. Group by intent:
- BUY (ready to purchase), RESEARCH (comparing or learning),
- VISIT (want a physical shop or event), QUESTION (need an answer).
Include question-style searches (how, what, can I, where).
Don't include search volumes. Flag phrases where big online retailers
will dominate the results.

The expanded list included useful finds: "how to find a book when you only remember the plot", "book subscription gift", "signed books as gifts", "independent bookshop gift card", "book club questions for [popular title]". It also included the fantasy that every AI list contains:

• "Cheap books online" and "free ebooks": a small shop can't and shouldn't compete here.
• "Best books of all time": huge competition and no buying intent.
• "Bookshop open 24 hours": the shop isn't.

Cut anything you can't serve, anything a giant owns, and anything that doesn't lead towards what you sell. She kept about 90 phrases.

Check demand with Google's free tools

This is the step that replaces AI's guesses with data, and it's worth seeing why it can't be skipped. Before starting properly, the owner had asked an assistant for "a table of the best keywords for my bookshop with monthly search volumes". It returned a confident table: "signed books, 2,400 a month", "book club [town], 880 a month", "independent bookshop [town], 1,300 a month". Keyword Planner later put "book club [town]" at 10-100. The assistant hadn't looked anything up; it had produced numbers of the right shape. Had she planned her pages around that table, she would have spent a month on the wrong ones. If an assistant ever gives you search volumes without naming a data source you can check, treat them as fiction.

Four free sources, each answering a different question:

• Google Keyword Planner (inside a Google Ads account; you don't have to run ads) shows monthly search estimates. Without active ad spend it shows broad ranges, such as 100-1K, rather than exact figures. That's enough to tell a phrase with real demand from one with almost none. Setting up the account may push you towards creating a campaign; look for the option to switch to expert mode and skip it.
• Google Search Console shows the searches your site already appears for, with impressions (how often you were shown), clicks and average position. For an existing site this is the most valuable data you have, because it's about you.
• Google's own search box. Type the start of a phrase and note the autocomplete suggestions; check the "People also ask" questions and related searches at the bottom of the results. These reflect what people actually search.
• Google Trends shows whether interest is seasonal. "Book advent calendar" is dead for ten months and huge in November.

The bookshop's findings for a few phrases (illustrative):

Low-volume phrases aren't failures. "Order a book in" might be searched a few dozen times a month, but almost everyone who searches it wants exactly what the shop offers. For a small business, ten well-matched visitors can be worth more than a thousand vague ones.

Group phrases into page-sized clusters

A cluster is a set of phrases that one page could satisfy completely. The test is simple: would the same page answer all of them well? AI is quick at the first pass:

Group these search phrases into clusters, where each cluster could be
fully answered by ONE web page. For each cluster: a working page title,
the main phrase (the clearest, most searched-sounding one), 3-6
supporting phrases, and the intent (BUY, RESEARCH, VISIT, QUESTION).
If a phrase needs a different answer from the others, give it its own
cluster. Phrases: [paste your checked list]

A few clusters from the result, after her corrections:

The correction: the AI had put "book club questions for [title]" in the book club cluster. Someone searching that wants a discussion guide, not a club to join. It became a separate idea for a monthly page of discussion questions, which could bring book-club organisers to the site every month.

Can you realistically rank? Look at page one yourself

Demand is half the question; competition is the other half. Search each main phrase in a private browser window and look at what already ranks:

• Giant retailers and marketplaces across the whole page: hard to beat. Only target the phrase if you have something they don't.
• A mix of small businesses, articles and directories: winnable with a genuinely useful page.
• A map pack of nearby businesses at the top: this is a local search, where your Google Business Profile matters as much as your website. Optimising your Google Business Profile with AI covers that side.
• An AI Overview at the top: the question is being answered directly. A clear, specific page still earns clicks from people who want the detail or want to buy. How AI Overviews are changing local search covers what that means for small businesses.

For "signed books", page one was dominated by large retailers and specialist dealers, but the second page included several independent shops. With a proper page, a list of what's in stock and a steady flow of signed stock from author events, the bookshop had a realistic chance. For "independent bookshop" on its own, most results were directories and famous shops elsewhere; she decided not to chase it and to focus on "[town] bookshop" through the Business Profile instead.

Turn it into a keyword map

The keyword map is the useful output of the whole exercise: one row per page, existing or planned. Keep it in a spreadsheet:

Two rules keep the map useful. One main phrase per page, so pages don't compete with each other. And every row has an action, because a keyword list with no pages attached changes nothing.

A quick sum on the time involved, from the bookshop's first run: gathering customer phrases took about an hour, expanding and cutting the list 40 minutes, checking demand in the four tools about 90 minutes, and clustering and mapping another hour. Call it four hours, most of it the checking, which is exactly the part AI can't do for you. The map then drove six months of website work: two new pages, three rewrites and one retitled page.

Using the map when you write

With the map done, each page has a brief. Use the main phrase in the page title, the main heading and the opening lines, naturally, and answer the supporting phrases in the body. Don't repeat the phrase awkwardly; Google understands variations, and customers notice stuffing. When you draft with AI, give it the row from the map and your own facts: "Write the opening section of our 'order any book' page. Main phrase: order a book in. Answer: how it works, how long it usually takes (7-10 days for most books, longer for out of print), what it costs (nothing extra), and how to ask. Use plain, friendly language." Then edit it into your voice.

For writing pages that rank without sounding machine-made, writing website articles with AI that still rank goes further, and writing your home, about and services pages with AI covers the core pages. On AI search specifically: Google's guidance is that its AI Overviews and AI Mode need no special optimisation beyond normal good practice, so a clear page built on real customer questions is the right foundation for both.

Winnable phrases for groomers, cafés, bike shops and florists

The steps don't change, but the winnable phrases do. Some illustrations:

• A dog groomer: broad terms like "dog grooming" are dominated by directories and big chains. Specific phrases customers use, such as "hand stripping [breed]", "puppy first groom" and "cat grooming [town]", are often winnable and match services customers actually search for by name.
• A café: most searches are local and handled by the map pack, so the Business Profile does most of the work. The website can still win specific ones: "gluten-free brunch [town]", "dog-friendly café [town]", "private room for meetings [town]".
• A bicycle repair shop: "bike service cost" and "how much is a bike service" are question searches worth a page with actual prices, since the page that answers them honestly tends to win the booking. "E-bike repair [town]" is growing and worth its own page.
• A florist: "flower delivery" is owned by national networks; "same day flowers [town]", "funeral flowers [town]" and "wedding florist [town]" are local and high value. Seasonal peaks around Mother's Day and Valentine's Day show clearly in Trends, so pages need to be ready weeks ahead.
• A bakery: "celebration cakes [town]", "gluten-free birthday cake" and "how many people does an 8 inch cake feed" (a question worth answering with a sizing page) beat "bakery".

In each case, the phrases that win are the ones where the searcher's need and what you sell line up exactly. What AI can and can't do for your local SEO covers the local side in more depth.

Refresh the map every quarter from Search Console

Once pages are live, Search Console becomes your keyword research tool. Every three months, export the Queries report for the last three months (the export button gives you a spreadsheet) and ask an assistant to find opportunities:

Here is my Search Console export: query, clicks, impressions, CTR,
average position. Find:
1. Queries with 50+ impressions and position 8-20 (close to page one)
2. Queries with high impressions but CTR under 1% (title or snippet problem)
3. Queries that don't match any page in my keyword map (pasted below)
4. Queries that suggest a new page is needed
Only use the data provided. Show the numbers for each.

Sample output for the bookshop, six months on (illustrative):

Close to page one: "signed books" (position 11, 1,200 impressions), "book club [town]" (position 3, rising). Low CTR: "order a book in" shown 410 times, CTR 0.7%; page title reads "Special Orders", which doesn't match the search. Unmatched: "book club questions [title]" appeared 14 times. New page suggested: "second-hand books [town]", 230 impressions, no matching page.

Each line is an action: retitle the order page to use the customers' phrase, add a discussion-questions page, and decide whether the second-hand shelf deserves its own page. Check every number against the export before acting, because an assistant working from a large spreadsheet can misread a column. Twenty minutes a quarter keeps the map matched to how people actually search, which drifts more than most owners expect. If you'd like a wider health check at the same time, auditing your website's SEO with AI in one afternoon pairs well with this refresh.

Keyword research with AI: follow-up questions

Can ChatGPT tell me the search volume for a keyword?

Not reliably. A chat assistant doesn't have live access to search volume data, so any number it gives is an estimate at best and invented at worst, even when it sounds precise. Use it to brainstorm and group phrases, then check demand in Google Keyword Planner, Search Console or a keyword tool that shows its data source.

Do I need a paid SEO tool as a small business?

Usually not to start. Google's free tools, Keyword Planner, Search Console, autocomplete and Trends, cover most of what a small site needs. Paid tools add competitor data and more precise volumes, which helps if you publish a lot or compete online nationally. Try the free route for a quarter before paying for anything.

Does keyword research still matter with AI Overviews and AI search?

Yes. Google says its AI features need no special optimisation beyond normal search best practice, so pages that clearly answer the questions people search for remain the foundation. The research also tells you what customers ask, which is exactly what AI assistants are trying to answer when they recommend a business.

How many keywords should one page target?

One main phrase and a handful of close variants that the same page would satisfy. If two phrases need different answers, they need different pages. Trying to rank one page for everything usually means it ranks well for nothing, and it reads badly to customers too.

Further reads

• Does Google Penalise AI-Written Content on a Business Website? — Whether AI-written pages hurt your rankings.
• What Is Generative Engine Optimisation and Does Your Business Need It? — What optimising for AI search actually involves.
• How to Track Whether AI Search Is Sending You Customers — See whether AI search is sending you customers.
• How to Choose an SEO Agency That Understands AI Search — If you'd rather hand SEO to an agency, how to choose one.
• How to Do Competitor Research With AI in One Afternoon — See which searches your competitors already win.
• How to Get Your Business Recommended by ChatGPT and AI Search — Get recommended when customers ask AI assistants.
• How to Use AI for Etsy Titles, Tags and Descriptions — Search phrases first, then prompts for each field: a title under 15 words, 13 non-repeating tags of 20 characters or fewer, and a description buyers can act on.
• How Wedding Businesses Get Found When Couples Ask AI for Suppliers — How AI assistants build supplier shortlists for couples, a test to see what they say about you, and the five fixes that get a wedding business named.
• Free vs Paid AI Marketing Tools: What Is Worth Paying For — A category-by-category comparison of free and paid AI marketing tools, the upgrades that pay for themselves, and the ones that rarely do.
• Can AI Manage Your Google Ads Without an Agency? — What Google's own AI handles, the jobs it leaves with you, a catering company's move from agency to DIY and a 45-minute weekly routine.
• How to Test a New Service Idea With AI Before You Launch It — A six-week way to test a new service with AI doing the drafting and real customers supplying the evidence, with pass marks set before the results arrive.
• AI Website Builder or Web Designer: What Each Really Costs — Builder prices for Wix, Framer, Durable and Hostinger, what designers charge according to Clutch, and a three-year cost that counts your own hours.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Ads help on Keyword Planner; Google Search Console help on the Performance report; Google Trends; Google Search Central guidance on AI features in search (facts sheet).