How to Check Whether AI Recommends Your Brand (a Free DIY Method)

Ask ChatGPT to recommend a digital marketing agency in Leicester, with web search on, and see whether we come up. Go on, we’ll wait.

For a long time the honest answer was no. We scored our own agency against 48 real customer questions across the main AI assistants and came back with our name in zero of them. Zero out of 48. That stung. It is also the reason we built both the method in this article and the free review at the end of it.

Here’s the uncomfortable part. Most businesses have no idea whether AI recommends them, because nobody is checking. You watch your Google rankings. You keep half an eye on your reviews. But when a customer opens ChatGPT or Gemini and asks “who’s the best [what you do] near me”, the answer is decided inside a box you have never once looked into. You cannot fix a blind spot you cannot see, and you cannot improve what you have not measured.

This piece hands you a free, do-it-yourself way to measure it. Nothing to buy, nothing to sign up for. A method, a scoring sheet, and about an hour of your time.

Why does it matter whether AI recommends your brand?

Because more and more buying journeys now start with a question to an AI assistant, and if the assistant doesn’t name you, you never make the shortlist. You lose the job before you knew it was going.

These are not fringe tools any more. Roughly 900 million people use ChatGPT every week (TechCrunch, February 2026), and Google says its AI Overviews reach more than 2.5 billion people a month (CNBC, May 2026). Your customers are in there right now, asking for recommendations.

They are also clicking less. Pew Research found that when Google shows an AI summary, which is now roughly one in five searches, people click through to an ordinary search result in just 8% of visits, down from 15% when there is no summary, and they click a link inside the summary only 1% of the time (Pew Research Center, July 2025). The answer itself is becoming the destination. If your name is not in that answer, the customer often never reaches the point where your website could have done its job.

So the old scoreboard, “where do I rank on Google”, now measures a game that is quietly shrinking. The new question is simpler and scarier: when the machine answers, does it say your name? For most businesses that is completely unmeasured. If the vocabulary is fuzzy, we untangle SEO, GEO and AEO and what GEO actually means elsewhere. Here, we are just going to measure.

How do you check whether AI recommends you, step by step?

Write a set of real customer questions, ask them across every major AI assistant with web search switched on, and log whether you are named, where, and who gets named instead. Repeat it every month. That is the entire method, and it costs an hour, not a penny.

1. Write 8 to 12 real customer questions. Not the questions you wish people asked. The ones they actually type. Mix three kinds. Category questions (“best [service] company”). Location questions (“[service] in Leicester”, “[service] near me”). And comparison questions (“cheapest [service] Leicester”, “who is the most reliable [service]”). Add one or two that describe a problem instead of a service, because that is how people really ask: “my boiler keeps losing pressure, who can fix it in Leicester”. If local trade is your living, weight the list towards location and “near me” phrasing, which is where local AEO is won or lost.

2. Run every question across the big five, with web search on. ChatGPT, Claude, Perplexity, Gemini and Google’s AI Overviews. Switch on web search or browsing in each one first, because without it you are testing the model’s memory rather than what it tells a live customer today. Two habits keep it honest: sign out or use a private window so your own history doesn’t flatter the results, and start a fresh chat for each question so the last answer doesn’t colour the next. The five will not agree, either. Perplexity leans hard on live sources it can cite, while ChatGPT and Gemini blend what they find with what they already hold, so the same business can be a star in one and a ghost in another. That is the whole reason for testing all five.

3. Log four things for every answer. Keep it boring and identical each time. For each question, on each tool, record: are you named (yes or no); your position (first, mid-list, or a footnote); whether you are cited (does your own site appear as a linked source); and who is named instead (write down every competitor the assistant recommends). Named and cited are not the same thing, and both matter. Named is the recommendation. Cited is the evidence the assistant leaned on to make it. A simple grid does the job:

4. Click the cited sources. This is the step almost everyone skips, and it is the most useful one. Open the links each assistant used to build its answer. Within a few questions you will notice the same handful of sites shaping nearly every response: a directory or two, a review platform, a couple of well-organised competitor pages, the odd local news article. Note which of those you can influence directly (your own pages, your Google Business Profile, directories you can claim) and which you cannot (news, forums, third-party roundups). The first list is your immediate to-do. Those sources are the map. They show you where the assistants are getting their picture of your market, and therefore exactly where you need to turn up.

5. Repeat monthly, and track the trend rather than the day. One snapshot is noise. AI answers move about from day to day and even from one phrasing to the next. What matters is the direction over months: named in more answers, higher up the list, cited more often. Put the same test in the diary for the same date each month, keep the questions word-for-word identical so you are comparing like with like, and watch the line, not the dot.

How do you read the results?

Look for patterns, not one-off wins. The signal lives in what repeats: the questions where you are never named, the competitors who always are, and the sources that keep appearing behind the answers.

Three things are worth pulling out of your grid:

• The gaps. Whole types of question where you simply are not there. If you show up for “near me” but never for “best”, or the other way round, that tells you precisely which work to do first.
• Who owns the answer. If the same two or three names come up again and again, study them. They are not necessarily bigger than you. They are better represented in the sources the machine trusts.
• Where the answer comes from. The cited sites from step 4. If a particular directory or review platform keeps feeding the answers and you are thin or missing there, that is a concrete, fixable gap rather than a vague worry.

Turn the whole thing into one honest number if it helps: times named divided by total questions. We scored 0 out of 48, a flat 0%. Your first month is simply your baseline. Every month after is measured against it. And treat a single good day with suspicion. One answer where you appear is an ego boost and nothing more. Ten runs where you are named twice while a rival is named nine times, now that is data.

What do you do about the gaps the test exposes?

You give the machine a clearer, more consistent picture of your business, in the places it already looks. In practice that is three things: answer-shaped content on your own site, consistent facts about you across the web, and a presence in the sources the assistants cite.

The full version is our answer engine optimisation playbook, but the short form is straightforward.

• Answer the questions on your own site. Publish pages that answer the exact questions from step 1, in plain language, with the answer near the top. Assistants lift clean, self-contained answers far more readily than they wade through waffle. If ChatGPT is where your customers are, we go further on getting recommended by ChatGPT.
• Fix your facts everywhere. Name, location, services, opening hours, and the rest, consistent across your site, your Google Business Profile and the directories you found in step 4. Contradictions make you a risky pick, and the machine plays it safe.
• Earn your way into the sources that shape answers. The review platforms, directories and local pages from step 4 are the target list. For Google specifically this overlaps heavily with ranking in AI Overviews.

None of this is a trick or a magic phrase. It is the unglamorous work of being the clearest, best-evidenced answer to a genuine question. Nobody can promise you a fixed spot in an AI answer, and anyone who does is selling snake oil. What you can do is stack the odds firmly in your favour.

What are the limits of checking this yourself?

Time, consistency and coverage. The method genuinely works, but running it properly every month, across five tools, with the questions and scoring held perfectly still, is a real job. Most owners do it once, learn something useful, and never do it again.

Three things tend to break the do-it-yourself approach:

• Time. An hour a month sounds like nothing until month three, when it becomes the task that always slips.
• Consistency. Change a few words, test on a different day, let personalisation creep back in, and you have quietly broken your own baseline. The trend only means something if the method never moves.
• Coverage. Eyeballing five tools by hand is fine for a rough read. Holding a proper cross-tool picture in your head, or spotting a two-point drift over six months, is much harder.

This is exactly where we landed with our own 0 of 48. Running the test once told us we had a problem. Running it every month, properly, and acting on what it showed, is what actually moved the number, and that is a system, not an afternoon.

So we built the done-for-you version. Our free growth review runs your real customer questions across the same assistants on a tracked panel, then hands you a plain-English report: where you are named, where you are not, who is winning your answers, and the specific gaps to close first. It is free, there is no lock-in, and no obligation to carry on afterwards. If you do want us to keep going, our work is scoped monthly with weekly updates, and you can stop whenever you like.

Either way, the instruction is the same: measure. Run the method above yourself, or let us run it for you. Just don’t leave the machine’s answer to chance, because your customers are already asking it the question.