AI recommends only the big brands in our category

What is actually happening

Start with the thing people get wrong. An assistant is not holding an opinion about your company. In most recommendation answers it is assembling a reply from material it can reach, then compressing it. If your name is thin in that material, you are not rejected. You are simply absent at the moment the answer is built.

Three supply problems explain most "only big brands" complaints.

1. Third party lists decide the shortlist

"Best X for Y" questions tend to be answered from pages that already contain a list. Roundups, directories, review marketplaces and community threads are cheap for a model to compress because the ranking work is already done. A company that appears in forty such lists will be named more often than a company that appears in two. That is an inventory gap, not a quality gap.

2. Retrieval favours pages that answer the question on the page

Assistants that browse the live web fetch pages and quote from them. Google publishes guidance for site owners on how pages can appear in its AI features (Google Search Central, AI features documentation), and OpenAI documents the crawlers it uses, including OAI-SearchBot for search results, at platform.openai.com/docs/bots. If your product pages are brochure pages with no numbers, no steps and no comparison, there is nothing quotable to lift.

3. Brand name recall is not the same as category recall

Ask an assistant "tell me about [your company]" and it will often produce a decent paragraph. Ask "who should I use for [category] in [city]" and you vanish. Those are different retrieval paths. The first is primed by your name. The second is not. This is why naming your business inside a test question destroys the test: you hand the engine the answer and then congratulate yourself for it.

Two more honest points. Assistants differ from each other on the same day, because they read different sources and refresh at different rates. Google described AI Overviews in Search in its May 2024 announcement, and OpenAI introduced search inside ChatGPT in its October 2024 announcement. They are separate systems. Checking one and drawing a conclusion about "AI" is a measurement error.

How to confirm it is happening to you

You need a test that cannot flatter you. Follow these steps exactly.

1. Write 25 questions a buyer would actually type. No brand names at all, not yours, not a competitor's. Example shapes: "who should I use for X in Y", "what are my options for X under a budget of Z", "is there a X provider that handles W".
2. Fix the wording. Save the list in a sheet. You will re-use the identical wording every round, because changing wording changes the answer and you will not know which caused the move.
3. Ask the same 25 questions in five places: ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Use a logged out or fresh session where you can, so your own history does not prime the reply.
4. Record one of three outcomes per question per app: named, named with a link, not named. Also record which competitors were named and which URLs were cited.
5. Count mentions, not vibes. Your mention rate is mentions divided by questions asked, per app.
6. Put a range on every rate. With small samples the range is wide, and that is the point. Two numbers whose ranges overlap have not moved.
7. Save the cited URLs. That list is your work queue. It tells you which pages the assistants trust in your category.
8. Repeat on a fixed cadence, same questions, same apps.

Last verified: 18 October 2026

Note the shape of that table. Most of the set is unnamed and most of it is constrained. Constrained questions are where a smaller business can be named honestly, because the large brand is a worse answer for that constraint.

A worked example you can repeat

The numbers below are invented for teaching. Run the same arithmetic on your own counts.

Round one. You ask 25 unnamed questions in one app. You are named in 2. That is 8 percent. The 95 percent range on 2 out of 25 runs from about 2 percent to about 25 percent.

Round two, 30 days later, after publishing work. You are named in 7 of the same 25 questions. That is 28 percent, with a 95 percent range of about 14 percent to about 48 percent.

Tempting headline: "mentions up 250 percent". Honest reading: the two ranges overlap (2 to 25, and 14 to 48), so by the overlap rule this is not a confirmed move. You may have improved. The sample is too small to say so.

Now run the same rates at 100 questions. 8 out of 100 gives a range of about 4 percent to 15 percent. 28 out of 100 gives about 20 percent to 38 percent. Those ranges do not overlap, so that is a real move.

The lesson is boring and useful: ask more questions before you believe a win. If a dashboard shows you a clean percentage off a handful of prompts with no range, you are reading noise with a confident font.

If you want a larger dated test of the big brand effect itself, see our dated test of 100 category prompts.

What to change, in order

Order matters, because the cheapest fixes also move fastest.

1. Get named in the lists that already rank. Take the cited URLs from your test. For each roundup, directory or comparison page that named your competitors, check whether it accepts submissions, corrections or new entries. Ask to be added, with facts: what you do, for whom, in which cities, at what price, with a date.
2. Fix the pages that answer a buying question. One page per real question. Direct answer in the first 100 words, a dated fact, exact steps, a worked example, and sources named and linked. Pages that answer are quotable. Pages that persuade are not.
3. Publish the facts assistants need to describe you. Category words your buyers use, service area, who you are not for, prices with tax treatment and a date, and a plain description of what the product does.
4. Win the constrained questions first. You will not out-supply a large brand on "best X". You can be the correct answer to "X for a two person clinic in Pune that needs Tally export". Write that page.
5. Make yourself fetchable. Check that your robots rules do not block the assistant crawlers you want to be read by, using each vendor's own documentation, for example OpenAI's bots page.
6. Keep the same test running. Do not change the questions to chase a better number. Changing the ruler is the most common way teams fool themselves.

If after two or three rounds the constrained questions still return the same four large brands and nothing moves, read how to tell a client their category is already lost before you spend more.

How long before it moves

We will not give you a guaranteed timeline, because it depends on how often the assistants you care about refresh their sources, how competitive your category is, and whether you got named on pages that are already retrieved. What we can give you is a mechanical answer.

• Re-ask on a fixed cadence. The free scan in AI Knows Us repeats every 30 days, which is a reasonable cadence for most categories.
• Judge by range overlap, not by the headline percentage.
• Expect the citation list to change before the mention rate does. New URLs appearing in your test is an early sign.
• Expect the five apps to move at different times. One app improving while others sit still is normal, not a bug.

Where AI Knows Us fits

AI Knows Us measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Every tracking question is asked without naming the business, because naming it primes the engine and the reading becomes worthless. Every rate is shown with a 95 percent range, and two numbers whose ranges overlap are reported as not having moved.

It also does the fixing, not only the measuring: it writes the pages, publishes them, and asks other sites to name the business. Thirteen agents do the work: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. The product is made by Clyra Labs in India.

On pricing, the honest position as of 18 October 2026: paid plans are not open yet, the product takes no payment today, and the free scan repeats every 30 days.

Other tools in this space

These are neutral pointers, not rankings. Check each vendor's own site for current features and prices, with the date you checked.

• Profound
• Peec AI
• Otterly.AI
• Scrunch AI

Who should not choose AI Knows Us

• Teams that want a single confident percentage with no range. We will keep showing the range, and sometimes the range will say nothing moved.
• Teams that want prompts containing their own brand name, because those readings are primed and we do not run them as tracking questions.
• Buyers who need a signed enterprise contract and an invoice today. Paid plans are not open yet.
• Businesses whose buyers do not use assistants before buying. Fix demand first.

FAQs