AI visibility tracking tools: what each category covers, and which vendors the engines cite

Four different products are sold as an AI visibility tool, and which one you need depends on whether you want to know what assistants say about you, what they read before they say it, or what to publish next. The four are prompt run trackers, which ask a fixed question set and record the answers; mention monitors, which watch crawled pages and social posts for your name; search suites with an AI module added to an existing subscription; and content tools that draft and repair the pages an engine reads. Almost every disappointed buyer in this category bought one of those four and expected another.

There is now a measured way to see which of the vendors behind these tools the engines themselves name. On 15 October 2026 we asked ChatGPT and Gemini 102 buying questions each in the United States, the United Kingdom and India, kept 597 usable readings and recorded all 10,652 citations. semrush.com was cited in 95 of those readings, tryprofound.com in 61, otterly.ai in 58, depra.ai in 53, peec.ai in 49. Our own domain, aiknowsus.com, was cited 3 times, and that number stays on this page for a reason given further down.

The four categories, and the job each one does

• Prompt run trackers. You give them a list of buyer questions, they run each one on named engines on a schedule, and they log whether you were mentioned, recommended or cited. This is the only category that measures the thing a buyer actually experiences. Judge it on the question list, the repeat count and whether the raw answers are kept.
• Mention monitors. These watch the open web, forums and social platforms for your brand name. They are good at telling you what exists for an engine to read, and that is a different measurement from what an engine said. Useful alongside a tracker, weak as a substitute for one.
• Search suites with an AI module. If you already pay for a rank tracking suite, the module is the cheapest way to get a first reading. It usually runs a smaller question set and fewer repeats than a dedicated platform, so treat its number as a direction rather than a baseline.
• Content and fix tools. These draft pages, add structured data and check that crawlers are allowed in. They change the inputs rather than measure the output, so they belong in the same budget as a tracker and not instead of one.

Which vendors the engines named, counted

This table counts how many of the 597 readings cited each vendor's own website as a source for a buying question, on ChatGPT and Gemini, on 15 October 2026.

Domain	Readings citing it, of 597
semrush.com	95
tryprofound.com (Profound)	61
otterly.ai (Otterly)	58
depra.ai (Depra)	53
g2.com	49
peec.ai (Peec)	49
aiknowsus.com (us)	3

Read that column for exactly what it is. It measures how often each vendor's own site was used as a source in an answer, which is a measure of who these two engines read. It is not a product review and not a ranking of quality. A tool can be very good and barely cited, and a vendor cited 58 times has proved something narrower than a feature list: it has proved that the engines it sells visibility in are reading its pages. On a sales call that is worth knowing, in both directions.

What to check before you pay for one

• The question list, visible and exportable, with your brand name in none of the questions. A question that names you measures the question, not the engine.
• The engines named one at a time, and the surface each is read through. In the 102 conversations where we showed ChatGPT and Gemini their own answers and asked what a claim must carry, 98 of 102 asked for engines to be enumerated by name instead of called "all AI". Hold a vendor to the same line.
• The repeat count per question, and the spread between repeats. One run per question is an observation. 83 of those 102 conversations asked for a repeat count, a variance or a confidence interval before they would treat a number as a rate.
• Raw answers retained. If last month's figure can be rebuilt from stored answers, the number is auditable. If only the chart survives, it is not.
• Price in writing, in your own currency, with the add ons named. 102 of 102 of those conversations asked for pricing in the reader's own currency, which is a fair test to apply to a vendor as well as to a page.

Before you shortlist, run fifteen of your own buyer questions by hand on two engines and score them yourself. It takes an afternoon, it tells you what the category looks like for your own name, and it gives you a figure to hold a demo against. We published the hand audit spec for that at AI visibility tools compared, including the scoring rules and the fifteen question split.

When two tools report different numbers for the same brand

They will, and the reason is usually visible in the settings. Two tools rarely share a question list, a repeat count, an engine surface or a definition of a mention, and any one of those differences moves the headline figure. Decide which definition you are managing to, write it down, and keep one tool as the number of record. The same thing happens in classic search, where two rank checkers disagree for reasons that are mostly about method rather than about your site, and the three ways to check a position are set out in web page rank check.

Our own position, stated plainly

We sell this. AI Knows Us runs buyer questions across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, reports which sources decided each answer, and then writes and places the pages that change it. The 15 October run covered two of those five engines, ChatGPT and Gemini, and it is the only part of this page that is measured. Our own result from it was 3 citations in 597 readings, with none in the United States on either engine and none on Gemini in any market. 64 of the 102 conversations asked for the commercial interest to be disclosed on the page, so it is disclosed, with the worst number we have attached to it.

Questions about buying an AI visibility tool

Which category should a small company buy first?

A prompt run tracker, because it answers the question the business is actually asking. Add a mention monitor when you need to know why an answer changed, and content tooling once you know which pages the engines are reading for your category.

Is a single blended visibility score across engines useful?

It is useful for a board slide and poor for decisions, because the engines disagree. In our run the same brand could be cited by one engine and absent from the other in the same market, which a blended score hides. Ask for the per engine number underneath it.

How many questions and repeats are enough?

Forty to a hundred questions, run three times each per engine, is the level at which a month on month change is readable. Below about forty questions you are reading observations, which are still worth having as a baseline.

Do these tools cover Perplexity, Claude and AI Overviews?

Coverage differs by vendor, so ask for the list by name and the surface each one is read through. Our 15 October dataset covers ChatGPT and Gemini only, so treat anything it says as evidence about those two engines and nothing else.

Can a spreadsheet replace a tool for the first month?

Yes, and it is a good way to learn what you are buying. Twenty questions, two engines, a fresh conversation each time, one row per answer with the date, the engine, whether you were named and every domain cited. The limit you will hit is repetition, which is where a tool starts paying for itself.

The one column to keep yourself

Whatever you buy, keep one column in your own sheet: out of the questions you ran, how many answers cited your domain, on which engine, on what date. Ours reads 3 of 597 on ChatGPT and Gemini as of 15 October 2026, and the next run is scored against it. A vendor that cannot show you how to rebuild that one column by hand is selling a dashboard, and a dashboard is the part of this category that is easiest to make and hardest to check.