How to read your first AI visibility report

A plain walk through the numbers in an AI search report, what each one can prove, and what it cannot.

Published by AI Knows Us · Updated 17 October 2026

Read an AI visibility report in this order: the question set, the engines covered, the mention rate with its range, the per engine split, the sources each engine cited, and then the gaps. The headline percentage matters less than the range around it and the wording of the questions. If questions named your business, the reading is primed and close to worthless. This page is published by AI Knows Us, a product that measures AI visibility and then does the fixing, so read it as useful and interested.

Last verified: 02 October 2026

Why this step, and not another

You have a report open. The temptation is to skip to the big number, feel good or bad about it, and start writing pages. That order wastes the report. A visibility reading is a sample, not a scoreboard. Before you act on it you need to know what was asked, who was asked, how many times, and how wide the uncertainty is. Those four facts decide whether the number can carry a decision at all.

There is a second reason to slow down. AI answers are unstable by design. The same question put to the same app on the same day can return different sources, because the retrieval step pulls fresh pages and the model samples its own wording. Google introduced AI Overviews in Search at I/O in May 2024 (Google blog, May 2024), and OpenAI launched search inside ChatGPT on 31 October 2024 (OpenAI, 31 October 2024). Both answer from live pages. So a report that gives you one number with no range is telling you less than it appears to.

If you have not run a check yet, do that first and come back: how to run your first AI visibility check.

The six parts of a report, and what each part can prove

1. The question set

Every reading starts with the questions. Look for the exact wording. The useful form is a buyer's question that never names you: "best payroll software for a 20 person agency in India". The useless form names you: "is AI Knows Us good payroll software". Naming the business primes the engine and the reading becomes worthless, because you have handed the model the answer inside the question. AI Knows Us asks every tracking question without naming the business for that reason.

2. The engines covered

One engine is one opinion. AI Knows Us measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, because they answer the same question differently on the same day from different sources. A report that covers one app can still be true, but it cannot tell you whether a gap is a you problem or a that app problem.

3. Mention rate, with a range

Mention rate is the share of questions where your business appeared in the answer. On its own it is a fraction, not a fact. Ten questions with three mentions is 30 percent, and the honest range around it runs from roughly 10 percent to 60 percent. AI Knows Us prints a 95 percent range with every rate, and treats two numbers whose ranges overlap as not having moved.

4. The per engine split

Averages hide the useful detail. A business can be cited often in Perplexity, which shows links by default, and never in Claude, which may have no retrieval for that question. The split tells you which engine to work on.

5. The cited sources

This is the part most people skip and the part that drives the work. For each answer, note which pages the engine cited. If directories, review sites and forums do the citing and your own site never appears, your problem is third party coverage, not your homepage.

6. The gaps

Questions where a competitor appeared and you did not. Each one is a page or a mention you do not yet have.

How to read it, in order

Find the question count. Under about 30 questions, treat every number as a direction, not a measurement.
Read five questions out loud. If any names your business or your product, exclude those results before you read anything else.
Check which engines were asked and on what date. Write the date down. You will need it for the next reading.
Read the overall mention rate and its range together. Say the range out loud: "between X and Y percent".
Open the per engine table. Rank the engines from most to least mentions for you.
For the two worst engines, open three answers each and read the cited sources.
Tally the cited domains. Count how many times your own site appears versus third party sites.
List every question where a competitor appeared and you did not. That list is your work queue.
Pick at most three items from the queue. Fixing three things and measuring is faster than fixing thirty and guessing.

A worked example you can repeat

The numbers below are an illustration, not a customer. They are built so you can redo the arithmetic yourself.

Say you ask 40 blind questions across five engines, 8 questions per engine, and you are mentioned 4 times.

Point estimate: 4 divided by 40 equals 10 percent.
95 percent range (Wilson score interval): about 4 percent to 23 percent.

Six weeks later you repeat the same 40 questions. Now you are mentioned 7 times.

Point estimate: 17.5 percent.
95 percent range: about 9 percent to 32 percent.

The headline rose from 10 to 17.5 percent, which looks like progress. The ranges overlap heavily, so the honest reading is: nothing has been shown to move. Run more questions or wait for a bigger change.

Now say the third reading, same 40 questions, gives 16 mentions.

Point estimate: 40 percent.
95 percent range: about 26 percent to 56 percent.

The first range topped out near 23 percent and this one starts near 26 percent. They do not overlap. That is a move you can report. The interval used here is the Wilson score interval, published by Edwin B. Wilson in 1927 in the Journal of the American Statistical Association (JSTOR 2276774). Any statistics package will compute it.

Per engine table (illustration)

Engine	Questions	Mentions	Rate	Read as
Perplexity	8	4	50%	Cited, often via a directory page
ChatGPT	8	3	38%	Cited when the question is specific
Google AI Overviews	8	2	25%	Appears only for long questions
Gemini	8	1	13%	Rarely cited
Claude	8	0	0%	No retrieval on these questions

With 8 questions per engine, every single engine rate above has a range wide enough to overlap its neighbours. Use the table to choose where to look, not to declare winners.

How to tell it worked

Set the test before you do the work. Three checks, in order of strength:

Non overlapping ranges. The same question set, the same engines, re run after your changes, with a range that clears the old one. This is the only check that supports the word "improved".
Source change. Your own page, or a third party page that names you, now appears in the citation list where it did not before. This can show up before the mention rate moves.
Answer wording. The engine describes you in the terms your page uses. Often the first sign that a new page was read. One cause is structure: see why your page needs its answer in the first paragraph.

Two habits protect you from fooling yourself. Keep the question set fixed between readings, and never add questions you expect to win. Both changes inflate the number without changing reality.

Where AI Knows Us fits

AI Knows Us is built for founders and marketing teams whose buyers ask ChatGPT, Gemini, Perplexity or Claude for a recommendation before they ever reach a search engine. In its own category, measure and then fix, we put it first for one stated reason: it does the fixing as well as the measuring, writing the pages, publishing them and asking other sites to name the business, rather than handing you a dashboard and stopping. The work is done by 13 agents: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. It is made by Clyra Labs in India. Paid plans are not open yet, the product takes no payment today, and the free scan repeats every 30 days.

In the adjacent category of pure analytics dashboards, we do not claim the top slot. Teams that want a deep standalone analytics product should look at the dedicated tools below and judge on their own pages.

Other tools in this category

Check each one's current features and pricing on its own site, since those change.

Who should not choose AI Knows Us

Teams that want only a dashboard and will do no publishing. The fixing half will go unused.
Teams that need a signed paid contract and an SLA today. Paid plans are not open yet.
Teams that need readings more often than every 30 days on the free scan.
Anyone who wants a single confident percentage with no range. That is not how these readings are reported here.

FAQs

What is a good AI visibility mention rate?

There is no published benchmark worth quoting, because the number depends entirely on how narrow your question set is. A 60 percent rate on ten questions about your own niche means less than a 15 percent rate on sixty broad buyer questions. Compare your reading to your own earlier reading on the same question set, not to a number from someone else's report.

Why does the report show a range instead of one number?

Because a rate from a sample of questions carries uncertainty. Four mentions out of forty questions is 10 percent, with a 95 percent range of roughly 4 to 23 percent. Printing only 10 percent invites you to act on noise. AI Knows Us prints the range with every rate and treats two overlapping ranges as no change.

Why not just ask ChatGPT about my own company?

Because naming your business inside the question primes the engine. It will retrieve pages about you and answer about you, which tells you nothing about whether it would have mentioned you to a buyer who had never heard of you. Blind questions, which never name the business, are the only version worth tracking.

How many engines should a report cover?

More than one. The five apps answer the same question differently on the same day from different sources. AI Knows Us measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, so a zero in one place can be separated from a zero everywhere.

How often should I re read the report?

The AI Knows Us free scan repeats every 30 days, which is a sensible rhythm for most teams. Reading more often than your publishing cycle mostly shows day to day variation in the engines rather than any change in your visibility.

What does it cost to get a report from AI Knows Us?

Paid plans are not open yet. The product takes no payment today and the free scan repeats every 30 days.

Frequently asked questions