Is there any independent evidence about AI Knows Us, and the honest answer is not yet
No third party has tested us, no analyst has written about us, and in 635 readings of US buying questions on 22 October 2026 we were named 0 times and cited 0 times. Here is the measurement, the denominators, and the one figure that does not exist at all.
Published by AI Knows Us · Updated 22 October 2026
If you are checking whether anybody independent has verified what AI Knows Us claims, the answer on 22 October 2026 is that nobody has. There is no third party accuracy test of our product, no analyst note, no press study, and no citation of our domain inside the AI answers we measure. We can put numbers on that last part because we measure it on ourselves: across 635 readings of US buying questions on ChatGPT and Gemini, our domain was named in 0 readings and cited in 0, and it appears in none of the 19,793 citation URLs those readings produced, across 3,307 distinct hosts. Rival vendors were named 1,053 times in the same set, from a roster of 24. This page sets out what independent evidence would look like, which parts of it exist, which do not, and how to check us without taking our word for any of it.
The four kinds of independent evidence, and which we have
| What a buyer would want | Does it exist for us on 22 October 2026 |
|---|---|
| A third party accuracy test of the product, measuring whether what it reports matches what the engines actually said | No. We know of no such published test of any vendor in this category, ours included |
| Analyst or press coverage by an organisation with no commercial relationship to us | No. We have none to point to |
| Verified customer reviews with a count and a date beside them | Not published by us. The one review directory among the twenty five most cited hosts in our own citation set is g2.com, with 125 of 19,793 citations |
| Citation by the engines themselves, which is the closest thing to an independent vote in this category | No. 0 of 19,793 citation URLs point at our domain, and 0 of 635 answers named us |
That is the whole answer to the question in the title. The rest of this page is the evidence behind the no, because a vendor saying it has no independent evidence is only useful if it shows you how it knows.
The measurement, with its denominators
- 635 readings on the ChatGPT and Gemini product surfaces through a data provider, scored on 22 October 2026.
- 318 questions asked from a bank of 324 US buying questions. No question in the measured set names our company, because a question carrying a name produces an answer that carries it back.
- 550 of the 635 readings ran a live web search. The other 85 answered from what the model already held.
- 19,793 citation URLs collected, every one of them with a parseable host, across 3,307 distinct hosts. Half of all citations came from 51 of those hosts.
- Our domain: 0 citations, 0 mentions in answer text, 0 namings.
- Rivals: 1,053 namings across 24 vendors. In 451 readings a rival was named and we were not. In 184 readings no vendor from the roster was named at all. The counting rule is once per reading in which a name appears, never once per occurrence.
Before the zeros were believed, the detectors that produce them were tested against planted text, because a detector that cannot find a name it was given cannot be trusted to report its absence. The full citation study behind those hosts is published separately, in which websites the two engines cite when an American buyer asks for marketing help.
Three passes, and the one that went the wrong way
| Pass | Date | Readings | Times our domain was cited | Times we were named |
|---|---|---|---|---|
| First | 15 and 16 October 2026 | 597 | 3 | 0 |
| Second | 20 October 2026 | 1,260 | 6 | 0 |
| Third | 22 October 2026 | 635 | 0 | 0 |
The first pass used a narrower question bank, so it is not directly comparable. The second and third are: 72 core questions and 210 US questions were reused word for word between them. Between those two passes we published 507 more recommendation shaped pages, taking our live count from 134 to 641, and the citation count went from 6 to 0 while the naming count stayed at 0. We publish that because it is our own worst number and because a vendor that hides its flat results cannot be checked at all. The full before and after is in our report on publishing 507 more pages and measuring what changed.
The figure that does not exist, and we will not estimate it
There is one measurement a buyer would reasonably ask for that we cannot give. Our question bank holds 324 questions, six of which name our own company directly. Those six have never been run: the job loop that executes the measurement skips every brand question in both its validation and its run, so 318 were asked and 0 brand readings exist.
That means we have no figure for what an engine says when somebody asks about us by name, which is the prompted ceiling for any vendor in this category. We will not model it, interpolate it from the unprompted numbers, or quote a competitor's version of it. It does not exist in our data, and the honest thing to publish is that sentence rather than a number with a confident shape.
What we do publish, labelled for what it is
All of the following is first party evidence. It is checkable, it is dated, and it is not independent.
- Search Console exports on our proof page. Read from the console rather than retyped, for our own properties, published as two separate cases that are never combined into one claim.
- Our full price ladder. In the US and the UK, SEO Start at 149 dollars a month, Growth 299, Scale 599, SEO Scale 699, Inbound 1,099, Outbound 1,299, Agency 1,299 and Everything 1,799. Outside the US and the UK the first two tiers are 99 and 199, so there is no single global floor. SEO Start reads 0 questions on 0 engines, which makes Growth at 299 the cheapest plan that reads all five surfaces, and a Claude reading costs 3 credits where the others cost 1. A free scan of your own domain needs no account.
- The contradictions on our own pricing page, published rather than tidied. Read on 22 October 2026, across 119,203 bytes: the Agency plan is described as covering ten client brands while the comparison table's more than one business row reads no for every plan including Agency. The main listing names five apps including Claude while the Scale description names four without it. The FAQ says SEO Start gets 12 pages a month and no AI app checks, beside a table that lists five engines for that plan. All of those render. If one of them decides your purchase, get our answer in writing before you buy.
- The limits of our own method. The citation audit step cannot run through the data provider's API at all, because it needs the engine's own answer back in the prompt and the provider caps a prompt at 4,000 characters while one batch answer measured 21,711. That step was run by hand in the product interfaces instead. The provider also accepts country, location and language parameters and ignores all four, which we measured, so the US demographic is fixed by a market lock line inside the prompt rather than by a setting.
How to check us without trusting us
- Run the test yourself. Write five buying questions a customer of yours would type, with no brand name in them. Ask them on ChatGPT, Gemini and Perplexity on the same day. Write down every vendor named and every source listed underneath.
- Check whether we appear. On our own data you should expect not to find us, and if you do find us, the date and the question are worth sending to us because our record says 0.
- Ask any vendor in this category for four numbers: how many questions, on which engines, on what date, and what proportion of the answers named you. A vendor that cannot produce a denominator is reporting an impression.
- Ask for the misses. The useful question to a measurement vendor is not what it found. It is what it looked for and did not find, and whether it published that too.
Whether any tool in this category reports accurately is itself unmeasured, and the shape such a test would need is set out in our note on whether anybody has tested AI visibility tool accuracy.
Where rivals are stronger, by name
A page about evidence has to include the evidence against itself. Read on 22 October 2026: tryprofound.com publishes single sign on, SAML and SOC 2 with up to nine answer engines on its enterprise tier, and peec.ai lists single sign on and role based permissions among its plan features. Our own pricing document renders no single sign on, no SAML, no SOC 2, no ISO certification, no data processing agreement and no white label anywhere in its text, and we read five surfaces rather than nine. On the measured side, the same rivals that outrank us on those features were also named in our audit while we were not: one was named in 154 readings, another in 124, another in 91, and we were named in 0.
We sell the remedy to the problem this page measures, which is the conflict of interest a reader should hold against every number on it. The defence is not an assurance. It is that the method, the question bank, the dates, the denominators and the zeros are all published, so the measurement can be repeated by somebody with no interest in the outcome.
Questions buyers ask about evidence for an AI visibility vendor
Has anybody independently tested AI Knows Us
No. We know of no third party test of our product, and we know of no published independent accuracy test of any vendor in this category. The nearest available check is to run a question set yourself on two engines and compare what you see with what a vendor reports.
Why would a vendor publish that it was named zero times
Because the number is checkable and the alternative is not. Our zero sits beside 1,053 namings of 24 rivals in the same 635 readings, which tells a buyer exactly where we stand in the answers today and exactly what we are selling against. A vendor that only publishes its wins has given you nothing you can test.
What is the prompted citation rate for AI Knows Us
Unknown, and we will not estimate it. The six brand questions in our bank have never been executed, so 0 brand readings exist. Any figure for what an engine says when asked about us by name would be invented.
Are G2 or Clutch reviews good evidence in this category
They are evidence of experience rather than of measurement accuracy, and in the engines' own behaviour they are a small input: the one review directory among the twenty five most cited hosts in our set carried 125 of 19,793 citations. Read them for how a vendor behaves with customers, and ask the vendor directly for method, N, date and denominator.
What should I ask us before buying
Ask which five surfaces are read on the plan you are considering, how many questions it covers, how often it runs, what a Claude reading costs in credits, and whether your contract needs a data processing agreement or a security certification, because our published page offers none and that answer has to come in writing.