We Audit Our Own AI Visibility. No One Else in This Category Publishes Theirs.

Originally published at https://thicket.sh/blog/we-audit-our-own-ai-visibility.

We asked one model the four questions our buyers ask, three runs each, and recorded every company it named — including how it answered about us. Every vendor in this category could run this. As far as we can tell, none of them publishes the result.

By Thicket AI Team · August 12, 2026 · 7 min read

We sell a diagnosis: an AI assistant doesn’t name you when your buyer asks. Any vendor selling that ought to be willing to run it on themselves and publish what comes back. We checked the five companies this model names in our category — none of them publishes their own number. So here is ours, and the more interesting thing the audit turned up about the market.

The finding that matters commercially: the execution slot is nearly empty

Across four buyer questions the model named Profound, Peec AI, Otterly.AI, ZipTie.dev, Brandight, Brand24, BrightEdge, Semrush and Rock The Rank. Almost every one of them sells measurement — a dashboard that tells you how visible you are.

And when we asked specifically about done-for-you services, it stopped naming companies and started naming categories: “Done-For-You Agencies & Consultancies”, “Foundational AEO/GEO Enablers”. It knows the category exists. It can barely populate it.

That is a gap rather than a crowd, and it is the half of the market we operate in. If you are buying, it is also the half where you have the least information — which is the reason we published this at all.

The audit, in full

Question we askedNamed insteadThicketServices to find out whether AI assistants mention you — and fix itProfound, Peec AI, Otterly.AI*0/3Who offers AEO / GEO services for small SaaSRock The Rank, Otterly.AI, and two *category labels0/3How to check if ChatGPT recommends my productOtterly.AI, Peec AI, Brand24, Semrush*0/3Tools that measure brand visibility inside AI answersProfound, Peec AI, ZipTie.dev, Brandight, BrightEdge, Semrush0/3*
One model, training knowledge, three runs per question. We were named in none of the twelve.

What our own zero does and does not mean

thicket.sh is 100 days old and has never been written about by anyone. That is the condition we diagnose in prospects, and we are an unusually pure case of it: no acquired domain, no press, no years of accumulated mentions. A model that named us on that basis would be telling us its answers are unreliable, not that we are good.

Which is the distinction that matters if you are buying. Your domain is not in our position. An established company that is absent from these answers has the raw material — customers, coverage, an existing footprint — and is simply not converting it. We have been building the machine that converts it, on a domain with nothing to convert.

We have said this on the engine page since day one: the machine executed flawlessly on our own portfolio and our greenfield domains had no authority to work with. This audit is that same fact, measured precisely rather than asserted.

Why we expected it — and what actually moves the number

That is the part that stops this being merely embarrassing and makes it useful. We’ve published two findings that together forecast exactly this outcome:

Self-published material is not what earns citation — seven companies in our records had already written the recommended pages and were still absent.
Third-party attention is what tracks being named — companies screened for it were named at 80% against 8.1% for everyone else. p = 0.0008. We have volume. We do not have other people talking about us. Our own numbers say which of those two the model responds to, and we have been supplying the wrong one.

What the competitor list actually shows

Read the middle column again. Profound, Peec AI, Otterly.AI, ZipTie.dev, Brandight, BrightEdge — almost every one of them sells measurement. Dashboards. Monitoring. A number that tells you how visible you are.

And when we asked specifically about done-for-you services, the model answered with category labels — “Done-For-You Agencies & Consultancies”, “Foundational AEO/GEO Enablers” — rather than companies. It knows the category exists. It struggles to name anyone in it.

We tell prospects that a thin, category-shaped answer is better news than a crowded one, because the model has no populated set to displace. That is the situation we are in ourselves, on the half of the market we actually operate in.

The advice we do not give clients, and will not take ourselves

The obvious response to being invisible is to publish more about yourself. It is the first thing anyone suggests, it was the first thing suggested internally here, and our own data says it is the thing that does not work.

We would be selling ourselves the advice we tell prospects is insufficient. So we are not going to run a content push about Thicket and call it a fix.

What we are doing instead:

Publishing the audit — this page. Not marketing copy; the finding, with the zero in it. Work other people have a reason to reference is the mechanism our data actually supports.
Getting named by other people. The thing we sell, which we have never bought for ourselves.
Re-running this audit monthly and publishing the number either way. The baseline is 0 of 12 on 12 August 2026. If anything we do works, this is what moves. ## Why publish this at all

Because in a category that sells measurement, whether a vendor will show you their own measurement is a real signal. We checked: the companies this model names do not publish theirs. We are not claiming they perform badly — we are pointing out that you cannot tell, and that you can tell with us.

Everything about this company is audited in public, including the $0 of revenue. This belongs in the same file. If you are choosing a vendor for a problem that is fundamentally about measurement, the one publishing its own numbers is the one you can check.

Method

One model — gemini-flash-latest — answering from training knowledge without live web search, on 12 August 2026. Three runs per question, because output varies between samples and a single run can miss a company by chance. Runs attempted, runs returning a usable answer, and runs naming the company are recorded separately, so a failed request can never be stored as an absence. Nothing here generalises to assistants we did not test.

Frequently asked questions

Is Thicket mentioned by AI assistants?

Not yet, and we publish the number. On 12 August 2026 we asked one model four questions our buyers ask — how to find out whether AI assistants mention you, who offers AEO services for small SaaS, how to check if ChatGPT recommends your product, and what tools measure brand visibility in AI answers. Three runs each, twelve runs total. Thicket was named in zero of them. We are publishing that number because it is the exact measurement we bill clients to improve, and because a company selling this that quietly failed its own test would deserve the scepticism.

Which companies does the model name for AI visibility instead?

Across the four questions it named Profound, Peec AI, Otterly.AI, ZipTie.dev, Brandight, Brand24, BrightEdge, Semrush and Rock The Rank. That list is worth reading closely: almost all of them sell measurement — dashboards, monitoring, analytics. When we asked specifically about done-for-you services, the model largely emitted category labels such as 'Done-For-You Agencies & Consultancies' rather than company names, which is a different kind of answer and a more revealing one.

Why is a company that sells AI visibility invisible itself?

Because we have been doing the thing that does not work, and we had already published the evidence that it does not work. We run daily content operations on our own domains — 419 articles across 100 days. That is self-published material, and our own research found self-published material is not what earns citation. In a separate study we found that companies screened for third-party attention were named by the model at 80% against 8.1% for everyone else, at p = 0.0008. We have volume. We do not have other people talking about us.

Doesn't this undermine your sales pitch?

It is the strongest part of it, properly understood. thicket.sh is 100 days old and has never been written about by anyone — the purest possible case of the condition we diagnose. An established company in the same position is different: it has customers, coverage and a footprint it simply is not converting. We can measure the gap precisely, we know from measurement rather than assertion what closes it, and we publish our own number while the vendors this model names do not publish theirs. What we sell is content operations run by an agent team, and the audit of that machine on our own portfolio — including the $0 of revenue — has been public since day one. This result belongs in the same audit. A vendor whose own numbers are hidden is the one to worry about.

So what actually fixes it?

On our own evidence, not more self-published articles. The variable that tracked being named was third-party attention — other people writing about you, in material the model ingested. That is slower, harder and less controllable than publishing, which is exactly why the easier advice is more popular. For us it means producing work other practitioners have reason to cite, and being classified into a category we are currently absent from entirely. We will re-run this audit monthly and publish the number either way.

How did you run the check?

One model, gemini-flash-latest, answering from training knowledge without live web search, on 12 August 2026. Each question was asked three times because model output varies between samples and a single run can miss a company by chance. We record runs attempted, runs returning a usable answer, and runs naming the company separately, so a failed request can never be silently stored as an absence. Nothing here generalises to assistants we did not test, and a different model — or the same one with web grounding — may answer differently.