We asked four AI engines which AI visibility tools they recommend

Original research · Data collected 6 September 2026 · 40 answers · Written by the Kartafla team

Everyone in this category publishes a "best tools" list. We wanted to know what the AI engines themselves say — so we asked them 10 real buying questions and counted the answers.

How we checked: We asked 10 commercial questions to ChatGPT, Claude, Perplexity and Gemini on 6 September 2026, one run per question per engine, giving 40 answers. For each answer we recorded every tool the engine named — not a fixed shortlist — and which URLs were cited. Every question was answered by every engine (10/10 coverage). Limitations are listed at the bottom of this page and they matter — read them before quoting these numbers.

Which tools the engines named most

Across all 40 answers, counting every time a tool was named:

Counts are mentions across answers, not a quality ranking. A tool named often is one the engines know about — which is precisely the thing this category sells.

The engines barely agree with each other

This was the clearest finding. Only 7 of the 152 tools named in this run were named by every engine — Otterly.AI, Profound, Peec AI, Semrush, Ahrefs, Scrunch AI, Awario. Every engine had a different most-named tool, and the volume of naming varied enormously.

Gemini was the most generous namer, listing 102 tools across its 10 answers. ChatGPT named 52 — roughly 2.0× fewer. Claude declined to name any tool at all in 1 of 10 answers, sometimes answering with criteria instead of products; Perplexity declined to name any tool at all in 1 of 10 answers, sometimes answering with criteria instead of products.

The practical consequence: "we rank well in AI search" is close to meaningless as a single claim. Visibility in Gemini says little about visibility in ChatGPT, and a tool that tracks one engine is measuring a fraction of the picture.

What the engines actually cite

Across these answers the engines cited 500 links spanning 270 distinct domains. The most-cited:

Vendor websites are a small minority. What the engines lean on is third-party roundups, community discussion, and reference sites — the "10 best AI visibility tools" genre, Reddit threads, and encyclopedia entries. Citation counts per answer also varied by engine: ChatGPT cited sources in 10 of 10 answers (7.5 links on average); Claude cited sources in 9 of 10 answers (4.7 links on average); Perplexity cited sources in 10 of 10 answers (19.7 links on average); Gemini cited sources in 10 of 10 answers (18.1 links on average).

The practical consequence: the highest-leverage work for AI visibility is usually not on your own website. It is getting written about, accurately, on the sites the engines already trust.

Where we came in: 0 out of 40

Kartafla was named in 0 of 40 answers. Zero. We build an AI visibility tool, we ran this study with it, and it found that the engines have never heard of us.

We are publishing that for two reasons. The first is that a study of who gets recommended, run by a vendor who quietly omits their own result, is marketing wearing a lab coat — and you should discount any such page you find, including ours if we ever do it. The second is that it is the honest demonstration of the product's premise: a new company can be genuinely useful and completely invisible to the engines at the same time, and you cannot fix that until you measure it.

The data above also tells us exactly why. We are not in the roundups, not discussed in the communities, and not cited by the reference sites the engines lean on. That is the work, and it is the same work this study would prescribe for any brand starting from zero.

Limitations — read these before quoting the numbers

• One run per question per engine. These models are non-deterministic; ask again tomorrow and the wording, and some of the names, will change. This is a snapshot, not a ranking, and a single run cannot separate a real preference from a coin flip.
• Names are read from the answer text, and that is imperfect. Unlike our earlier runs we no longer count only a shortlist we chose in advance — every tool an engine names is captured, which is why 152 tools appear here. The cost is that the same vendor arrives under several spellings and the engines sometimes name a category rather than a product, so the counts are normalized before publication; the raw run and every normalization decision are in the dataset.
• AI engines are not counted as tools. Asked to recommend tools, the engines mostly name each other — ChatGPT, Perplexity, Gemini and Claude were among the most-mentioned names in the raw data. They are the subject of the question, not answers to it, so they are reported separately rather than ranked.
• 10 questions is a small sample, chosen by us, in English only. Different phrasing surfaces different tools.
• We sell one of the products measured. We chose the questions and the vendor list. We have published our own zero and the full method so you can weigh that, but it is not a neutral third-party study and we are not going to call it one.
• Point in time. Collected 6 September 2026. AI answers move quickly; treat anything here as stale within months.

Related