Are AI Visibility Tools Visible in AI Search? We Tested 30 of Them.

We build an AI visibility scanner. So we did the obvious thing and pointed it at our own industry, including at ourselves.

We took 12 questions a real buyer would type when shopping for AI visibility software, ran each one against Perplexity and Claude web search five times, and recorded every source the engines cited. 120 calls, no failures, all on 26 July 2026. Then we asked which of 30 vendor websites showed up.

Our own site showed up zero times out of 120. More on that below.

The companies selling AI visibility are worse at it than the SEO incumbents

We sorted the 30 sites into three groups before running anything: AI-visibility specialists (16), established SEO platforms that have added AI-visibility features (10), and independent audit tools (4, including us).

The legacy SEO platforms are cited at roughly 2.6 times the rate of companies whose entire product is AI visibility.

Twelve of the sixteen specialists were never cited once across all 120 calls. We are not naming them. They are real companies with real products, several of which are better funded than we are, and a name-and-shame list is not the point. The point is the count: 12 of 16.

The vendors that were cited, we will happily name, because the result is theirs to keep and anyone can reproduce it:

Note the ceiling. The best-performing vendor in an entire category, on the twelve questions most central to that category, is cited in under a third of answers. Nobody is dominating this. Nobody is close.

Nobody owns this category

Across 120 calls the engines cited 278 distinct hosts. The single most-cited source across all twelve questions, siftly.ai, appears in 30.8% of answers. That is the top of the entire market, and it is a minority position.

For comparison, ask an answer engine about CRM software and you will get the same handful of names most of the time. This category has no such gravity yet. Whoever wants those citations can still take them.

The round-up articles and the answer engines disagree about who exists

We built our 30-site sample from 2026 best-AI-visibility-tools round-up articles, which is the same research any buyer would do. That sample missed the actual leaders.

siftly.ai (30.8%) and frase.io (28.3%) each outrank every single site in our sample, and neither was in it. Nor were useomnia.com (23.3%), nightwatch.io (16.7%), presenc.ai (12.5%), llmclicks.ai (12.5%), trysight.ai (10.8%), or cognizo.ai (9.2%).

If you are doing competitive research by reading listicles, you are looking at a different market than the one your buyers see when they ask an AI. That gap is not a rounding error. It is most of the leaderboard.

What actually gets cited is content, not products

The fourth-most-cited source in our results is amplitude.com, a product analytics company, at 25.8%. It outranks all but three of the vendors that actually sell AI visibility software. Also well represented: Zapier (20.8%), Reddit (20.0%), YouTube (15.8%), HubSpot's blog (12.5%), LinkedIn (11.7%).

And look closely at the winner. SE Ranking was cited in 38 of 120 calls. Their dedicated blog subdomain accounted for 23 of those on its own; the main domain for 19. (The two overlap, because some answers cited both.) The category leader's AI visibility is driven as much by their blog as by their product site.

The engines are not citing the best tool. They are citing the best page about the question. Those are very different assets, and most vendor sites are built almost entirely out of the first kind.

The two engines barely agree

Perplexity and Claude cite different sources for the same question, and the disagreement is not subtle:

• Rankability: 36.7% on Perplexity, 11.7% on Claude.
• Otterly.AI: 20.0% on Perplexity, 35.0% on Claude.
• In an earlier pilot on a single question, the two engines shared zero cited domains across four calls.

If you are tracking your AI visibility as one blended number, you are averaging across systems that fundamentally disagree. Split it by engine or the number means very little.

Our own number is zero

Fetchling was cited in 0 of 120 calls, on 0 of 12 questions. We are in the group of four independent audit tools that scored a collective zero.

We are including this for a reason beyond honesty. We build the tool that measures this, we ran the study, and we still lost, which is the whole argument we have been making. Being technically ready for AI crawlers and actually being cited are different problems. Our site is technically fine. It is cited nowhere. Readiness is table stakes, and almost nobody is measuring the part that matters.

We are going to re-run this exact study in 60 days, with the same 12 questions and the same 30 sites. This post is the only deliberate thing we will have changed. We will publish what happens either way.

Methodology

Reproduce it or pick it apart.

• Date: 26 July 2026, single session.
• Engines: Perplexity sonar, and Claude Sonnet 5 with web search via OpenRouter's web plugin.
• Runs: 5 per question, per engine. 12 x 2 x 5 = 120 calls, 0 failures.
• Why repeated runs: research on AI search citation stability found pairwise source agreement for identical prompts run minutes apart averages only about 32 to 43 percent (Schulte, Bleeker and Kaufmann, University of St. Gallen, April 2026, arXiv:2604.07585). A single check is close to noise. Every number here is a rate over repeated sampling.
• What counts as a citation: the engine returned the domain, or a subdomain of it, in its list of sources. A mention in prose without a linked source does not count.
• Failed calls are dropped, not scored as non-citations. There were none.
• Sample: 30 public vendor sites, all verified reachable that day. No customer data was used anywhere in this study.

The 12 questions, verbatim:

• What tools can I use to see if my brand is being cited by ChatGPT and Perplexity?
• How do I track whether AI search engines mention my website?
• What is generative engine optimization and what software helps with it?
• What are the best AI visibility tracking platforms right now?
• Which AI search monitoring tool is best for a small marketing agency?
• What are the alternatives to enterprise AI brand monitoring platforms for a small budget?
• My website ranks well on Google but never shows up in AI answers. What should I do?
• How can I audit my site for AI crawler accessibility and llms.txt support?
• How do I find out which websites ChatGPT cites for questions in my industry?
• How much do AI visibility monitoring tools cost per month?
• Is there a free tool to check my AI search visibility?
• What should an agency use to give clients an AI visibility report?

No question names a vendor. A branded question tests navigation, not discovery, and hands that vendor a guaranteed citation.

Limits, before you quote the numbers

• Two engines. No ChatGPT, no Gemini, no Google AI Overviews. Everything here describes Perplexity and Claude web search specifically.
• Five runs is thin for Claude. Its run-to-run variance was visibly higher than Perplexity's. The gap between groups is wide enough to survive that; any single vendor's individual rate is softer than the decimal places suggest.
• Two of the twelve questions name ChatGPT or Perplexity. They are genuine buyer phrasings, but asking Perplexity a question containing the word Perplexity may bias its retrieval.
• One category, one day, US English. Answer engine indexes move.
• The sample is judgment-selected, assembled from round-up articles, a limitation this study happens to demonstrate rather dramatically.
• We are not a neutral party. We sell a product in this category, we chose the questions, and we are in the sample. We put ourselves in at zero and published the prompts so you can check us.