Science-backed data on why AI says what it says
AI-visibility research · Free instruments · No signup

AI decides who gets recommended. We study why and show you what it sees on your site.

We distill the research on why AI says what it says, how generative engines choose their sources, and where brands disappear. Then we built the instrument that shows what those systems can read when they reach your site.

Free, about 10 seconds, shows its work. Or read the research first ↓

Fetches as · GPTBot · ClaudeBot · PerplexityBot · OAI-SearchBot · Google-Extended · Applebot · + more

From the papers we've distilled

What the research actually says

Every number links to its paper · updated as we publish
94% / 40% LLMs ace multiple-choice cultural knowledge, then stumble when answer choices disappear. Source · arXiv:2606.01879 89% A self-evolving LLM agent beat human fact-checkers on health misinformation. Source · arXiv:2606.02215 ~16% Across major answer engines, cited sources were AI-generated surprisingly often. Source · arXiv:2605.23684 12-38% Six frontier LLMs hallucinated scientific citations at double-digit rates. Source · arXiv:2605.14306 21-32 pts Cosmetic rewording changed AI brand recommendations more than switching models. Source · arXiv:2605.27440 48-52% Lower-stature specialists stayed invisible in RAG while leaders still surfaced. Source · arXiv:2605.27439 65.7% Schema.org metadata raised FAIR-compliant precision, but cut query coverage. Source · arXiv:2605.28787 65% → 18% One false top search result collapsed frontier agent accuracy in a Microsoft study. Source · arXiv:2603.00801 And what does AI see on your site? Run the free instrument. No signup, no score theater, no gate. Develop your page →
The research, distilled

Latest digests

All articles →
01 Jun 25, 20269 minarXiv:2606.25787 LLMs mostly source "brand reputation" from other people's pages What LLMs treat as brand reputation is mostly other people's web pages, not the brands themselves. 02 Jun 23, 202610 minarXiv:2606.23057 AI brand "ownership" is moderately concentrated, but the winner changes by model Who gets the top pick in AI recommendations, and how consistent is it across models? 03 Jun 23, 202610 minarXiv:2606.23165 English-only AI reputation monitoring misses local champions in multilingual markets English-language prompts create a measurable local-visibility blind spot across languages.
How we work

Honesty is the product

SHOW THE WORK

Every digest links the paper it came from. Every checker result cites its source. If we cannot source a claim, we do not say it.

SAY WHAT DOESN'T WORK

The checker tells users llms.txt is hygiene, not a citation lever, citing SE Ranking's 300k-domain study instead of selling a generator.

NO GATES

No signup, no email wall, no book-a-demo trap. Read it, use it, act on it, leave.

Now you have no excuses not to act on it.