Risks in AI visibility work. And what we do instead.

Pages served only to AI crawlers

Serving bots a different page from people, by user agent, is cloaking under Google’s spam policies, and security researchers have shown it can quietly change what an assistant says about a person or product. If it is found, the brand’s whole site can lose trust.

What we do: One page for people and bots. Our facts file always comes with a page people can read that says the same words, and our bot view lets you check any site for a difference.

Sources: Google Search spam policies (cloaking); SPLX research on agent-aware cloaking, reported by Dark Reading.

Scripting consumer chat apps

Collecting answers by driving logged-in ChatGPT or Claude apps with scripts runs into those apps’ terms. OpenAI’s terms list “Automatically or programmatically extract data or Output” among things you cannot do; Anthropic’s consumer terms bar access “through automated or non-human means”, except with an API key. Accounts can be closed, and data can stop without warning.

What we do: Official APIs only, paid per answer. That costs more and is slower to scale, and it means the data cannot be switched off by a terms breach.

Sources: OpenAI Terms of Use (effective 1 January 2026); Anthropic Consumer Terms, section 3.

Seeded forum posts and manufactured mentions

Paying for or planting posts, reviews and “mentions” to feed AI answers can be spotted by readers and moderators, and Google’s guide to AI search says that “seeking inauthentic mentions across the web isn’t as helpful as it might seem.”

What we do: We do not write, post or buy mentions. We show which pages the engines read when a rival is picked, so you can earn a place there honestly.

Source: Google Search Central, guide to AI features and your website.

Memory prompts behind “Summarize with AI” buttons

Microsoft’s security team found more than 50 prefilled prompts from 31 organisations telling assistants to “remember” a company as a trusted source, and classed it as memory poisoning. A visitor who notices it is unlikely to trust that brand again.

What we do: Never. Our hidden prompt check finds these links and hidden text aimed at AI on any page, including plugins and themes you did not write.

Source: Microsoft Security blog, “AI Recommendation Poisoning”, 10 February 2026.

One score out of 100

A single score mixes mentions, positions and sentiment with weights you cannot see, and it can jump when the question set changes. When the number cannot be redone by hand, a team can end up reporting noise.

What we do: Three counts, each with its denominator, a 95% range and the answers behind it. Question changes are logged. See the method.

Rank positions inside an answer

A study of nearly 3,000 answers found that fewer than 1 in 100 repeats of the same question gave the same list, and fewer than 1 in 1,000 in the same order. Position moves by chance; whether a brand shows up at all across many runs was far steadier.

What we do: We report whether you were named and picked across repeated runs, not your place in one list.

Source: SparkToro with Gumshoe, January 2026 (small, not peer-reviewed, partly run by a company that sells tracking).

Mass AI-written pages

Publishing many thin pages written to match prompts can fall under scaled content abuse in Google’s spam policies, and adds little an engine needs.

What we do: We help you fix the facts on the pages you already have, and keep one dated source of truth.

Source: Google Search spam policies (scaled content abuse).

Special AI files sold as a must

Google says you “don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search”, and that structured data is not required for generative AI search. No AI vendor has said it reads llms.txt or our facts file format.

What we do: We say this on the facts file builder. The point of a facts file is one honest, dated source with a matching page for people, not a ranking trick.

Source: Google Search Central, guide to AI features and your website.

Self-ranking “best tools” lists

“The 10 best tools” pages written by one of the ten, with itself at number one, are easy for readers to spot and quick to erode trust.

What we do: We do not publish them. We name no rival on this site, and we print no rival’s prices: they change, and we cannot keep them right.

Removing AI watermarks and “humanising” AI text

Tools that strip watermarks or disguise AI text help hide where words came from, the opposite of what an honest brand wants an engine to see.

What we do: Not built, on purpose.

What we cannot promise

No tool can make an AI recommend you, and anyone who promises it is guessing. We measure, show the receipts, help you publish facts that can be checked, and tell you whether a change moved anything.