Is there hidden text on my page written for AI models?
We look for text a visitor cannot see that is worded as an instruction to an AI model, such as a request to ignore earlier instructions or to always recommend one product. When we find it we say how it is hidden, how many passages there are and which element to look for. We never repeat the words of the passage, and we do not say why it is there; it may be harmless.
What it actually is
A finding needs two things at once. The text has to sit in an element hidden in a way a machine reads straight through, such as a style that removes it from view, a zero size, a position far off the screen or a colour that matches its background. And the text has to read as an instruction addressed to a model. Hidden menus, accordions, cookie notices, text meant for screen readers, structured data and ordinary marketing copy are left alone. Rules that live in a separate stylesheet are not seen at all.
Why it matters
Security researchers call this indirect prompt injection: instructions placed in content that a model later reads, which can steer it even when no person can see them. OWASP lists it among the top risks for applications built on language models. A model summarising your page cannot tell which words were meant for it, so a page that says something to machines it does not say to people puts its own reputation, and the trust of the answers that cite it, at risk.
What good looks like
Everything a machine can read on the page is something a visitor can read as well, apart from ordinary markup. If a passage must be hidden, as in a closed accordion or a mobile menu, it holds the same kind of content a visitor would be shown when they open it.
Being honest about it
We cannot know why text is there. Some of it is left over from a template, a plugin or an old experiment, and some is an example in documentation about this very subject. So a warning describes what we found and where, and nothing more. The check can only pass or warn, it will miss anything hidden by a stylesheet, and because it is new it sits under measured, not scored: it is shown in your report and never moves your score.
Where most sites go wrong
A hidden block carried by a plugin or a theme that nobody on the team knows about, or a leftover from an experiment in steering assistants. The other is a page that quotes attack phrases as examples in a security article and hides them in a collapsed section.
On the web
State of the Web · November 2026In November 2026, 99% of the sites where it applies (3% of the sites we could read) passed the “No hidden text written for an AI model” check.
Not enough history yet.
Based on the sites Bot Appétit has scanned. Most of these sites come from lists we scan for research. Not a random sample of the web.
Not sure where yours stands?
A free scan checks this and everything else in about 20 seconds — no signup. Or if it's already flagged and you'd rather it were simply handled, we do that too.
Flagged for this one? We handle it. No pitch, just a plain-English chat.
Book a call