Leftover AI phrases: find and remove them

What it means

We found text in the page body that reads like leftover output from an AI chat assistant — phrases such as "As an AI language model" or "my knowledge cutoff is …" that a person would never write on purpose. This is a deterministic match on the literal text, not a guess about how the page was written.

Why it matters

Google judges content by value, not by whether AI was used — but leftover assistant phrases are one of the specific "tells" its Search Quality Rater Guidelines cite as a signal of unedited, low-value content, which can be rated Lowest. Their presence means nobody read and edited the page before publishing, and that lack of care is exactly what the guidelines penalize.

How to fix it

1. Read the page and delete any leftover assistant phrasing ("As an AI language model", "I cannot browse the internet", knowledge-cutoff admissions).
2. Edit the surrounding text so it reflects real, first-hand knowledge and adds something a reader can't get elsewhere.
3. Add a review step to your publishing workflow so drafts are read by a person before they go live.

When it's not a problem

If the phrase appears because the page is legitimately about AI assistants (e.g. a tutorial quoting a model's refusal), it's not a defect — it's quoted, in context, and clearly attributed.

How common is it?

4 audited sites in our corpus currently show this issue. The breakdowns below show which platforms, gatekeepers, verticals and countries are most exposed.