Challenge · AI search visibility
Will AI-written content get you penalized?
It is the first question every marketing team asks before it lets a tool near the website, and the answer people repeat to each other is usually wrong in both directions. Some believe any AI involvement is fatal. Others believe nothing is. Neither survives reading what the search engines actually published.
By Priyadharshini Krishnan, Co-Founder & Chief AI Officer · 17 September 2026 · 7 minute read
Search engines do not penalize a page for having been written with AI. What they act on is content produced at scale with nothing added for the reader, and the published guidance says the test is whether the page helps the person who landed on it. That means the question worth asking about a draft is not how it was produced. It is whether anyone could have written this particular page: does it carry your own numbers, your own customer situations, a named author who knows the subject, and an edit that cut everything generic. A draft that fails those is a risk whoever typed it. A draft that passes them is a page, and the engines have no reliable way to tell you otherwise.
What the search engines actually said
Google's position has been public and unchanged for years: how content is produced is not what decides whether it ranks. Quality does. The same guidance names the thing that does get acted on, which is content generated at scale to manipulate rankings without adding value for the reader. That behavior has a name in the spam policies and it long predates language models. Article spinners did it. Content farms did it. A model just makes it cheaper.
So the line is not drawn where most teams assume. It runs between a page that earns its place and a page that exists to fill a slot in a spreadsheet, and both sides of that line can be written by a person or by a model.
The practical version
If you could publish the page under your own name, with your own author byline, and defend every claim in it to a customer, the way it was drafted is not the risk. If you could not, the way it was drafted is not the risk either. The page is.
What makes a page read as machine-written
There is a recognizable shape to unedited output, and readers spot it before any algorithm does. Three things account for most of it.
- No author who could have written it. Generic advice with no name attached, or a name with no reason to know the subject. A model cannot supply the twelve years somebody spent in your category.
- No specifics only you could supply. Your own figures, a customer situation you actually handled, the thing you got wrong last year. This is the material a model has no access to, which is exactly why its absence is the tell.
- Sentences that could sit on any competitor's site unchanged. If swapping the logo changes nothing, the page is not about you, and an answer engine looking for something to attribute to you finds nothing to hold on to.
Notice that all three are also the reasons a page fails to get cited. The work that keeps you out of the penalty conversation is the same work that gets you into answers, which is a convenient thing to be true.
What to do with a draft before it goes live
- Put a real author on it, one who could answer a question about it on a call.
- Add at least one thing only your company knows. A number from your own account, a situation from your own customers, a decision you made and why.
- Cut every sentence that would be true of any company in your category.
- Check the claims. A draft that cites something is only as good as the thing it cited, and this is the step people skip.
- Publish it on your own domain, under your own name, where you have to stand behind it.
The question is never how the first draft appeared. It is whether anybody added anything before it went out.
Where the volume argument actually bites
The risk is not one page. It is the decision that follows the first page, when a tool makes drafting cheap and somebody works out that two hundred pages cost the same as two. That is the behavior the spam policies describe. It is a decision a person makes, and the software has no part in making it.
A practical guard: tie publishing volume to how many pages a person on your side can genuinely edit and sign off in a week. If the number of drafts exceeds that, the constraint is not the tool.
If you are an agency
Clients ask this in the first meeting and the answer decides whether the engagement goes anywhere. Saying "AI content is fine" loses the room, and so does saying it is dangerous. What holds up is the distinction: the engines act on scaled content with nothing added, the test is whether the page helps the reader, and your process is what puts the page on the right side of that. It also sets the expectation you want, which is that drafting gets faster and reviewing does not go away.
How we do this ourselves
Every draft Signal AI Visibility Optimization (AIVO) produces is written against your Brand DNA, your evidence and your competitor set, so it starts with the specifics a generic draft is missing. A person on our side edits before it reaches you, and nothing goes live until somebody on your side says so. We produce the brief, the draft and the structured data. You publish, because it goes on your domain and should read like your brand.