Is AI content bad for SEO?

What Google says

Two documents cover it.

Google's guidance on AI-generated content, from February 2023, says its focus is on the "quality of content, rather than how content is produced". It adds that using automation, AI included, "with the primary purpose of manipulating ranking in search results" breaks its spam policies, and that not all use of automation is spam.

The spam policies name the offence: scaled content abuse, which is when "many pages are generated for the primary purpose of manipulating search rankings and not helping users", "no matter how it's created".

So the question Google asks is about purpose and usefulness. A hundred pages written by people to catch every variant of a keyword are covered. One article drafted by AI that answers a real question well is not.

What gets a site removed

Publishing at scale without substance. Otterly, an AI visibility company that publishes its experiments, built two sites of 1,000 AI-generated posts each. Google removed both from its index, one within 8 days and the other within 12 weeks, by its account. That is two sites in one test, and it matches what the policy says it is for.

We sell AI-written articles, so here is where we stand

Weigh this section knowing that. Our product drafts an article a day with AI. We think that is defensible when each article is tied to a question people search for, written from sources that were read, checked claim by claim, and held back when it is thin. We think it is indefensible when the aim is to fill a site. The difference is not the tool. It is whether anyone would miss the page if it were gone.

These guides were drafted with AI too, and every figure in them was checked against its source before publishing. We would sooner say so.

What gives an AI draft away

Before writing these guides we read 39 posts on the blogs of eight companies that sell AI SEO or AI visibility software. Many of them read as machine-drafted, and two of the companies say outright that their product writes their blog. We expected to find the famous phrases. We found few: "in today's digital landscape" and its relatives have been edited out of all but one of those blogs.

What remained was structural, and it was the same on every blog that read as machine-made:

• A stamped template. The same takeaways box, the same closing paragraph and the same call to action, word for word, on every post.
• The same block for every item in a list. Price, pros, cons, best for, twelve times over, each the same length.
• Precise numbers with no source. A percentage or a number of days stated as fact, with nothing behind it.
• Facts that disagree with themselves. One blog gave two different word counts and two different language counts for its own product in two posts. One priced a competitor in euros in the text and dollars in the table. One had a "5 best" list with six entries.
• Nothing first-hand. No test the author ran, nothing that went wrong, no one named.

The posts that read as written by people had the opposite: a dataset with its size and dates, an experiment that failed and what it cost, a section arguing against their own conclusion.

None of that is about detection. A reader notices a template after two posts, and a reader who catches one wrong fact stops trusting the rest. Those are the costs.

What to do with an AI draft

1. Trace every number to its source and open the source. How to fact-check an AI-written article has the method. Three claims in our own first drafts were wrong.
2. Add the thing only you know. A price you were quoted, a result you measured, a mistake you made. If you have nothing, the article may not need to exist.
3. Delete the last paragraph. In a machine draft it restates the article. Then delete the last sentence of each section for the same reason.
4. Break the pattern. If every section is the same length and every bullet opens in bold, rewrite until the shape follows the content.
5. Check it against your other posts. The template only shows when you read three in a row.

We would not spend time on AI detectors or "humanizer" tools. Google's policy is about purpose and quality, and rewording a draft changes neither.