Does AI-written content rank on Google?

The short answer

The short answer Yes. In a June 2026 Ahrefs study of 1 million top-10 pages, 5.3 percent of positions one through three were fully AI-generated and 9 percent were at least 80 percent AI. But pages under 50 percent AI got two to three times the impressions. Google penalizes scaled low-value content, not AI itself. What the data.

The short answer

Yes. In a June 2026 Ahrefs study of 1 million top-10 pages, 5.3 percent of positions one through three were fully AI-generated and 9 percent were at least 80 percent AI. But pages under 50 percent AI got two to three times the impressions. Google penalizes scaled low-value content, not AI itself.

What the data actually shows

Ahrefs sampled 1 million pages from the top 10 positions across 100,000 SERPs in June 2026, with roughly 331,000 analyzable and 150,000 carrying enough text for AI detection. The distribution across positions one to three:

AI content level	Share of positions 1–3
100% AI-generated	5.3%
80% or more AI	9.0%
Under 50% AI	82.2%
Under 20% AI	54.7%

Fully AI-generated pages rank in the top three. That settles the direct question. But the gradient matters more than the headline. Average AI share was 27.1 percent at position one and 30.9 percent at position 10, rising steadily down the page. Indexation followed the same pattern: 49.28 percent of low-AI pages were indexed against 40.35 percent of very-high-AI pages. And pages with low or moderate AI content received two to three times the impressions of high or very-high AI pages.

The study’s own conclusion: Google is not trying to punish AI-generated content, it is applying the same quality signals it always has. Scaled AI output tends to be shallow, uncited and unoriginal, and that is what underperforms.

Context for why this question now matters to everyone: AI drafting is no longer a minority practice. Ahrefs’ compiled 2026 SEO statistics put 87 percent of marketers using AI for content creation and 74 percent of newly published web content containing AI-generated material. If AI origin were disqualifying, three quarters of the web would be. It is not.

What the policy says

Google’s spam policies do not classify AI content as spam. They define scaled content abuse as generating many pages “for the primary purpose of manipulating search rankings and not helping users,” and list “using generative AI tools or other similar tools to generate many pages without adding value for users” as an example of it.

Read that clause carefully. The violation has two conditions: many pages, and no added value. One well-researched AI-assisted page is not scaled content abuse. Two hundred templated pages with the city name swapped are, regardless of who or what wrote them.

The rule for where you fall

AI-assisted, human-directed, cited: publish it. If a person set the argument, supplied the specifics, verified the numbers and edited the output, the origin of the first draft is not a ranking problem. The Ahrefs distribution shows most top-ranking pages sit under 50 percent AI, which is roughly what “assisted” looks like when measured.

Fully generated, generic, uncited: expect the ceiling. It can rank. It indexes less reliably and earns two to three times fewer impressions. In a competitive category that gap is the whole game.

Generated at volume with no differentiation: expect enforcement. This is the pattern the policy names.

There is a second reason to add specifics beyond ranking. Research from Princeton, IIT Delhi and Georgia Tech built GEO-bench across 10,000 queries and 25 domains and measured what lifts a source’s visibility inside AI-generated answers: statistics +25.2 percent, quotations +27.2 percent, cited sources +24.6 percent, fluency +24.7 percent. Keyword stuffing scored below the unoptimized baseline. The things that make AI-assisted content rank are the same things that make it get cited by assistants.

What to add to any AI draft before it ships

A number that only you have. Client data, a survey, an internal benchmark, a tested result. This is the single hardest thing for a competitor’s model to reproduce.
Cited external figures with live links. Not “studies show.” A named source and a URL you actually checked.
A direct answer in the opening 40 to 60 words. The block an assistant lifts.
A judgment call. What you would do, and what you would not. Models generate consensus by construction; a stated position is differentiation.
Fact verification. Every statistic, date and name checked against the source. Fabricated citations are the fastest way to lose a page’s credibility with both readers and reviewers.

What we’d do

TACK™ are AI integrators, not trainers, and we use these tools in production every day. Our rule on content is simple: AI accelerates structure and drafting, humans supply the specifics and the position. Every page carries named figures with real citations, opens with a direct answer, and states what we would actually recommend.

Measured on indexed status, impressions by query theme, and citation share across assistants, not on word count or publishing cadence. That is Get Found, and it hands to Get Leads once the traffic arrives. See the full capabilities or the case studies. Engagements start at $5,000 per month.

Common mistakes

Treating detection scores as the target. AI detectors are unreliable and Google does not run yours. Optimize for specificity, not for a percentage on a tool.

Scaling before validating. Publish ten pages, watch indexation and impressions for 60 days, then scale what worked. The policy risk is entirely in the volume.

Publishing unverified statistics. Language models fabricate plausible numbers and sources. Every figure needs a URL a human opened.

Assuming length substitutes for substance. A padded 3,000-word page is worse than a specific 1,000-word one on both search and AI surfaces.

The bottom line

AI-written content ranks, including at positions one to three. It ranks worse on average because it is usually thinner, and it indexes less reliably. Use AI for drafting, supply the specifics and the position yourself, and never publish a number you have not verified.

If you want a content program that uses AI properly without inheriting its ceiling, book a call with TACK™ or call 310-620-1141.