5.3% of Google's Top Three Results Are Fully AI-Written. The Number Your Dealership Should Worry About Is 40.35%.

Last week the same study got two headlines that point in opposite directions.

Ahrefs published its own research under the title "Google Doesn't Punish AI Content; It Punishes Bad Content." Four days later, Search Engine Journal's weekly roundup led with "AI-Detected Pages Rank Lower." Same dataset. Same numbers. Completely different message depending on which one crossed your desk.

Both are defensible readings, which is exactly the problem. If you run a store and your SEO vendor is publishing AI-assisted model pages, comparison posts, and blog filler on your domain every month, you need to know which reading applies to your invoice. So we went to the actual study.

The short version: the ranking panic is overblown, and there is a real finding underneath it that almost nobody led with.

What Ahrefs actually did

On July 27, Ryan Law and Xibeijia Guan published an analysis of 1,000,000 pages pulled from the top 10 positions across 100,000 search results in June 2026. About 300,000 of those pages existed in Ahrefs' crawler database, and 150,000 had enough text on them to run AI detection against. The organic performance slice covered 80,861 pages sorted into four bands by how much of the text scored as AI-written.

Credit where it is due, they front-load the caveat most vendors would bury. In their words: "AI content detection is not perfect, and the way we detect AI content will be different from how Google does (if it does)." Every number below is a detector's probability estimate, not a confession from the publisher.

Here is what the ranking data shows.

54.7% of pages ranking in the top three have less than 20% AI-detected text. Pages scoring under 50% AI account for 82.2% of all top-three positions. The web's best-performing pages are still mostly written by people.

But heavily AI-written pages rank fine too. 5.3% of top-three results scored 100% on AI detection. 9.0% scored 80% or higher. Those pages are not being scrubbed from the results.

And the slope from first place to tenth is almost flat. Average AI score at position 1 was 27.1%. At position 10 it was 30.9%. Pages scoring 80% or higher took 8.4% of first-place spots and 11.7% of tenth-place spots. Search Engine Journal, covering the study on July 28, called that correlation "meaningful but far from disqualifying." That is the right characterization. A three-point average difference across ten positions describes a mild lean, nothing more.

Ahrefs also found no hard cutoffs anywhere in the distribution, which is the tell that there is no binary AI classifier flipping a switch on your content. Law's own read: "I don't think Google is trying to punish AI-generated content; I think it is relying on the same old hallmarks of content quality that it always has." And more bluntly: "It's just that AI-generated content is usually lower quality than human-generated content."

So if the fear on your rooftop was that Google has an AI detector and will torch your rankings when it trips, the data says relax. That is not happening.

The number that should actually stop you

Ahrefs also measured something different from ranking: how many of these pages made it into Google's index at all.

Pages with the least AI text were found in the index 49.28% of the time. Pages scoring between 20% and 50% AI: 43.38%. Between 50% and 80%: 40.72%. Pages at 80% or higher: 40.35%.

Read those honestly, because the honest version is more useful than the scary version. Indexation is bad across the board here. Even the most human-written band did not quite reach half. So "60% of AI pages never get indexed" would be a dishonest way to state this, since the human-written pages sat at roughly 51% unindexed too.

The AI-attributable gap is about nine percentage points. That is the real finding, and it behaves like a tax rather than a cliff.

Ahrefs pairs it with an explanation borrowed from Dan Taylor: Google will burst-crawl a batch of new content, then throttle its crawl resources if the domain's authority does not justify continuing. Publish a lot, get sampled, then get metered.

And then comes the caveat that most coverage treated as a reason to dismiss the finding, which is precisely backwards for our industry: "Sites that lean heavily on AI content may tend to be newer or lower-authority in the first place." Ahrefs is being careful about causation, and they should be. But look at what that sentence describes. A mid-authority domain publishing content faster than its authority can carry.

That is a dealership website. That is the exact profile.

Why now, and why this reversed

There is one more piece of context worth holding, because it explains why this landed as news at all. SEJ's roundup notes that Ahrefs' own analysis a year earlier found a correlation of just 0.011 between AI score and ranking position, with no clear link.

A year ago: no relationship. Now: a gradual, consistent one across every band, plus an indexation gap. Nothing suggests Google built an AI detector in the interim. What changed is the volume and the average quality of what got published. When AI writing was a novelty, the people using it were mostly careful. Now it is the default production method for every content package sold to every small business in America, and the median AI page got worse.

Your dealership is buying into that median.

What this means for your dealership

The penalty you were worried about is the wrong worry. Nobody is going to demote your VDPs because a detector says your model overviews read as machine-written. Stop paying for "AI-content-safe" or "human-verified" line items sold on that fear. The threat model is wrong.

Getting indexed is the actual bottleneck, and volume is what breaks it. The vendor pitch that sounds like a bargain, 40 blog posts a month, a landing page for every model and trim and nearby city, is precisely the behavior that triggers the metering. You are spending budget to manufacture pages that have roughly a 40% chance of entering the index. Then you are reading a report that counts them as delivered.

Your domain authority is the constraint you cannot AI your way around. A franchise dealer site is competing against manufacturer pages, Cars.com, Edmunds, and CarGurus for the same queries. Those domains can publish sloppily and get indexed anyway. Yours cannot. The authority gap means the same page costs you more to get seen, which makes wasted output more expensive for you than for anyone else on the results page.

Fewer, better pages is now a measurable position rather than an aesthetic preference. Pages with less than 50% AI text hold 82.2% of top-three rankings. Use AI all you want. Just stop shipping the first draft it gives you.

AI in the workflow is fine. AI as the workflow is what shows up in the data. A page where a human decided what to say, brought your actual inventory and pricing and market into it, and used AI to speed up the writing does not look like the pages in the failing band. A page where a prompt produced the whole thing does.

What to do about it

1. Run an index check before you run a rankings report. Pull the last 90 days of pages your SEO vendor published and check Google Search Console's Pages report for how many are actually indexed. If it is under half, you have found where the money went. This one query settles more vendor arguments than any ranking dashboard.
2. Change the deliverable from page count to indexed page count. Rewrite the contract line. You pay for pages Google accepted, not pages a CMS accepted. Any vendor who resists that is telling you something.
3. Cut the volume and move the budget into depth. Ten pages that carry your real inventory, your real market, your real pricing, and your staff's actual answers will outperform forty spun variations of the same model overview. We made this argument before the data existed. It exists now.
4. Kill the near-duplicate city and trim pages first. These are the highest-volume, lowest-differentiation pages on most dealer sites, they are almost always AI-produced at scale, and they are the most likely to be sitting outside the index right now.
5. Put something on the page only you have. Your live inventory, your service pricing, your trade-in numbers, photos of your actual lot and staff. A model cannot generate it, a competitor cannot copy it, and it is the difference between the two bands in this study. Same principle we covered on OEM-vendor template sites and used car inventory pages.
6. Do not overcorrect into banning AI. The study is explicit that fully AI-written pages rank in the top three. Making your team write everything by hand fixes nothing if the output is still generic.

The honest version

One vendor, one detector, one month of data, and a correlation the authors themselves refuse to call causation. Ahrefs says so directly: "There are many reasons why AI-generated content might earn fewer impressions, so it would be wrong to assume that there is any kind of automatic suppression happening." Nobody has run this on dealership domains specifically, and the first team that does may find the numbers move.

But the direction is not really in dispute, and it is not even a new idea. Google has spent twenty years getting less generous with pages that do not earn their place. The only thing AI changed is how cheap it became to make those pages, and how many of them your vendor can put on your invoice.

The nine-point gap is the price of the shortcut. It is survivable. It is also entirely avoidable, and you are probably already paying it without seeing it on any report you receive.

If you want to know how many of the pages you bought this year actually made it into Google, book a walkthrough and we will pull the numbers with you.