AI Content Ranks on Google. I Built a 100% AI Site to Prove It.

Every week someone tells me AI content cannot rank. An agency says it in a pitch. A LinkedIn post says it under a screenshot of somebody else’s traffic drop. The claim has been repeated so often that it gets treated as settled, and almost nobody repeating it has tested it.

Read the transcript

Every week someone tells me AI content can't rank. Almost nobody saying it has actually tested it. I did. In April I launched a brand new site, every page written by AI. By September: 1,422 clicks, nearly 97,000 impressions, rankings for more than a thousand queries, and not one link bought. Google put this in writing back in 2023. Appropriate use of AI or automation is not against our guidelines. In 2024 Google rewrote its spam policy to target content produced at scale to boost rankings, whether automation, humans or a combination are involved. Behavior, not the tool. Think of a building inspector. He checks the foundation, the wiring, the roof. He never asks if you used a hammer or a nail gun. Google grades what a page does for the reader, not who typed it. The data agrees. Ahrefs checked 600,000 top-ranking pages. 86.5 percent contained AI content. Three Google updates hit my site in its first four months. What happened next is below.

I did. In late April 2026 I launched a brand new domain in a consumer hobby niche. No history, no authority, no audience, no links. Every page on the site was produced with AI: Claude for the research and the writing, Replit for the build, one person at the keyboard. Then I watched it in Google Search Console every day for four months while Google rolled out a core update and two spam updates.

Between launch and September 1 the site earned 1,422 clicks and 96,913 impressions from Google Search. It shows up for more than 1,000 different queries. Of the 171 keywords I track for it in Ahrefs, 52 were in the top 10 by the end of August, up from zero in May, and five of them sit in the top three. Its Ahrefs Domain Rating is 0. It has zero dofollow backlinks. I did not buy a single link.

The site is not named in this post, for reasons I explain further down, but the numbers are real and the method is at the bottom. If you want the name and the full playbook, send me a private message on LinkedIn. Here is what happened, and more importantly, why Google’s ranking systems let it happen.

What Google actually says about AI content

Start with the source material, because most of the “Google penalizes AI” crowd has never read it.

In February 2023 Google published its guidance on AI-generated content. The section that matters is titled “Rewarding high-quality content, however it is produced.” Inside it, Google wrote: “Appropriate use of AI or automation is not against our guidelines. This means that it is not used to generate content primarily to manipulate search rankings, which is against our spam policies.” It noted that automation has produced helpful content for years, including sports scores, weather forecasts and transcripts. And it drew the comparison that ends most of these arguments, pointing out that “about 10 years ago, there were understandable concerns about a rise in mass-produced yet human-generated content. No one would have thought it reasonable for us to declare a ban on all human-generated content in response.”

In March 2024 Google rewrote the rule that people still cite as the “auto-generated content penalty.” The replacement is called scaled content abuse, and Google’s announcement describes the target as “producing content at scale to boost search ranking” and adds that it applies “whether automation, humans or a combination are involved.” The policy stopped being about the tool and became about the behavior. A room full of freelancers cranking out 5,000 thin pages breaks it. One person using AI to build 1,000 pages that answer real questions does not.

Google’s spam policies page, last updated August 28, 2026, lists “using generative AI tools or other similar tools to generate many pages without adding value for users” as its first example of scaled content abuse. The operative words are “without adding value for users.” The tool is named as an example of how people do it, not as the offense.

The people who grade Google’s results have the same instruction. In January 2025 Google added generative AI to its Search Quality Rater Guidelines for the first time. The Lowest rating goes to pages where “all or almost all of the MC on the page … is copied, paraphrased, embedded, auto or AI generated, or reposted from other sources with little to no effort, little to no originality, and little to no added value for visitors to the website.” AI generated sits in the same list as copied and paraphrased. The test is the back half of the sentence.

Then in May 2026 Google widened the definition of spam itself. The opening line of the spam policies now covers “attempting to manipulate generative AI responses in Google Search,” which pulls AI Overviews and AI Mode under the same enforcement as the classic results. Both spam updates my site grew through rolled out after that change. What the overview rewards once a page ranks is a separate question, and I measured it across 400 searches in how to rank in AI Overviews.

So the written rules have been consistent from 2023 through this month. Google grades what a page does for the person who lands on it. It does not grade who, or what, typed it.

Why Google does not run an AI filter, and could not if it tried

There are two practical reasons the policy reads the way it does.

The first is that Google is now one of the largest publishers of machine-written text on the web. At I/O in May 2026, Google said AI Overviews have more than 2.5 billion monthly active users and AI Mode passed 1 billion in its first year. Those answers are written by a model, from web content. A company whose flagship results page is machine-written cannot treat machine-written text as spam by definition. It has to judge by quality, because quality is the only test its own product survives.

The second is that detection does not work at web scale. AI detectors flag human writing as AI, and wave through lightly edited AI writing, often enough that OpenAI shut down its own AI text classifier in July 2023 over its low accuracy. If Google demoted pages on a detector’s say-so, it would be demoting real publishers every day. What Google relies on instead are signals: whether the page satisfies the query, whether people bounce back to the results and try again, whether other sites reference it, whether the site has a track record of being useful. Google’s own guide to its ranking systems says the helpful content system “evolved and became part of our core ranking systems” in March 2024, so “is this helpful” is no longer a filter bolted on after the fact. It is the ranking algorithm, and none of it needs to know how the text was produced.

The third-party data lines up with this. In July 2025 Ahrefs analyzed 600,000 top-ranking pages across 100,000 keywords and found that 86.5% contained some AI-generated content, 4.6% were entirely AI-generated, and the correlation between a page’s share of AI content and its ranking position was 0.011, which is zero for practical purposes. In July 2026 they repeated the exercise on a million pages across 100,000 search results. In positions one through three, 5.3% of pages were 100% AI-generated and 9% were at least 80% AI. Every position on page one held between 8.4% and 11.7% pages that were mostly AI. Ahrefs’ conclusion: “There are no obvious hard cutoffs suggesting a binary AI classifier is preventing AI-generated pages from ranking highly or making it into the index.”

The same study found something the “AI content is free traffic” crowd should read too. Pages with low or moderate AI content earned two to three times the impressions of pages that were mostly AI, and mostly-AI pages were indexed at a lower rate, 40% against 49%. That is not an AI penalty. That is a quality gradient, and most mass-produced AI pages sit at the bottom of it.

Three Google updates, one AI site, zero drawdowns

Google posts the start and end of every confirmed ranking update on its Search Status Dashboard. Three landed on my site in its first four months. For each one I compared the seven days before the rollout started against the seven days after it ended, using the daily Search Console export.

Read the transcript

Google posts every confirmed ranking update on its Search Status Dashboard. Three hit my AI-written site in its first four months. Here is the daily click chart, with each rollout marked. First, the May core update. A core update is Google re-grading content quality across its whole index. The site was three weeks old. Impressions went from 25 a day to 204. Clicks went from under one a day to four. Then the June spam update. A spam update is a sieve, not a scanner. It catches thin pages, scraped pages, doorway pages. It does not ask who wrote them. My site sped up through it. Clicks up 34 percent. Impressions up 43 percent. August is the strongest data point I have. Impressions stayed flat. Clicks more than doubled, from 14 a day to 33. Same visibility, higher positions. Google re-ranked the site upward in the middle of an enforcement rollout. That same week, Glenn Gabe documented a programmatic AI site losing rankings across more than 200,000 queries. Same storm, different boats. Those sites were built to game rankings. Mine was built to answer questions. The tool was the same. The behavior was not. Ahrefs studied a million top-ten pages this summer. About one in twenty top-three results was written entirely by AI, with no cutoff anywhere. And Google's own AI Overviews reach two and a half billion people a month. Google cannot call AI text spam without calling its own results spam. Month by month, clicks went 23, 187, 448, 761. No update produced a dip. Google grades what a page does for the person who lands on it, not who typed it. People lie. Results don't.

May 2026 core update, May 21 to June 2. Google described it as “a regular update designed to better surface relevant, satisfying content for searchers from all types of sites.” A core update is a broad re-evaluation of content quality across the whole index, and my site was three weeks old when it hit. Impressions went from 25 a day the week before to 204 a day the week after, up 714%. Clicks went from 0.7 a day to 4.0. The first quality review Google ran on an entirely AI-written site accelerated it.

June 2026 spam update, June 24 to June 26. Spam updates are a different animal. Google describes them as improvements to its automated spam-detection systems, SpamBrain among them, and those systems enforce the spam policies, which explicitly list AI-generated pages that add no value. This is the direct test for an AI site. Clicks went from 7.4 a day to 9.9, up 34%. Impressions went from 566 a day to 812, up 43%. The site passed the filter built to catch auto-generated spam and kept climbing.

August 2026 spam update, August 18 to August 21. This is the strongest single data point I have. Impressions stayed flat, roughly 1,690 a day before and 1,640 after. Clicks went from 14.1 a day to 33.1, up 135%. Click-through rate went from 0.84% to 2.02%. Same visibility, more than double the clicks. The site did not pick up new queries; it moved up on the queries it already appeared for. Google re-ranked existing visibility in the site’s favor during an enforcement rollout.

For context, the August update was brutal for the wrong kind of AI site. Glenn Gabe documented a programmatic, AI-generated site in what he calls an “ultra-YMYL niche” (Google’s shorthand for topics that touch health, money or safety) losing rankings across more than 200,000 queries, a thin affiliate operation losing more than 14,000, and a site with more than 1.5 million indexed URLs, about 85% of them programmatic, getting cut down. Same update, same week, opposite outcome. Those sites and mine were all built with AI. The difference was whether the pages did anything for the person who landed on them.

Month by month, clicks went 23 in May, 187 in June, 448 in July and 761 in August. By the last week of August the seven-day average was 42 clicks a day. No update produced a dip. That is the shape of a site the algorithm trusts, not one riding a loophole that has not been closed yet.

A second site, same result, different industry

The hobby site is the clean experiment because it started from nothing. I have a second data point from the opposite end of the web: a family-owned industrial equipment distributor, in business since the 1970s with branches across the Midwest, whose website was due for a rebuild.

I rebuilt that site with AI tools in Replit and relaunched it in early August 2026. Over the last four full weeks of July, the old site averaged about 12,700 impressions and 170 clicks a week in Search Console. Over the first four full weeks of August, the new site averaged about 31,800 impressions and 240 clicks a week, with average position holding steady. That window includes the August spam update. The week the update rolled out, the site logged 30,812 impressions and 227 clicks, above anything it did in July.

A rebuild is not a pure AI content test the way the hobby site is, because the company’s expertise and product catalog carried over. But it answers a question clients ask me constantly: will Google punish us if we rebuild with AI? The numbers say no. Impressions more than doubling while position held means the site began appearing for queries it never showed up for before, in the same month Google ran a spam update. It was also generating leads within weeks, which is the number the owner cares about.

What this site did that the penalized sites did not

I am not going to publish the full playbook, because the site is a live business and the niche is small enough that the strategy would give it away. But the overarching approach is not a secret, and every piece of it maps onto the rules above.

The pages that rank best are built on live data. Each one answers a question people actually type into Google with a real number from a real source, refreshed as the data changes, instead of an opinion paragraph a model wrote from memory. That is the “added value” test from the quality rater guidelines, answered in the most literal way possible: the page contains information the reader cannot get from the pages above it.

The editorial blog grew from roughly 150 posts at launch to more than 1,000, but it was released on a schedule over four months, not dumped in a weekend. Scaled content abuse is about generating “many pages without adding value.” Volume by itself was never the offense. Volume without value is. Every post had a distinct question to answer and a place in the site’s structure.

An internal link mesh ties every related page to its neighbors so nothing is orphaned. A thousand pages that link to each other by topic look like a reference library. A thousand pages that only link to the homepage look like a content farm.

And a few things stayed out on purpose. No paid links, not one placement, sponsored post or “guest post package.” No link swaps, private blog networks, expired domains or redirected authority. No cloaking, hidden text, doorway pages or scraped content. No pages parked on someone else’s domain. The June spam update looks for exactly those things, and the site sped up through it. The Ahrefs link profile on September 3 showed a Domain Rating of 0, a few dozen nofollow mentions and not one dofollow link. Whatever moved this site, it was not link equity.

That last point is worth sitting with. For most of my 22 years in this business, the standard advice for a new site was to buy or build links, because content alone could not carry a domain with no authority. This site ranks for more than 1,000 queries on content and structure alone. Links still matter in competitive spaces, but the floor for what one person with the right tools can do without them moved this year.

The question Google is actually asking

Glenn Gabe put it the right way after the August update: “If anyone can set up a site programmatically like this with almost no human involvement at all, what value is there?” That is the question Google’s systems are built to answer, and it has been the question since long before ChatGPT existed. AI did not change what Google penalizes. It made the penalized thing cheaper to produce, so the enforcement got louder and more frequent.

Danny Sullivan, who spent years as Google’s Search Liaison and still works on the Search team, said the same thing from the other direction on the Search Off the Record podcast in January 2026. People kept asking whether they should chop content into bite-sized chunks because “LLMs like things that are really bite size.” His answer: “We don’t want you to do that. I was talking to some engineers about that. We don’t want you to do that. We really don’t.” Content written for a machine, whether the machine is producing it or consuming it, is the thing Google keeps saying it does not want. Content written for a person, by whatever means, is the thing it keeps rewarding.

So when someone tells you AI content is unrankable, ask to see their Search Console data. Then ask what their site does for the reader. The answer to the second question predicts the first.

What I tell clients about AI content

AI is fine as the writer. It is a poor strategist and a worse editor unless you make it one. Bring the specifics: the data, the numbers, the sources, the structure, the schedule. Decide what each page is for before a model writes a word of it. Release on a cadence you can sustain. Do not buy links to “make up” for AI content; that adds a real policy violation to a page that did not have one. And watch Search Console every day with Google’s update calendar next to it, because that is the only way to read cause and effect instead of guessing.

I have done SEO since 2004, and I will tell you plainly that nobody is an expert in AI and SEO right now. The ground moved and it is still moving. Be careful with anyone selling certainty, because they are usually selling something else. What I can offer is data from my own sites and the willingness to show you all of it.

People lie. Results don’t. If you want the name of the site, the stack and the playbook, send me a private message on LinkedIn. If you want to know what this would look like for your business, book a free SEO consultation and I will tell you, including whether it is the right spend.

Method. Daily data exported from Google Search Console for the property (name withheld), April 27 to August 31, 2026, with the partial day of September 1 excluded. Before and after figures compare the seven days preceding a rollout start with the seven days following the rollout end. Update windows are the rollout dates Google posted on its Search Status Dashboard: May 2026 core update (May 21 to June 2), June 2026 spam update (June 24 to 26) and August 2026 spam update (August 18 to 21). Keyword counts are from Ahrefs Rank Tracker, 171 tracked keywords, desktop, August 31 snapshot. Link profile from Ahrefs Site Explorer, September 3, 2026. The distributor figures are weekly Search Console totals for the four full weeks of July and the first four full weeks of August 2026. Correlation across three independent windows with zero negative outcomes is the claim here; no single window proves causation on its own.