Does Google Penalize AI Content? A Forum Just Found Out
Someone will offer you 50 blog posts a month before this quarter is out, and the price will look like a bargain against what a writer costs. Before you sign anything, look at what just happened to a forum that has been online for roughly 20 years. Google did not punish that site for using AI. It punished pages that answered questions without adding anything a real person knew, at a volume no human could have produced, and that distinction is the entire argument here.
After reading the news coverage, the panicked forum threads, and the studies, a business owner still cannot tell which side of the line their own blog sits on, or how to check. I can answer that from a live operation rather than a theory. My content is AI-assisted and human-directed: I set the angle, the experience in it comes from 17 years in search and 300+ businesses, and every claim gets checked against a real source before it publishes. That process is disclosed on my about page, and it gives me a test you can run on your own pages today.
Google flagged one URL pattern and buried 168,290 threads with it
Windowsforum.com received a partial manual action around July 17, 2026, code WNC-651700, labeled "Thin content with little or no added value," according to Roger Montti's report in Search Engine Journal on July 27, 2026. A partial manual action means a human reviewer at Google looked at a section of the site and demoted it by hand. The section was the URL pattern /threads/, which holds 168,290 threads and more than 500,000 posts. That is not a bad page. That is the entire library.
Translate it into your own business for a second. If your blog lives at yoursite.com/blog/ and Google applies that label to the pattern, every post you have ever paid for stops ranking, and the way back is a reconsideration request reviewed by a person on Google's timetable, not yours. Whatever share of your leads came in through blog content dries up while you wait. For most owners I work with, that is the single cheapest source of booked calls they have.
Age did not save that domain. Neither did the fact that hundreds of thousands of the posts were written by real humans with real Windows problems. The label was applied to the container, and everything inside the container went down together. That is the part owners underestimate when they let a vendor bolt an auto-publishing tool onto a blog that was previously fine.
The label was "no added value," which is not a synonym for AI
Read the wording Google actually used and the panic shrinks into something manageable. Nothing in "thin content with little or no added value" mentions how the text was produced. It describes what a reader gets when they arrive: an answer they could have gotten anywhere, from nobody in particular, about a problem the writer never had.
Montti made the point about forums specifically, and it applies to your service pages just as cleanly.
Received knowledge is the operative phrase. A language model can restate what has already been written about hip replacement recovery times, HVAC sizing, or conveyancing fees. It cannot tell you that the last eleven patients who called your clinic all asked the same question about driving, and here is what you tell them. One of those pages earns its place in search results. The other one is a slightly reworded copy of what already ranks.
The counter-evidence matters just as much, because vendors on both sides are overstating their case. Matt G. Southern reported in Search Engine Journal on the same day, July 27, 2026, that pages scoring high on AI detectors do tend to sit in lower positions and get indexed more thinly, yet heavily AI-flagged pages still show up across Google's top 10. Both things are true at once. AI-drafted text is not an automatic death sentence, and it is not a free pass either.
Rankability's 2026 study analyzed 487 search results pages and reached the same conclusion: quality predicts ranking, production method does not. Google Search Central has said the plain version for a while now. AI-assisted content is not penalized by default. Scaled content abuse, meaning mass production of low-value, near-duplicate pages made mainly to catch search traffic, is a policy violation regardless of whether a human or a machine typed it.
So the question to ask your agency is not "do you use AI." Almost everyone does, including me. The question is what gets added after the draft exists, and who is accountable for the claims inside it.
The 50-posts-a-month offer is the same machine with a smaller invoice
Picture a physiotherapy clinic with four practitioners and a website that has never earned a booking. A tool sells them a content package: 40 FAQ pages, one per condition, published over six weeks, each answering a question like "how long does frozen shoulder take to heal." Nobody at the clinic reads them. The pages go live, the dashboard shows 40 new indexed URLs, and the owner feels like progress happened.
Every one of those pages restates what the NHS site, the local hospital, and dozens of other clinics already say. Not one contains the sentence the clinic could actually own, which is what their own patients report at week six, in their own words, with the treatment protocol this clinic uses. The pages are competent. They are also interchangeable, and interchangeable at volume is precisely the shape Google's scaled content abuse policy describes.
The same offer arrives dressed as a subscription, an "AI SEO platform," a white-label content partner, or a junior hire with a prompt library. The invoice size changes. The mechanism does not.
| Attribute | Safe: AI-assisted, human-directed | Risky: scaled content |
|---|---|---|
| Where the angle comes from | A person who talks to your customers decides what the page argues | A keyword list decides, and the tool fills in the rest |
| First-hand material | Your results, your pricing, your failures, your before-and-after numbers | Received knowledge only, restated from pages that already rank |
| Fact-checking | Every stat traced to a named source and date before publish | Unsourced claims and invented statistics nobody checks |
| Publishing rate | Limited by how fast a human can add something real | Limited only by the tool's quota, which is the pattern Google's policy names |
| Overlap between pages | Each page answers a question the others do not | Near-duplicates for keyword variations of one question |
| Accountability | A named author who can defend the advice on a phone call | No author, or a byline nobody could reach |
Nothing in that right-hand column requires AI. Human freelancers producing rewritten summaries at commodity rates land in exactly the same place, which is why "we only use human writers" reassures nobody who understands the policy. The volume plus the emptiness is the violation. AI just made both cheaper to reach.
Run the added-experience test before anything goes live
Take any page on your site and try to answer one question about it: what does a reader learn here that they could not learn from the three pages already ranking above you? If the honest answer is nothing, the page is a liability whether a person or a model typed it. Five checks make that judgment concrete enough to hand to a junior marketer.
That test is deliberately coarse, because it has to survive being run by a busy person on a Tuesday. When you want the sharper version for a single page, the tenth-sentence method in The 10% Rule, the test that decides if Google shows your page, forces you to read the top three results first and write the one sentence they leave unanswered. Coarse test for the whole library, precise test before you publish.
I hold my own site to that gate. Before any post here gets drafted, I read the top three ranking pages for the target query and write the one sentence they leave unanswered; when no honest sentence exists, the idea gets killed and logged instead of softened and published anyway.
A good partner answers that last question in a minute. A content mill goes quiet, and the silence is your answer.
Volume itself is not the sin. A team can publish weekly for years without risk if each piece carries something real. The way to keep that sustainable is a documented plan rather than a content quota, which is the whole point of building a content strategy playbook that decides what gets published and why before anyone opens a draft.
Watch the reports a manual action would show up in, not your post count
Three numbers tell you whether this is going wrong, and you can check all of them in Search Console in under ten minutes a month. The Manual Actions report is binary and non-negotiable; check it monthly and the day after any big publishing push. If the report is not empty, fix the flagged pattern first, strengthen or remove the empty pages, then file the reconsideration request from the same screen, and expect Google to review it on its own schedule, not yours. The Pages report, indexed versus not indexed, is the early warning: Southern's July 2026 reporting noted that heavily AI-flagged pages tend to get indexed more thinly, so I read a rising count of "crawled, currently not indexed" URLs as Google looking at the new pages and passing. Then the Performance report filtered to your blog path, watching clicks rather than impressions, because impressions rise on volume alone.
| The vanity metric | What to watch instead |
|---|---|
| Posts published per month | Inquiries, calls, and bookings from organic search |
| Total indexed pages | Pages earning clicks in the Performance report |
| Keyword count | Clicks on queries a customer would actually type |
| Average position across all terms | The Manual Actions report staying empty |
Give it 60 to 90 days before you judge a content change, since Google needs to crawl, index, and rank before anything moves. Judging a post at two weeks is how owners kill the pieces that were about to work.
The vanity metrics are easy to name because vendors report them first. Number of posts published is a production statistic, not a business one. Total indexed pages goes up when you add 40 interchangeable FAQ pages and it goes up right before the demotion, so it measures the risk as happily as the progress. Keyword count and average position both improve on branded and near-zero-volume terms without adding a single lead. What you actually want is the count of inquiries, calls, and bookings attributable to organic search, and if your reporting does not show that, the reason it does not is worth more attention than the report itself. I have made this argument at length in the case for grading your content on something better than Google traffic.
One more discipline: keep a record of which pages carry which first-hand claim. When a manual action does land on somebody, the sites that recover fast are the ones that can show a reviewer what changed and why. The ones that cannot are still writing reconsideration requests six weeks later.
Frequently Asked Questions
Does Google penalize AI-generated content in 2026?
Not for being AI. Google Search Central's position is that AI-assisted content is not penalized by default, while scaled content abuse, meaning mass low-value or near-duplicate pages made mainly for search traffic, is a policy violation regardless of who or what produced it. Rankability's 2026 study of 487 search results pages reached the same conclusion: quality predicts ranking, production method does not. Matt G. Southern reported in Search Engine Journal on July 27, 2026 that heavily AI-flagged pages still appear across Google's top 10. What gets punished is a page with nothing added, especially when there are thousands of them.
What actually happened to the forum that had 500,000 posts demoted?
Roger Montti reported in Search Engine Journal on July 27, 2026 that windowsforum.com received a partial manual action around July 17, code WNC-651700, labeled "Thin content with little or no added value." It hit a single URL pattern, /threads/, which contains 168,290 threads and more than 500,000 posts on a roughly 20-year-old domain. SEOs including Gagan Ghotra suspect the trigger was the site's AI chatbot "staff member," which had auto-posted about 111,050 replies since March 2023, rather than the human threads. A partial manual action means a person at Google reviewed that section and demoted it by hand.
An agency offered me 50 AI blog posts a month. Should I take it?
No, not on those terms. Ask them one question in writing first: which sentence on each page could only have been written by your business, and what is the source for every number in it. If the answer is that the pages summarize what already ranks, you are buying the exact pattern Google's scaled content abuse policy targets, and you are buying it 50 times a month. A smaller number of pages that each carry your own results, pricing, or customer patterns will outperform them and will not put your existing pages at risk. Volume is fine when every piece adds something; it is the emptiness at scale that costs you.
The forum that had half a million posts demoted did everything the old playbook rewarded. It aged, it accumulated, it answered questions, and it grew its page count every single month. None of that saved it, because Google stopped counting pages and started asking what was inside them. If your marketing report still leads with how many posts went out last quarter, you are measuring the exact thing that got that site demoted. When you want a straight read on whether your own library adds anything, book an AI visibility call and we will run the swap test on your last ten pages together.
