If AI is writing all of your content, you are producing the one thing Google and the AI answer engines have both started filtering out: pages that add nothing. In September 2026, unsealed court filings showed Microsoft's own scientists warning that AI answers had started a "doom loop" for the web, with OpenAI's own experts estimating search referrals to affected publishers down as much as 60%. A week later, Google ran its fourth spam update of the year against content produced at scale with no value, whether a person or a tool made it.
Using AI is not the problem. Using it instead of having something to say is. At R17 Ventures AG, a digital performance agency, we use AI to shape original material, never to replace it.
The content playbook used to be one word: more. Then the companies that built the AI answer engines said, in their own documents, that this was hollowing out the web they depend on. And Google started filtering the output. One distinction now decides whether AI helps your content or hurts it.
Yes, on one condition: the substance has to be yours. Neither Google nor the AI answer engines penalize AI-assisted writing. Google's spam policy is explicit that scaled content abuse applies however the pages were produced. What both filter out is content with nothing new in it, and a model prompted with a keyword produces exactly that, because it can only restate what it has already read.
Our working rule: AI works on original material. It does not generate it. Every two weeks we record an interview with the people running the accounts. AI helps turn that into blogs, short-form video and carousels. It does not decide what we think. Outline, draft, tighten, repurpose: all fine. Be the source: no.
On September 17, 2026, a 92-page summary judgment brief was unsealed. Three details matter for anyone producing content:
Whatever the court decides on copyright, the commercial message is out. Clicks from an AI answer are a fraction of clicks from a search result, and the companies building those products have known since January 2024.
Google's September 2026 spam update started on September 24, rolled out over roughly two weeks, and was the fourth of the year after March, June and August. No new policies came with it. It enforces the existing ones, and the one that matters here is scaled content abuse: publishing large volumes of pages that add little or no value, however they were produced.
A thousand thin pages written by freelancers fail the same test as a thousand written by a model. "Adds nothing" is the operative phrase, and Google now runs four updates a year to act on it.
The AI platforms took the traffic. The search platform is filtering out the content made to win it back. Volume, the one thing that is cheap to produce, is the thing both sides have stopped rewarding.
|
Who is filtering |
What they discount |
What gets through |
|---|---|---|
|
AI answer engines (ChatGPT, Copilot, Gemini, Perplexity, Claude) |
Pages that restate what the model already knows |
A source with a new fact, figure, method or position, from a named person |
|
Google Search |
Scaled, low-value pages, AI-made or not |
A page that answers the query better than what already ranks |
|
Buyers |
Content that reads like every competitor's |
A point of view they could not get anywhere else |
Original means original substance, not original phrasing. If a competitor's logo could sit on the piece and nothing would change, it is not working for you.
Original content contains something that did not exist before it was published. The test: could this have been written by someone who never did the work? If yes, it is not original, however well it is phrased.
Qualifies: first-party data; a position someone could disagree with; the reasoning behind a decision and what happened next; a named author who knows the subject.
Does not qualify: summaries of the top-ranking pages, by hand or by model; "ultimate guides" assembled from other guides; anything published to hit a cadence rather than to say something.
Pages published measures effort, and both Google and the AI platforms have just said effort is not what they reward. We track three measures instead:
|
Measure |
How to track it |
What it tells you |
|---|---|---|
|
AI citations |
Run a fixed set of 20 to 30 buyer prompts through ChatGPT, Gemini, Perplexity and Claude each month. Count how often your brand or page is named. |
Whether your content says something the models need |
|
Rankings that survive updates |
Track your commercial pages across every core and spam update. Note which held or improved. |
Whether the page answers the query better than the competition |
|
Replies and conversations |
Comments, shares with a comment, inbound messages, sales calls that reference a piece. |
Whether a buyer found it worth responding to |
No. Google's spam policies penalize scaled content abuse, meaning large volumes of pages with little or no value for users, regardless of whether a person or a tool made them. AI-assisted content that adds something new is treated like any other content.
Yes, as a tool. Use it to outline, draft, tighten and repurpose material that came from your own data, experience or position. Content where the model is the only source restates what already exists and is unlikely to rank, be cited or be read.
An internal Microsoft presentation, unsealed on September 17, 2026, in the New York Times case against Microsoft and OpenAI, used "doom loop" to describe AI answers reducing traffic to the sites that produce knowledge, which in turn weakens the models trained on those sites and the wider web.
GEO is the practice of making a brand's content and entity signals likely to be retrieved and cited in AI-generated answers from ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. It depends on original, attributable content, consistent entity information across the web, and technical access for AI crawlers.
Original content is slower to make and harder to fake. It is also the only kind Google and the AI answer engines still reward. R17 is a digital performance agency, paid on results. Our SEO & GEO work is built so your brand shows up in search results and in the AI answers behind them.
Want to know what we would change about your content plan?
We will tell you straight. Let's talk.