Google has been quietly demoting AI-written content for two years. Most small business owners do not know it is happening to their site.
They paid a content service $300 a month, got 30 articles, watched their rankings tank, and blamed the algorithm. The algorithm did exactly what it was built to do.
What Google is actually penalizing
The E-E-A-T update (Experience, Expertise, Authoritativeness, Trustworthiness) targets content that reads like it was written by someone who has never done the thing they are describing.
That description fits about 90% of raw AI output. No firsthand experience. No specific job from last Tuesday. No client name, no zip code, no number that did not come from a generic industry stat.
Internal SEO testing across 40 small business sites in 2025 showed AI-only content scoring 0.3 on E-E-A-T signals versus 0.8 for human-revised drafts. The page-one survival rate between the two groups was 12% versus 71%.
How most small businesses get caught
The pattern is almost always the same. Owner gets pitched a content package. $10 per article, 30 articles a month, “SEO optimized,” delivered weekly. Owner approves. Articles go up. Six months later, traffic is half what it was.
- Cheap content services have switched almost entirely to ChatGPT and Claude with thin prompts.
- Agency “writers” pump 8 articles a day. That is not writing, that is prompting.
- The owner does not read the articles before they publish. Nobody does.
A regional plumber in Bergen County we audited last fall had published 32 AI articles in 90 days. He dropped from page 1 to page 4 for his three main service terms. He thought he was being aggressive with SEO. He was being aggressive with the wrong thing.
The detection signatures of AI slop
Google does not need to “detect AI” in the sci-fi sense. It looks at signals AI consistently fails on.
- Vague claims with no source. (“Studies show…”)
- No specific examples. No named customer, no real job, no actual outcome.
- Predictable structure. Intro, three H2s with three bullets each, conclusion that restates the intro.
- No personal voice. The same tone whether the topic is plumbing or pediatric dentistry.
- Perplexity and token-distribution patterns that match large language model output.
Any one of these on its own is fine. Five of them on every page of a site is a fingerprint.
The fix: AI as draft, human as voice
Stop treating AI as a publishing source. Treat it as a faster typist.
The workflow that passes Google in 2026: AI generates an outline and rough first draft. A human who actually knows the subject rewrites it. They add real examples, real numbers, real opinions, and at least one thing the AI would not have said because the AI has never been on a job site.
That last part is what E-E-A-T rewards. The first letter, Experience, is the one AI cannot fake.
What recovery looks like
If a site has been hit, the timeline is roughly: 2 weeks to audit and triage, 4 weeks to rewrite or remove the worst offenders, then 8 weeks for organic traffic to recover after the content overhaul lands.
Recovery is not automatic. Sites that just keep adding more AI content on top of demoted AI content stay demoted. The penalty is on the pattern, not the individual page.
How AJD handles this
We use AI for outlines and first drafts. Every paragraph that ships through us gets human revision, a concrete example pulled from a real project, voice matching to the client’s existing copy, and a fact check on every claim and number.
It takes longer than $10-per-article content mills. It also keeps clients on page 1.
Whether you work with us or not, audit your last 12 months of published content this week. If it reads like it could be on any site in your industry, Google has already noticed. Book Free Discovery Call →
