AI-written content ranks on Google when the draft carries information the model did not already have. Every post on this blog is drafted with AI, published in 7 languages, and the site's organic clicks grew 63% in a single week this May.
The page you are reading right now was written by an AI that works under instructions to delete most of what it produces.
If you run a business blog, you have probably tested AI drafts and worried about a Google penalty. This article shows real Search Console data from 105 AI-drafted posts, the ones that rank and the ones that stall. You get the March 2026 core update context, the editorial gates this blog runs, and a decision table for your own content.
- Google's March 2026 core update targets scaled content abuse. pSEO sites lost 60 to 90 percent of their rankings, while pages carrying first-party data kept ranking, whoever or whatever wrote them.
- The top post pulls 205 clicks per week at position 6.0. Sister posts from the same pipeline stall at a few hundred impressions per week. The moving variable is the anchor, the piece of information no model already has.
- ChatGPT and Perplexity entered this site's top-7 referrers in May 2026, citing the posts a vanilla model could never have generated.
- Every draft passes a gate: an anchor requirement before writing, a mechanical scrub after, and a kill switch that already reverted one published post within hours.
- A decision table below shows when AI drafting grows a blog and when it burns the domain.
This Blog Is Written With AI. Here Is What It Earns.
webvise.io publishes in 7 locales: English, German, French, Spanish, Dutch, Polish, and Italian. The blog holds 105 posts, every one drafted by an AI pipeline and gated by a human editor before publication. All numbers below come from Google Search Console and Vercel Web Analytics for this domain.
The top-clicking post pulls 205 clicks per week at position 6.0, and it is the German locale of an English original. A French post converts at a 5.65% click-through rate. In the week ending May 3, ChatGPT and Perplexity entered the top-7 referrers for the first time, and clicks rose 63% week-over-week on flat impressions.
The pattern repeated with news content this month. A post on Anthropic's Fable 5 redeployment went live July 1, took 77 clicks in its first week at position 4.7 for its target query, and lifted site clicks 32.8% week-over-week on 74,032 impressions.
That entire production line, drafting, translation, and the gates in between, is an automation project. webvise builds the same kind of governed pipeline for client teams as an AI automation engagement, with your first-party material as the input.
What Google Enforced in the March 2026 Core Update
Google's March 2026 core update named scaled content abuse as an explicit violation. Sites publishing thousands of templated AI pages lost 60 to 90 percent of their rankings. The update never mentions authorship, and human-written generic content took the same hit.
The mechanism sits in AI Overviews. For every indexed page, Google effectively asks whether the model can already summarize it from existing knowledge. A summarizable page gets answered inline, and the click dies there. A page carrying unknown information ranks normally, because Google has nothing to show above it.
| Content type | What Google does with it | Traffic outcome |
|---|---|---|
| AI article written from the title alone | AI Overview answers it inline | Impressions without clicks |
| Thousands of templated pSEO pages | Classified as scaled content abuse | 60 to 90 percent ranking losses |
| Human-written generic listicle | Same inline summary treatment | Impressions without clicks |
| AI-drafted page with a measurement no model has seen | Ranks normally, no summary available | Clicks route to the page |
The fourth row is the entire strategy. The classifier reads what the page adds to the model's knowledge, and the byline never enters the calculation.
Same AI, Same Template: The Posts That Rank and the Posts That Stall
This blog is its own controlled experiment. Every post shares the pipeline, the schema markup, the internal link structure, and the translation process. Anchor strength is the only variable left.
In the week ending May 3, a 1,200-word decision-tree post pulled 15 clicks from 1,784 impressions at position 4.98, a 0.84% CTR. The same week, a long-form explainer on the same site pulled 4,201 impressions and exactly 1 click. Both were AI-drafted. The decision tree carried a framework with defended thresholds, and the explainer answered a question the model could answer itself.
Posts with weak anchors stall at a few hundred impressions per week and never break out. The stall is an input problem: no model generates your company's numbers, your project outcomes, or the position you are willing to defend in a sales call.
The Pipeline: What the AI Drafts and What the Gates Kill
Every article starts with an anchor requirement. A brief must contain at least one measurement, post-cutoff event, defensible position, or anonymized project example before drafting begins. Bare topics get rejected outright, and the AI assembles what the intake established.
- Anchor gate before drafting. A bare topic like "AI for e-commerce" gets aborted. The brief must name the data or position the article defends.
- Mechanical scrub after drafting. A script counts banned phrases, dash characters, paragraph lengths, and known AI structure tells. Any nonzero count blocks publication.
- Quality score with a floor. Drafts below the threshold get rewritten before translation, never after.
- Translation audit per locale. All 7 language versions are checked mechanically so every number, name, and link survives. A Polish reader gets the same claims as an English one.
The kill switch has fired on live content. On June 29, a cost guide went live, read like a generic price table on review, and was reverted within hours. The rewrite shipped the same day, rebuilt around the pricing principles webvise actually quotes in proposals, and that version is the one ranking today.
Should Your Business Blog Use AI? The Decision Table
| Your situation | Use AI drafting? | Condition |
|---|---|---|
| You hold internal data, client outcomes, or a position you will defend | Yes | Gate every draft on one of those anchors before writing |
| You want hundreds of pages targeting keyword permutations | No | The March 2026 update classifies this as scaled content abuse |
| You publish in more than one language | Yes | Translation is the strongest AI use case. Audit numbers and links per locale |
| Nobody can supply first-party material yet | Fix intake first | Anchor capture from project work comes before any drafting |
The economics follow the table. AI drafting cuts the copywriting cost to near zero, so the budget shifts to subject-matter capture: logging benchmarks, interviewing the founder, writing down what a project actually cost. That capture step is where most in-house attempts stall, and it is the part webvise runs as intake at the start of a content engagement.
What This Means for Your Content Budget
Plan for fewer pages and stronger briefs. A short post with one real measurement outperforms a 3,000-word guide assembled from training data. On this domain the gap was 15 clicks to 1 in the same week.
If your current agency sells volume, monthly post quotas, or word counts, the March 2026 update priced that model out. If your team can name one number nobody else has published, you have enough for a post that ranks.
webvise runs this pipeline for its own blog and builds it for client teams: intake, anchor capture, AI drafting, 7-locale translation, and the gates in between. Book an intro call to walk your existing blog through the same audit.
Development practices are aligned with ISO 27001 and ISO 42001 standards.