- 762 AI-generated articles published in 60 days. 31 clicks in 12 months.
- Position 27.89 for my own surname. The n8n stack ran perfectly. The strategy was structurally broken.
- If a CMO is being pitched “AI content at scale” for a regulated financial services brand, the same playbook is in the deck, and the math will be identical.
Thirty-one clicks. That is what 762 articles published over 60 days produced for this site over the last twelve months. The n8n flow ran every four minutes. The Gemini API bill came in under €50. Google Search Console’s verdict on my own surname: position 27.89.
This piece is the public confession of a content strategy that was technically perfect and strategically broken. If you are a CMO at a regulated financial services firm being pitched “AI content at scale” right now, read carefully, because the same playbook is being sold to you and the math will be identical.
The setup: what I built and why
The hypothesis was straightforward. European fintech and asset management is a fragmented, multilingual market. The semantic SEO model, Laurent Bourrelly’s cocon sémantique, works by establishing topical authority through a dense network of related content. AI now produces that content at near-zero marginal cost. So the plan was to scale the content, scale the authority, and capture the traffic.
I wired it together in n8n. Two workflows did the work.
The Post Creator ran on a schedule trigger every four minutes. It read a target keyword from a Google Sheet (for example, Fractional CMO Madrid), called Gemini 3.1 Flash Lite, the cheapest available model, generated a 1,000-word article, formatted it as HTML, and published it directly to WordPress.
{ "rule": { "interval": [ { "field": "minutes", "minutesInterval": 4 } ] } }
Fifteen possible runs per hour. Three hundred and sixty per day if it never sleeps.
The Internal Linker ran every ten minutes. It picked one published post and injected exactly two links, one to a hub page and one to a commercial page. Hardcoded. The same two destinations across every post in the corpus.
The keyword universe was a 226-row sheet of {Role} {City} combinations, the pSEO playbook applied to every European capital between Lisbon and Tallinn. By 28 April 2026 the WordPress sitemap counted 762 URLs. Of those, 414 had been published in March alone and 304 in the first four weeks of April. Eighteen posts a day, on average, with no human in the loop.
The data: what happened
Published Feb-Apr 2026 on roodhorst.eu. Google Search Console, last 12 months.
Google Search Console, last 12 months. 748 of 762 pages never produced a single click.
That is 0.04 clicks per page per year. Of 762 URLs, only 60 (8%) ever surfaced in Google's index even at low positions. Of those 60, only 14 received a single click in the entire year. The remaining 748 URLs sat invisible, burning Google's crawl budget, signalling thin content, and dragging the domain's average authority down.
Five structural failures, ranked by damage
Google's helpful content classifier looks at velocity as a signal. A site publishing twelve articles a day, every day, with no editorial gaps, no human review markers, and no images on 99% of pages is indistinguishable from any content farm the classifier has been trained to demote.
Gemini 3.1 Flash Lite Preview was selected for cost. That left nothing for Google to surface as worth ranking.
Seven hundred posts each linking to the same two commercial pages is not a cocon sémantique. It is a 1,400-edge link wheel pointing at two destinations, the exact pattern Google has detected and discounted since 2012.
"Fractional CMO Madrid" is searched by founders of Series A SaaS startups looking to hire. They are not the buyers of my offer, which is multilingual GTM launches for €50M+ regulated financial services firms.
Of 762 pages, 755 had zero images. None had an author byline block, Person schema, or a primary source. Google's E-E-A-T framework looks for exactly the signals that were missing.
If I had to do this again, here is what I would build
The rebuild is public, and you are reading part of it. The distribution operating model and the pieces around it are what "do it properly" looks like in practice.
- One pillar, not seven hundred. Three content pillars maximum: a narrative pillar, a tactical pillar, and a public test lab. Thirty hero pieces total. Maximum velocity of two pieces per week.
- Research before drafting. Every piece begins with eight verified sources. A reasoning model handles research; a strong writing model handles the draft.
- Drafts only, never auto-publish. Every piece passes a human review gate.
- Author byline and Person schema on every piece. A verifiable Person entity, the thing this experiment never had.
What this means for any CMO running an AI content motion now
The AI-content-at-scale pitch you are getting from agencies in 2026 is the exact playbook that produced the 31-click result above. If "scale your content output 10x with AI" is in the proposal, ask three questions.
One. What is the publishing cadence? Velocity is one of the strongest demotion signals Google currently uses.
Two. Show me the link architecture. If 100+ posts link to the same two or three commercial pages, the agency is building a link wheel.
Three. What is the E-E-A-T plan per piece? Volume cannot substitute for the trust signals Google's algorithm is trained on.
Follow the rebuild, week by week. Notes from a European Operator is one short email every Friday: one observation from the field, one tactical thing for FinServ marketers, one new piece in the Test Lab, one open question. Subscribe to the Friday log.
Frequently asked
Was the AI bad?
No, the strategy was. Gemini 3.1 Flash Lite produces clean, grammatically coherent prose. It was the wrong tool for the wrong strategy.
Is all AI-generated content failing?
No. AI-assisted content with real human curation, original research, and controlled cadence works well.
Should I fire my content agency?
Read your contract. If they are running this exact programmatic-SEO and auto-publish playbook on your domain, the question is how quickly you can apply the kill list, not whether.
What about Google's recent helpful content updates?
Every quarterly update since 2023 has tightened the demotion signal for AI-generated thin content at scale. Build for the algorithm of 2027, not 2021.
Can I see the raw data?
Yes. The numbers come from the Google Search Console export. If you want a second pair of eyes on your own content motion before it becomes a 31-click story, the full method and dataset are on this page.
who is building the AI bubble is the argument in full.
the AI Asset-Manager Observatory logs the answers behind the experiment.
how the AI financial advisor prompt gives itself away is the follow-up test.
1 thought on “I Generated 762 AI Articles for My Consulting Site. They Produced 31 Clicks.”