An AI-generated news publisher returned to Google Discover after an earlier blacklist, expanded across three sites and was blocked again following a sharp rise in exposure. Next INpact reports that the sites captured up to 4% of traffic in early October while their operator published 26,000 articles.
The episode establishes more than the presence of low-cost automated content. It shows that a previously sanctioned operator regained meaningful distribution through Google’s recommendation system before enforcement reduced its visibility. That sequence makes the durability of platform-supplied traffic the central issue for content teams.
The return turned publishing volume into reach
The publisher’s production scale supplied Discover with a large inventory of pages to select. Next INpact counted 26,000 articles from the operator, compared with 34,000 from Ouest-France, a publisher employing more than 700 journalists. The comparison concerns output, not equivalent reporting resources, accuracy or editorial depth.
Discover then provided the distribution mechanism. Rather than waiting for readers to enter a search query or visit a known publication, the recommendation system placed selected articles in front of users. High-volume production therefore created many opportunities for individual pages to receive impressions.
The available evidence does not reveal which subjects, formats or other signals caused those pages to be selected. It establishes the sequence: the operator returned, its sites obtained substantial exposure, and the measured presence of AI-generated material dropped after Google intervened.
Conceptual illustration of publishing volume feeding a recommendation system, not a photograph of the service.
Discover exposure measured distribution, not truth
The monitoring cited by Next INpact found that AI-generated sites initially received fewer impressions per unique article than human-produced sites. That relationship reversed in June, after which the AI-generated articles reportedly performed better through the summer and into October.
This is evidence about recommendation performance. It is not evidence that the articles were accurate, original or useful. An impression records distribution, while editorial quality requires different evidence: identifiable sourcing, support for factual claims and a record of how errors are handled.
Next INpact characterises the operator’s earlier activity as involving thousands of articles and fake news. The supplied evidence does not present a representative sample of those stories, measure their error rate or compare their accuracy with that of established publications. The quality claim should therefore remain attributed to the report rather than expanded into a general conclusion about all AI-assisted publishing.
Conceptual illustration showing that recommendation exposure is separate from editorial verification.
The blacklist worked quickly but leaves key gaps
The clearest observed consequence was a sudden change in the composition of recommendations. The share of articles from sites identified as AI-generated approached 12% in early October, then fell below 5% after Google blacklisted a spammer. The dashboard behind these figures does not claim to cover every Discover recommendation.
That decline supports a narrow conclusion: removing one high-output operator can materially alter the monitored feed. It does not explain how Google identified the network, whether enforcement applied to domains or to the operator, or which ranking and policy signals were decisive.
Durability is also unresolved. The same report describes the publisher as returning after a previous blacklist, but it does not document the route used to regain distribution. The current action therefore demonstrates immediate suppression, not permanent exclusion.
Content teams should watch durability, not the spike
For SME content and commercial operations teams, the relevant decision concerns channel dependence. A sudden increase in Discover impressions can reflect recommendation access without demonstrating durable audience demand, brand recognition or editorial authority. Production volume does not remove that dependency.
This does not show that AI assistance automatically triggers exclusion. It shows that a high-output publisher associated with spam could obtain substantial recommendation reach and then lose it abruptly when Google acted. Treating that temporary reach as stable acquisition capacity would confuse platform distribution with an owned audience.
The next useful evidence is specific: whether the blocked network returns again, whether its share remains suppressed, and whether Google discloses the signals behind the action. Until then, teams should judge Discover-led growth by its durability and source quality, not by the height of a short-lived traffic surge.
