Article
Search google zero

Google Zero and the feedback loop: What this means for publishers

The thing every publisher is dreading: a version of search that doesn't send you any readers at all.

by Ian Lyall
The image features four stylized robots sitting at desks with laptops. Each robot has an illuminated 'AI' symbol on their chest and is engaging with the laptops in front of them, suggesting a focus on artificial intelligence and technology. — Credit: Photo by Mohamed Nohassi on Unsplash c Photo by Mohamed Nohassi on Unsplash

So, the next generation of search, as Google has framed it thus far, leans towards giving people an answer on the spot rather than a list of links to go and read one. For media companies, the floor disappearing.

This is Google Zero. It describes the scenario where referral traffic simply tanks, it hits zero (nada, zilch). For sites built on search discovery, rather than via a direct readership, that's terminal.

So, who feeds the machine?

Here's the rub, search needs publishers, just as much as publishers need search.

Google's answers exist because someone, somewhere, reported the story, ran the investigation, or wrote the review. If the direct-answer model kills the economics that pay for that work, then the well runs dry. Bottom line: no new reporting, no new material for the next answer.

An inbreeding problem

Does this then create a recursive loop? This is a scenario where large language models generate text, that text ends up back in the training data for the next generation of models, and the cycle repeats. Slop squared.

One commentator referred to this as 'inbreeding', which, I think, is self-evident. Each pass, the output drifts further from anything a person actually observed or verified, computers increasingly learning from what other computers made up.

What next?

For anyone still building a content strategy around search visibility, that's the uncomfortable bit. The traffic problem might be survivable with a different distribution plan. The provenance problem, models trained on models trained on models, is harder to route around.

by Ian Lyall
whatsnew. APPS · WEB TOOLS · SECURITY · AI

Know what’s new.

The useful side of the internet. Covered properly.

Set as preferred →