What If the SEO Playbook Was Wrong?

For ten years, the advice on how to be found in search was unanimous. Write more. Be comprehensive. Create

By Conor Healy

June 1, 2026

For ten years, the advice on how to be found in search was unanimous. Write more. Be comprehensive. Create the definitive resource. If someone could find the answer somewhere else, your content was not good enough.

Marketing teams published 8,000-word guides. They added sections until there was nothing left to add. Depth meant authority, authority meant rankings, rankings meant traffic.

Then HubSpot lost 70 to 80% of its organic traffic. Healthline lost approximately half. These were not sites with thin content. They were sites that had followed the rules better than anyone.

Information Burial

The Scientific Institute for Generative Intelligence ran controlled experiments on how content length affects AI citation probability. The finding inverts a decade of advice.

Citation probability rises with depth up to approximately 5,000 words. Then it drops. Content beyond that threshold performs worse than content half as long.

The researchers call it information burial. When a page contains too much material, AI systems struggle to extract the relevant claims. A 2,500-word article with clear structure and specific, verifiable statements outperforms a 10,000-word ultimate guide where the same information is scattered across dozens of sections.

The pillar pages designed to rank on Google are being systematically ignored by ChatGPT and Perplexity. The most comprehensive resource on the internet is now the least citable.

Why Optimisation Language Hurts

The second SIGI finding is stranger.

Researchers analysed websites across multiple service categories and found a consistent correlation: sites that use GEO-related vocabulary are less likely to receive AI citations than sites that do not.

Pages mentioning “AI citations,” “LLM optimisation,” “generative engine,” or similar terms correlate with lower citation rates. The meta-language about being found by AI correlates with not being found.

Two mechanisms explain the pattern. First, content about optimisation tends to be thin on the substantive claims AI systems want to cite. A page explaining how to get cited by AI contains less citable information than a page that simply answers a question well. Second, AI systems may have learned to associate optimisation language with promotional content, triggering a credibility discount.

The vocabulary designed to attract AI attention is the vocabulary that repels it.

Same Problem, Different Symptoms

Both findings describe counterproductive effort. The comprehensive guide buries its claims under volume. The GEO-focused page signals its intent so clearly that the signal becomes noise. In both cases, the marketer tries harder and gets less.

AI systems are not rewarding effort. They are rewarding extractability. A page that answers a specific question with a specific, verifiable claim in a self-contained paragraph is more citable than a page demonstrating how much work went into it.

What This Requires

For marketing teams that have spent years building content engines optimised for volume and depth, this is not a tactical adjustment.

Shorter pages. Less visible optimisation. Resisting the instinct to add more sections, more keywords, more signals. The opposite of what worked before.

The research is still emerging. SIGI notes that findings come from controlled experiments that may not generalise across all contexts. But the pattern is consistent: the sites that followed the old rules most faithfully are the sites losing the most ground.

The question is no longer how to appear in AI search. The question is what to stop doing.