In May 2024, Liz Reid stood on the I/O stage and announced that AI Overviews would be live in US Search by the end of the week. Within a year the feature had rolled out to more than a hundred countries, absorbed half the visible real estate above the fold, and started displacing the link list that twenty years of SEO had been built around. The effect on traffic was immediate. Studies from Search Engine Land and Semrush showed double-digit click-through declines on informational queries even when the brand still ranked in the blue links beneath.
The successor discipline has acquired a name nobody loves: Generative Engine Optimisation, or GEO. The work is two years old. Most brands are about a quarter into trying it.
Why GEO is not SEO with new keywords
The two disciplines target different artefacts. SEO targets a ranking position on a results page. GEO targets a citation in a generated answer. The mechanics rhyme but they do not match.
- SEO rewards a page
- GEO rewards a sentence inside a page
- SEO is scored against keyword intent
- GEO is scored against query embedding
- SEO content is one-shot
- GEO content is sliced, recomposed and partially quoted
- SEO authority signals matter for ranking
- GEO authority signals matter for being trusted enough to cite
Princeton's GEO benchmark and the academic GEO paper from 2023 have already established the optimisation moves that lift citation rate. They are not the moves SEO teams are trained on. Lead claim hardening, source attribution inside paragraphs, named-entity consistency, structured summary blocks. None of these rhyme with keyword density.
What we changed in our content brief
We rewrote the content production brief in February 2025 and again in November. The current version has six discipline checks layered on top of the old SEO recipe.
- State the answer in the first sentence so chunked content survives
- Cite the data inside the paragraph it supports, not in a list at the bottom
- Use the brand's named entities consistently throughout the piece
- Include a structured summary block that models can lift cleanly
- Avoid promotional fluff that lowers the trust signal to the model
- Update the canonical content rather than write a fresh post for every variation
The lift in citation rate has been worth more than the equivalent SEO work it replaced. Importantly, the same content also performs better in the residual link list, because the same writing discipline that helps a model attribute also helps a human read.
“The job is no longer ranking. The job is being trusted enough to be quoted.”
Distribution still matters
A common misconception is that GEO removes the need for distribution. It does not. Models are trained on what is widely cited and they retrieve from what is widely indexed. Outlets, citations and authority signals still drive what gets surfaced. The difference is what those signals translate into. Five years ago they translated into a higher rank. Today they translate into a higher probability of being the model's chosen source.
The thing buyers should be asking their agency
Before signing a content retainer in 2026, ask one question. Show me the citation rate of your last twelve months of work in Perplexity, ChatGPT search, and Google AI Overviews. If the agency cannot produce the number, they are still optimising for last decade's surface.