Insights · GEO

GEO: how brands get cited by ChatGPT and Perplexity

A growing share of buyers now ask an AI, not a search box. When the answer is one paragraph naming three brands, the game is no longer ranking - it is being the brand the model names. That is Generative Engine Optimization.

For twenty-five years, being found meant ranking - climbing a list of links a person then chose from. That surface is shrinking. A rising share of queries now return a single generated answer: a paragraph that names a few brands and cites a handful of sources, with no list to climb. In that world, position 8 does not exist. You are either in the answer or you are not, and Generative Engine Optimization - GEO - is the discipline of being the brand the model names.

What actually changed

The mechanics are different from search. A model doesn't rank your page against nine others; it reads across many sources, synthesises, and reproduces what it can state with confidence. So the winning content is not the most "optimised" - it is the most quotable: clear claims, defined terms, structured facts, numbers a model can lift without ambiguity. Pages written to hedge and impress read as noise to a system looking for something it can assert plainly. This is closer to being a good reference than to gaming a ranking.

Own a claim the model can repeat

Models gravitate to brands that occupy a distinct, well-defined position, because a clear position is easy to state and hard to confuse. This is the old Ries and Trout lesson - own a word in the mind - re-expressed for a machine: own a claim the model can attach to you without hedging. "A leading provider of innovative solutions" gives a model nothing to repeat. "The category's low-alcohol specialist" or "the fastest route-to-market audit in FMCG" gives it a sentence. Distinctiveness was always a growth lever; now it is also a retrieval lever.

In classic search you competed to be ranked. In AI answers you compete to be repeated. Vague brands don't get repeated.

Presence beyond your own site

A model's picture of you is assembled from the whole web, not just your homepage. Third-party articles, reviews, directories, comparisons and the language other people use about you all feed the answer - often more heavily than your own marketing copy, which the model discounts as self-interested. Mental availability, in Ehrenberg-Bass terms, now has a machine reader: the more places state clearly what you are and what you're for, the more reliably the model reconstructs it. Earned presence is GEO infrastructure, not PR decoration.

Let the crawlers in - then measure

None of this works if the AI crawlers can't read you. The most common self-inflicted GEO wound is a robots.txt or edge setting that blocks GPTBot, ClaudeBot, PerplexityBot or Google-Extended - quietly declaring your site off-limits to the exact systems you want to be cited by. Allow them deliberately. Then measure the thing that matters: not rankings, but whether you are actually named in answers to the questions your buyers ask. Share of AI answer is the new share of voice, and it is trackable.

How we build it

We treat GEO as part of Digit It, our digital work - the claims, the structured content and the earned presence that make a brand legible to a model - and we use Brand X-Ray to baseline where you're cited today versus the competitors the models currently prefer. The brands that win the AI answer will be the ones that were clear about what they are, in more places, earlier. That work compounds; starting late is the expensive option.