The short answer is yes, generative engine optimisation works, but not in the way it is usually sold, and not for every business. GEO is the practice of improving how often, and how favourably, AI answer engines like ChatGPT, Gemini, Perplexity and Google's AI Overviews name your brand. The evidence that you can move those outcomes is real. The evidence that it is easy, fast or stable is not. This is an honest read of what works, what does not, and how to judge it for your own case. If the term itself is new, start with what generative engine optimisation is.
What the evidence actually supports
There is now enough data to say the levers are real. Ahrefs' correlation work found AI visibility tracks with third-party mentions and video presence more than with anything on your own site, per Ahrefs. Community citation share is large and measurable, with Reddit alone accounting for 40.1 per cent of citations in one Semrush study. And naming a brand consistently changes what a model says about it. None of that guarantees traffic, but it confirms the mechanism: you can influence whether a model knows and names you.
Where GEO falls down
Be clear-eyed about the failure modes. The first is volatility. Citation shares move week to week, sometimes violently, so a win can evaporate on a model update; Reddit's ChatGPT share moved from roughly 60 per cent to about 10 per cent in a fortnight in late 2025, per Semrush. The second is attribution. A mention inside an AI answer often carries no link and no clear path into your analytics, and 62 per cent of citations never even name the brand, per the ghost citations study, which makes return genuinely hard to measure. The third is fit. If nobody asks AI for recommendations in your category, optimising for it is effort spent on absent demand.
What moves the needle, and what does not
The levers that work are consistent across the studies: third-party mentions and comparison inclusion, genuine community presence in places like Reddit and the forums your buyers use, video, consistent brand-and-category naming, and content structured so a model can quote it. The things that do not work are just as clear: keyword-stuffing your own pages, spinning up thin pages at scale, and the myth that putting community keywords into your slugs earns citations. On-page tricks are the weakest lever there is. How to get cited by AI and the best GEO and AEO tools cover the practical side.
Want to see this in action?
See how every major AI model talks about your brand. Free to start.
Who it works best for
GEO pays off soonest in categories where buyers actively ask AI for recommendations: software, professional services, considered purchases and B2B tools. It pays off slowly, or barely, for businesses whose customers do not use AI to choose at all. An honest check of whether your buyers ask AI in the first place should come before any GEO budget, not after.
How to tell if it is working for you
Do not judge GEO on a single screenshot. Measure a fixed set of buyer questions across all four models on a schedule, track both citations and named mentions, and read the trend over weeks. Compare against where you started, and against the competitors named alongside you. The point is a baseline and a direction of travel, not a vanity reading. What AI visibility is explains the metrics, and the free check gives you a starting baseline.
The verdict
GEO works as an influence exercise, not a growth hack. Treat it like PR and reputation rather than classic SEO with a guaranteed ranking. The brands that benefit build genuine third-party and community presence, describe themselves consistently, and measure honestly over time. The ones that treat it as an on-page checklist, or buy into slug tricks and thin-page schemes, waste the budget and sometimes invite penalties. Done properly, it is worth it. Sold as a quick fix, it is not.
