GEO
AEO measures what the engines say about you. GEO makes your content what they actually draw on.
Generative engine optimisation - GEO - is young, and it's already full of snake oil. My version of it isn't mystical, it's entity and salience analysis using Google's Natural Language API, the same class of analysis Google itself uses to work out what a page is actually about. I read the top-ranking competitor pages, extract the entities and their salience, and classify your own coverage as strong, partial or missing against them.
On a real engagement, for a private healthcare practice I'll keep abstract here, the finding was stark: its pages carried a fraction of the entity coverage of competitor pages, and its most important page didn't register the very entity it was supposed to be about. Staff names were absorbing the salience that should have belonged to the clinical terms. Once you can see that, the fix is obvious. Before, it wasn't visible at all.
Priority-scored gaps, not a wishlist
Every gap gets scored by competitor frequency times salience, so the highest-value gaps surface first. That's the difference between a list of everything competitors mention and a prioritised list of what's actually worth fixing. An entity every competitor treats as central to the topic, and that you're missing entirely, is a different priority to one mentioned in passing on one page. The scoring is what tells them apart.
A cannibalisation check, always first
Before recommending anything new, I check whether an existing page already covers the entity. If it does, the answer isn't more content, it's an internal link. New content that competes with your own existing page is a net loss, not a gain - it splits authority and salience between two pages instead of concentrating it in one, which is the opposite of what you're trying to achieve.
Surgical additions, not rewrites
Fixes are two to three additions per page, not a rewrite of the whole thing. The goal is closing a specific entity gap, not replacing content that's already doing its job elsewhere on the page. A page missing one clinical term doesn't need a new page. It needs that term added in the right place, with the right weight, and nothing else touched.
The structural layer underneath
Entity coverage only works if the technical layer supports it - clean structured data, answer-first sections an AI system can actually extract, and consistent entity signals across the site: sameAs links, organisation schema, the connective tissue that tells a machine this is one consistent entity, not a scatter of disconnected pages. Get the content right and the structure wrong, and an AI system still can't extract it cleanly. Both layers have to hold at once.
What's durable and what isn't
A lot of what's being sold as GEO right now won't survive contact with how these systems actually work. The durable version is entity coverage, technical cleanliness, and being the source worth citing in the first place, and that's largely the same work that helps classic SEO too. It's not a separate discipline bolted on. It's the same diagnosis, applied to a newer set of readers.
The brands showing up inside AI-generated answers aren't the ones gaming a new algorithm. They're the ones whose content was already structured to be understood, by machines and readers alike. GEO isn't a shortcut around good SEO. It's what good SEO looks like when the reader is also a language model. If you're being pitched a GEO package that isn't grounded in entity data and structural cleanliness, ask what it's actually built on, because right now most of the answers to that question aren't good ones.