GEO — generative engine optimisation — and AEO — answer engine optimisation — are two labels for one job: understand how models talk about a brand when a person asks, then change what those models can retrieve.
The category sells “visibility.” Visibility is a fog word. We split it. A mention is the brand name in the prose. A citation is a URL the model names as a source. A recommendation is the model telling someone to use you. Those three columns are the product.
Mixing them into one “ChatGPT rank” is folklore. Models name you while warning people off. That is a mention with hostile sentiment, not a win. Models cite a competitor’s domain while saying your name. That is a mention without the citation you wanted.
Share of voice is your slice versus named peers on the same prompt panel, same engine, same date. Without a peer list it is theatre. Position is where you sit in the reply — first named, or buried after three other brands. It is not a Google rank, and not a promise of “#3 in ChatGPT.”
Prompt tracking is the instrument: a frozen set of questions, run on a schedule, against ten named engines. Query fan-out is what happens inside the engine — one user ask becomes several retriever queries. Write for the splits, not for one polished sentence.
Consumer UI and a model API are not the same answer. How we measured belongs on the label. Stochastic answers need repeats. One lucky screenshot is not a method you can defend in a meeting.
Ten engines are surfaces, not a league table. ChatGPT is not Gemini Apps is not Perplexity. We name the surface. We do not sell a branded index with a locked formula nobody else can audit.
That is what GEO measures here. The next move is still ordinary: let the crawler in, publish extractable facts, earn mentions on domains the model already cites, and slice again in a month.