Four things, in order of measured impact
Answer engine optimisation has collected a lot of vocabulary and not much plain description. Here is the whole job, in the order the evidence says it matters.
1. Measure first, always
Before anything changes we record what the engines say today, against a prompt set built from your buyers' actual language, run in clean sessions. Everything afterwards is compared against that file. Without a baseline you cannot tell improvement from noise — and given AI answers are non-deterministic, there is a lot of noise.
2. Earned third-party coverage
The biggest lever by a distance. Answer engines lean heavily on what other people say about you: earned media outperforms your own website by roughly 325% for citation rates. In practice this means the "best X in Australia" roundups, industry publications, directories and review profiles that engines quote from. The citation map from the audit tells us exactly which ones matter in your category — usually it is a surprisingly short list.
3. Entity clarity
Engines need to be certain who you are before they will recommend you. That means consistent identification across authoritative sources and deep structured identity signals. In a 90-day controlled test this was the single highest-impact change measured: +34% citations overall, +52% on Gemini and Bing.
4. Content built to be quoted
Answer-first structure, real statistics, specific numbers, quotable expert lines, and nothing behind a form. The Princeton GEO study measured lifts of roughly 41% for quotations, 32% for statistics and 30% for inline citations. Gating kills it outright — 14 citations versus 1,847 for the same material ungated.
And the thing we don't do: sell schema markup as the service. We implement it because it helps machines parse pages, but the best causal evidence puts its independent effect on AI citations at approximately zero. The studies are all here.