GEO — generative engine optimization: getting recommended by AI answer engines the way SEO gets ranked by Google. The method the owner invented and ran, over a long period, for the FlexMesh site:
- SEO first, then wait for indexing — being in the index is the entry ticket.
- Key discovery: answer engines crawl live and generate live. They fetch pages at query time — so the feedback loop for GEO changes is minutes, not the weeks of a re-index cycle. This is what makes the rest of the method possible.
- Impersonate the target audience: put the real users' questions to the assistants ("best app for multi-carrier parcel drivers in Canada", …).
- When not recommended, interrogate the engine: "why didn't you recommend FlexMesh?" — the judge states its own rubric. The model's explanation is the gap report.
- Patch the site per the explanation; re-ask; loop.
Why this matters beyond FlexMesh: it turns GEO from guessing-the-algorithm into an eval loop — a machine-checkable verdict iterated against, the same shape as an adversarial test harness. The generator that passes you over is also the cheapest consultant on why.
Measured outcome (2026-07-10): it worked. The web property's channel split shows AI Assistant: 29 sessions / 27 users in 90 days — assistants are citing the site and sending people, to a zero-budget niche product, in a channel most marketers can't show a single session from. Alongside it, the SEO substrate the loop was built on delivers 339 organic sessions / 273 users per 90d. The remaining question moved downstream: citations arrive, readers arrive — installs don't follow (channel-fork-seeding-or-unlit, the site→install loss).