Short, honest explainers — what changed, how the engines decide, and what a store can actually do about it.
Generative Engine Optimization is not ranking on a page; it's being part of the answer.
Search gave you ten blue links and a fair shot at #4. AI assistants give one composed answer — a product is either in it or invisible. That flips the job: instead of optimizing a page for a crawler, you're making sure the engines know your facts (burn time, materials, sizing, returns), read them somewhere they trust (your store, UGC, editorial), and prefer them when comparing you to rivals.
GEO work therefore looks less like keywords and more like answering: publish the answers shoppers ask for, keep the machine-readable store impeccable, and earn mentions on the sources the engines cite. Then measure it the only honest way — by recording the actual conversations.
Blind category questions decide most of the outcome — brand questions are the minority.
Most shopping prompts never name a brand: “best scented candle for a small apartment.” The engine builds a candidate set from what it already knows plus live web search, compares along purchase dimensions (quality, safety, price, provenance), and composes a shortlist — often with one clear top pick.
Three levers move the outcome: coverage (do your product facts answer the question at all?), evidence (do trusted third parties corroborate you — Reddit, YouTube, editorial?), and comparability (can the engine line your specs up against rivals without guessing?). Talgeo instruments all three and shows which one loses you each chat.
AI agents don't browse like humans — they knock on a structured door.
Agentic commerce is arriving as a protocol, not a feature: assistants discover stores through /.well-known/ucp and transact through structured endpoints instead of scraping your theme. Stores with the door open get read accurately — products, prices, policies — and become safely buyable by agents; stores without it get guessed at.
The practical checklist is short: expose the discovery file, keep the product feed complete (identifiers, availability, policies), and verify what agents actually see. Talgeo's Technical GEO page runs this checklist against your store and re-checks after every fix.
If a number can't open its transcripts, it's marketing — not measurement.
Talgeo's rule: every metric is a count over a visible denominator. Visibility 55% is 108 of 198 recorded chats; win rate 53% is top-pick in 57 of the 108 appearances. Each count clicks through to the conversations it was made from, and every claimed improvement ships as a before/after pair of the same question.
This isn't pedantry — it's what makes AI visibility actionable. An index that drifts three points tells you nothing to do. A lost chat with its transcript tells you exactly which fact was missing, which rival was preferred and why.