Fix three things before work starts
Without a fixed protocol, no comparison means anything. These belong in writing before the work begins and must not be swapped at acceptance.
- Question set: a fixed thirty to fifty questions covering informational, comparison and decision intent
- Platform list: named explicitly, with Chinese and international engines listed separately
- How you ask: fixed time window and phrasing, so the wording itself does not move the result
Pass criteria differ by metric type
These measure different things and should not be held to the same clock. Applying one timeline to all of them is the usual source of acceptance disputes.
- Fundamentals: crawlers allowed, schema valid, pages fetchable — this should pass 100%
- Citation coverage: at least some pages you control get cited; whole-site coverage is not required
- Answer accuracy: a cited claim must be correct; volume never compensates for errors
- Business outcome: observed over a longer window, not used to accept a single delivery
Keep the raw responses
Save each full response, not just a summary number. Answers change, and without the original record you cannot later establish what was said at the time.
The query tool keeps each AI response and its cited sources for exactly this reason.
Handling disputes
'The answer I saw was different' is the common one. It usually reflects sampling variation rather than a measurement problem: reproduce the question live and compare timestamps.
'Cited, but not the page we wanted' is the other. Define which pages count as yours — often only a specific directory — in the agreement rather than at acceptance.
Writing the protocol down before work starts is the single most effective thing you can do for a clean GEO sign-off, because it removes most disputes before they exist. Sources reviewed on 2026-10-17. This article records a method with verifiable boundaries; it does not claim that any search or AI platform will index, cite or recommend anything.