FactRelay Docs
14-day GEO Baseline Protocol
The complete protocol behind the entry engagement — scope freezing, sampling, review separation, remediation limits, and the retest schedule, day by day.
This is the method contract behind the 14-day GEO Baseline. The commercial page describes scope and price; this page describes exactly how the work runs, so clients and reviewers can audit the engagement against it.
Fixed scope
One brand, one target market, one language, one primary measurement surface (authorized API; provider and model recorded per snapshot). 10–15 buyer questions frozen as Prompt Set v1.0; 3–5 independent samples per prompt; 10–20 approved facts; at most 3 remediation tasks; one early retest plus one scheduled follow-up.
Fixing the scope is not a limitation — it is what makes the result interpretable. Every variable added (second surface, shifting prompts, unbounded facts) multiplies the explanations available for any observed change.
Day-by-day
| Days | Work | Gate to proceed |
|---|---|---|
| D1–D2 | Confirm brand materials; approve facts; freeze prompt set | Fact owner has approved the round’s facts |
| D3–D4 | Baseline sampling; snapshots archived immutably | All runs carry full metadata |
| D5–D6 | Human review of candidates; client confirmation of findings | Every client-visible finding confirmed |
| D7–D8 | Remediation tasks and fact-constrained drafts | Drafts cite only approved facts |
| D9–D10 | Client publishes; URLs/versions/dates logged; retest run or scheduled | Publishing log complete |
The rules that hold throughout
- Panels stay separate. Natural prompts are byte-identical across runs; citation-diagnostic prompts run apart; brand-free controls measure drift (measurement standard).
- Models propose, humans decide. Automated analysis produces candidates only. A human reviewer, then the client, confirm each finding before it exists as a conclusion.
- Opinions keep their attribution. Complaints and reviews are never adjudicated; only generalization is corrected, with facts published alongside (evidence standard).
- Publishing is the client’s act. We draft; you publish; the log records it. No automated edits, ever.
- Retests are like-for-like. Same prompt-set version, same surface, same rules. Anything else is a new baseline, not a retest.
What the protocol does not promise
No visible change within 10 days is guaranteed or implied. Re-crawl and index cycles often exceed the engagement window, which is why the follow-up retest (30 days, optionally 60) is part of the deliverable. A null result at D14 is an expected, reportable outcome — the protocol’s value is that whatever happened, you can see exactly what was measured, changed, and observed.
Deliverables
A customer-facing report includes the approved question scope, observed answers, relevant public statements, recommended changes, and a like-for-like recheck summary. A sanitized example is available in the public demo case; internal processing records are not published.