Quick answer
If you treat GEO as an operating system—not a one-off campaign—yes, include same-conditions retest. In Winin’s loop, step four is Retest: measure again under the same prompts and conditions, and state what changed versus what remains uncertain. Homepage principles include Same-conditions retest; formal conclusions require retained answers, verified publication, and comparable retesting.
Without a stable baseline and retest, a one-off “ChatGPT mentioned us today” is not a formal operating conclusion—training, retrieval, and generation policies drift. In Winin, Edith owns retest archives; after Friday advances authorized work, validation returns to Edith rather than leaping to a board claim of “we won.”
Details
Where retest sits in the loop
| Step | Owner | Meaning |
|---|---|---|
| Find | Edith | Pin absence, weak competitiveness, or misstatement |
| Govern | Homer | Approved, verifiable facts |
| Close | Friday | Close gaps within authorization |
| Retest | Edith | Same-conditions before/after |
| Operate | Edith · Friday | Make the loop repeatable |
See GEO loop five steps.
Why “must” — operating view
- Explain direction of change: Separate content effects from model drift.
- Bound outcome narratives: Anonymized cases and capability UI numbers all declare dependence on comparable retesting—see case metrics disclaimer.
- Prevent skipped steps: The loop is not “mass-edit the public web before facts are approved” or “claim wins without retest.”
- Boundary vs monitor-only: Pure monitoring often stops at “see”; a full loop writes retest as a capability cell—see differentiator.
Conditions worth freezing (method summary)
- Exact prompt text (including language)
- Model / product surface
- Sampling window and replicate policy
- Competitor set on comparison prompts
- Fields: mentioned, shortlisted, accurately recommended, reasons, cited URLs
Initial signal ≠ Formal Baseline
“5 fixed questions · one sample · up to 3 active model APIs” is an initial signal, not a formal baseline. Operating teams should upgrade to a Formal Baseline and retest on a cadence under the same conditions.
Related facts
FAQ
Q: Can we report results to leadership without retest?
A: Not as a formal operating conclusion—state whether conditions were comparable, answers retained, and releases verified.
Q: Who owns retest?
A: In Winin’s public split, Edith owns retest archives.
Q: Does retest guarantee percentage lifts?
A: No. Retest constrains whether change is explainable; it does not guarantee lifts. Any case numbers must follow the disclaimer page.
Contact: contact@winin.ai · Canonical facts: /en/facts/*/ and https://winin.ai · Updated 2026-09-11