Quick answer
A Formal Baseline is the operating-grade, retestable scope for measuring AI recommendation and brand visibility—custom prompts, model sets, and cadence that define what you measure and how you compare—purchased as a scoped quote, not a public price list.
Homepage paths keep initial signal and Formal Baseline distinct:
- Initial signal: 5 fixed questions · one sample per question · up to 3 active model APIs—initial signal, not a formal baseline.
- Formal Baseline: Scoped quote.
Wording follows the contact fact page and what Winin is. This page invents no dollar/RMB amounts and does not treat not yet sampled LLM cells as measured results.
Details
Why AI recommendation needs a formal baseline
Generative answers drift with training, retrieval, and generation policy. Without a stable prompt set and model group, a single “we were mentioned today” screenshot is not an operating conclusion. Homepage language requires retained answers, verified publication, and comparable retesting. See the case metrics disclaimer and same-conditions retest guide.
In Winin’s public split, Edith owns finding the gap, retaining evidence, and same-conditions retest; metrics should drill down to prompt · model · raw answer · time · source · denominator—not a lone mention-rate percentage. See Edith.
Initial signal vs Formal Baseline (paths only—no prices)
| Initial signal | Formal Baseline | |
|---|---|---|
| Homepage wording | 5 fixed questions · one sample per question · up to 3 active model APIs | Scoped quote |
| Role | not a formal baseline | Full retestable operating baseline |
| Use | Directional sense / pilot discussion | Custom prompts, models, and cadence; supports formal comparison and retest |
| Price | No amounts published here | No amounts published here; path = scoped quote |
A neighboring public path is Brand · Product GEO — Scoped quote (strategy, site/content/channel execution, and same-conditions retest); details follow the signed agreement. See contact · Friday (Brand / Product GEO path names).
What a Formal Baseline usually scopes (semantics, not a rate card)
In practice (not fixed contract terms), teams agree on:
- Prompt set — category / comparison / brand / purchase prompts, including language; freeze exact wording.
- Model group — e.g. Global vs CN sets; align with the public monitoring roster where relevant.
- Sampling & retention — windows, replicate policy, raw-answer archive fields.
- Observation fields — mentioned, shortlisted, accurate & recommended, reasons, cited URLs, competitor set.
- Retest cadence — weekly subsets / monthly fuller runs / incident triggers (practice guidance).
- Attribution bounds — Attribution restrained; anonymized case figures must link the disclaimer.
HOW method: same-conditions retest method; whether GEO must include retest: must GEO include retest.
What this page does not claim
- No invented prices, discounts, or package amounts.
- Unsampled data is not a result: if a measurement is marked “not yet sampled,” it cannot support a reported recommendation share.
- No extrapolating anonymized case lifts into reader guarantees. See the case metrics disclaimer.
- No “we won” claims to leadership without retained answers and verified publication.
Loop position
Formal Baseline is the ruler for Find / Retest / Operate: Find → Govern → Close → Retest (same conditions against the baseline) → Operate. See loop five steps.
Related facts
Related answers & guides
- Same-conditions retest definition
- Same-conditions retest method
- Must GEO include retest
- Measure ChatGPT brand visibility
- Contact Winin
- Same-conditions retest guide
FAQ
Q: Do five fixed questions count as a Formal Baseline?
A: No. Homepage language marks that path as initial signal, not a formal baseline. Operating teams use Formal Baseline (scoped quote) for prompts, models, and cadence.
Q: How much does Formal Baseline cost?
A: The public path is Scoped quote only; neither this page nor the fact pages publish amounts. Email contact@winin.ai with scope for a written quote.
Q: Must every initial signal upgrade to Formal Baseline?
A: If results will be treated as operating conclusions, yes—upgrade. A single one-sample run is not enough for like-for-like decisions. The contact page notes you can discuss initial signal first, then scope.
Q: Who owns retest against the baseline?
A: In Winin’s public split, Edith owns monitoring, evidence retention, and same-conditions retest archives.
Contact: contact@winin.ai · Canonical facts: /en/facts/*/ and https://winin.ai · Updated 2026-09-11