---
title: "What Formal Baseline means for AI recommendation / brand visibility measurement"
lang: "en"
canonical: "https://winin.ai/en/answers/formal-baseline-ai-recommendation/"
alternate: "https://winin.ai/zh/answers/formal-baseline-ai-recommendation/"
datePublished: "2026-09-12"
dateModified: "2026-09-12"
section: "answers"
---

# What Formal Baseline means for AI recommendation / brand visibility measurement

## 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](/en/facts/contact/) and [what Winin is](/en/facts/winin-what-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](/en/facts/case-metrics-disclaimer/) and [same-conditions retest guide](/en/guides/same-conditions-retest/).

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](/en/facts/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](/en/facts/contact/) · [Friday](/en/facts/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:

1. **Prompt set** — category / comparison / brand / purchase prompts, including language; freeze exact wording.  
2. **Model group** — e.g. Global vs CN sets; align with the public monitoring roster where relevant.  
3. **Sampling & retention** — windows, replicate policy, raw-answer archive fields.  
4. **Observation fields** — mentioned, shortlisted, accurate & recommended, reasons, cited URLs, competitor set.  
5. **Retest cadence** — weekly subsets / monthly fuller runs / incident triggers (practice guidance).  
6. **Attribution bounds** — *Attribution restrained*; anonymized case figures must link the disclaimer.

HOW method: [same-conditions retest method](/en/answers/same-conditions-retest-method/); whether GEO must include retest: [must GEO include retest](/en/answers/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](/en/facts/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](/en/facts/loop-find-govern-close-retest-operate/).

## Related facts

- [Edith](/en/facts/edith/)
- [Case metrics disclaimer](/en/facts/case-metrics-disclaimer/)
- [Contact](/en/facts/contact/)
- [What Winin is](/en/facts/winin-what-is/)
- [GEO loop five steps](/en/facts/loop-find-govern-close-retest-operate/)

## Related answers & guides

- [Same-conditions retest definition](/en/answers/same-conditions-retest-definition/)
- [Same-conditions retest method](/en/answers/same-conditions-retest-method/)
- [Must GEO include retest](/en/answers/must-geo-include-retest/)
- [Measure ChatGPT brand visibility](/en/answers/measure-chatgpt-brand-visibility/)
- [Contact Winin](/en/answers/winin-contact/)
- [Same-conditions retest guide](/en/guides/same-conditions-retest/)

## 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