# How do I verify a GEO partner's work?

> Winin connects approved brand and product facts to content, structured data, feeds, channels and credible sources, then uses repeated measurement to observe whether discovery, citation or recommendation changed under comparable conditions.

- Canonical page: https://winin.ai/en/geo/verify-vendor/
- Language: English
- Last updated: 2026-09-04

**Look for six things.**A fixed question set, model coverage and test counts, the calculation scope of every metric, a relevant comparison design, the raw answer archive, and a complete optimization and post-test history. Together they show how the partner found the opportunity, acted and verified change.

- Ask for denominators, not case-study screenshots.
- Ask for a frozen question set, or two runs are not comparable.
- Ask for raw archives: every metric must drill down to source text.
- Ask for a control group, or drift and model upgrades stay unexplained.
- One vendor may execute and measure, but protocol, delivery and review need separate roles.

## Why execution and verification need separate roles

A professional program separates the measurement scope, delivery actions and result review, making every conclusion traceable to the raw record.

Fix the measurement scope before work starts, retain the full run register, and have someone outside the delivery work review the result. Material claims should also be reproducible by the customer or a third party.

## What the six artefacts look like

Question set: a versioned list, each item tagged with intent, immutable once frozen.

Sampling design: which engines, repeats per question, time window, closed-book or retrieval-augmented.

Metric contract: numerator and denominator per metric, plus the rule applied when the denominator is thin.

Control design: which question clusters are treated, which are controls, and how they were assigned.

Raw archive: the complete original answer per observation, with citations traceable back to an offset in the source.

Complete history: every test, optimization, publication and post-test result.

## Frequently asked questions

### The vendor says denominators are confidential. Fair?

The denominator describes the measurement scope of your own brand, not their trade secret. Withholding it means you cannot judge how representative the result is.

### Can we verify this ourselves?

Partly. Fix a set of questions, ask them on two or three engines on a regular cadence, and keep the raw answers. That does not replace systematic measurement, but it surfaces obvious discrepancies.

### Does winin.ai also provide execution?

Yes. winin.ai provides customer-approved content, data, feed, channel and credible-source improvements. To control self-review risk, the protocol is frozen before delivery, raw evidence is retained, and customers or third parties can repeat the same test. Nothing is auto-published without approval.

[Start with a free initial check](/en/free-ai-presence-check/)

## Source

This Markdown document is the machine-readable counterpart of https://winin.ai/en/geo/verify-vendor/. The canonical HTML page remains the public presentation source.