Skip to content
Vouch
Start free trial
Menu

How Vouch measures

Vouch asks AI assistants what your buyers ask, follows the conversation the way a buyer would, and records where your brand is named, recommended or dropped.

How we collect answers

  • Every conversation starts fresh, with no history about you or your rivals.
  • Assistants tracked today: Claude, Gemini, ChatGPT, Perplexity (API answers), Google AI Overviews, Google AI Mode and Copilot (Bing AI answer).
  • Claude is asked through Anthropic's API with web search on, so we see the pages each answer cites.
  • Gemini is asked through the public website, logged out, as a visitor sees it. If the website is unavailable, or cannot be reached from your country, we may ask Gemini through its API via a data provider instead. The run page says so, because that answer is not pinned to the country.
  • ChatGPT Threads and Perplexity (API answers) are asked through the assistant's API via a data provider.
  • For ChatGPT daily tracking, Google AI Overviews, Google AI Mode and Copilot (Bing AI answer), we record the search answer as a data provider retrieves it.
  • Answers collected through a data provider can differ from what a signed-in user sees in the app.
  • Threads are not available on Google AI Overviews, Google AI Mode and Copilot (Bing AI answer): each shows one answer per search, so there is no conversation to follow.
  • We never solve a human check. If an assistant asks for one, the run stops and we try again later.
  • Tracking runs once a day for every prompt, assistant and country. Where Threads are on, each of them gets 5 Threads a week.

The six stages of a Thread

A Thread is one buyer's conversation, six questions long. The first question is your tracked prompt. Vouch writes the next five for your category.

  1. 1. Discovery. The opening question: the tracked prompt itself.
  2. 2. Shortlist. Narrow down and ask to compare the leading options from the previous answer for your use.
  3. 3. Objection. Raise the typical doubts and downsides about the leading options.
  4. 4. Trust. Ask about reliability, after-sales service, warranty and real owner reviews.
  5. 5. Value. Ask whether the leading option is worth its price compared with the alternatives.
  6. 6. Decision. Ask for one clear recommendation and where to buy it.

How a conversation is scored

Each conversation scores from 0 to 1. Half of the score is how many of the six stages your brand survives: it survives a stage when the answer names it, and every stage before it named it too. The other half is whether the final answer recommends you. A change in how warmly the assistant talks about you moves the score by up to a tenth, up or down.

score = (0.5 × stages survived ÷ 6 + 0.5 × chosen) × (1 + 0.1 × change in sentiment), kept between 0 and 1

The Vouch Rate is the average score of the week's completed conversations, times 100. The salvation rate is the share of those conversations whose final answer recommends you.

Why results vary from run to run

Assistants rarely give the same answer twice. That is why each Thread runs 5 times a week, and why a week with fewer than 3 completed conversations shows "Not enough conversations yet" instead of a number. Treat a small change from one week to the next as noise until it holds for a few weeks.