Methodology
How Wadar measures, what it measures, and what it does not measure yet. This page is written to be verifiable, not reassuring.
Where Wadar stands today
The automated scan is not in production yet. The first reports are produced through AI-assisted manual curation: purchase questions are asked by hand, answers recorded and cross-checked by hand, and the products cited counted by hand.
It is slower, narrower, and covers fewer questions. But it is measured, not invented.
Full automation — thousands of questions asked every night, across several markets, with statistical repetitions — is the next build. This page will be updated the day it goes into production, with the exact switchover date.
What Wadar measures
Wadar measures neither search volumes, nor ad spend, nor sales. Wadar measures what AI assistants answer to real purchase questions.
Example questions asked: “best milk frother 2026”, “gift idea for a 30-year-old man under €50”. These are the wordings a buyer actually uses.
The models queried
Five assistants, chosen because they concentrate the bulk of purchase queries addressed to an AI:
- ChatGPTOpenAI
- GeminiGoogle
- PerplexityPerplexity AI
- ClaudeAnthropic
- Le ChatMistral AI
Queries go through each provider's official interfaces, in accordance with their terms of use. Web search is enabled wherever the model allows it: we want to measure what a real buyer sees, not what the model memorised during training.
The protocol
The same question is asked several dozen times. Models never answer twice identically: a single answer proves nothing, a distribution over thirty answers is a measurement.
A product cited in 648 answers out of 900 scores 72%. That is how often it appears — not a quality score, not an editorial ranking.
The gap between two successive readings. The movement matters more than the absolute level.
The pages models rely on to answer, recorded as they are.
A check of the product's presence in public ad libraries. A product recommended by AI and absent from ads is a golden window.
The lead in days: how it is calculated
Every product detected carries two dates: its first detection by Wadar, and its first appearance in advertising. The gap between the two is the lead.
Two notions that must not be confused, and Wadar never confuses them:
The window has closed, the figure is final: “detected 31 days before its first ad”.
The window is still open, the figure keeps growing: “absent from ads for 23 days”.
“First detection” means: the first time Wadar saw it — not the first time an AI recommended it. When the scan starts on a niche, every already-established product shows up at once, including those that have been advertised for months. Their calculated lead would be wrong. That is why a product detected during the first fourteen days of scanning a niche, or already present in advertising when detected, never displays a lead: it displays “not enough data”.
“Not enough data”
When a measurement does not rest on enough observations, Wadar writes “not enough data” and shows no figure.
This is a rule, not a stylistic precaution. An approximate figure presented as a measurement destroys the only thing Wadar sells: trust in the number.
The orders of magnitude quoted on the site
The home page frames the market in a few phrases: tens of millions of shopping queries asked to AI models daily, a major share of purchase decisions now influenced by their recommendations, traffic that converts better than average.
These are not Wadar measurements. They are orders of magnitude observed across the market, given to set the context — never presented as data verified by us. No precise figure is put forward until it comes out of our own readings.
With the first reports, these phrases will be replaced by our measurements, with their calculation method and reading date.
What Wadar does not do
- No fake reviews, no astroturfing, no manipulation of AI answers. The Playbook only suggests durable, verifiable visibility actions.
- No technical circumvention of the services queried: official interfaces only.
- No guarantee of commercial success. Wadar reduces risk and saves time; execution remains yours.
- No reselling of personal data.
A question about the method?
Write to contact@wadar.app. If a figure looks wrong to you, say so: a correction is published in the following report, in full.