Feed audit
Scores every product page on what an AI needs to read there, and tells you which ones to fix first.
The feed audit answers one question: do your product pages contain what an AI needs to read in order to recommend you? Every product in your catalogue gets a score out of 100, and the page tells you where to start. "Lancer l’audit" re-runs the measurement whenever you want to re-check.
A deterministic score, not an opinion
No AI is involved in the calculation. It’s a grid of 9 weighted checks, applied identically to every product page. Two audits over an unchanged catalogue return exactly the same result. So when a score moves, it’s because your page moved — not because the thing grading it moved. That’s what makes comparing two audits actually worth something.
- Description present: 20 points
- Title present: 15 points
- Detailed description, at least 120 characters: 15 points
- Attributes and tags: 10 points
- Image: 10 points
- Price: 10 points
- Descriptive title, at least 15 characters: 10 points
- URL: 5 points
- SKU reference: 5 points
Reading the counters
"Score moyen" (average score) gives the overall state of the catalogue, and the three counters next to it break that down: "Fiches bonnes" (80 and above), "À améliorer" (50 to 79), "Faibles" (under 50). "Évolution du score" shows up from the second audit onward, once there’s finally something to compare against. The table itself is sorted worst-first, with filter tabs and search: you start at the top and work down.
The per-product detail
Open a product to see every check passed or failed, the fixes ranked by priority, and which crawlers can reach the page: GPTBot, ChatGPT Search, ClaudeBot, Perplexity and Google AI. A blocked crawler is a more serious problem than a description that’s too short. If the bot can’t read the page, none of the rest of the work counts.
Pushing to Bing and ChatGPT
"Pousser vers Bing et ChatGPT" submits your URLs for a fresh crawl instead of waiting for the bots to come back on their own. The button appears after a first audit, once your Shopify store is connected, and can run once per hour. It reaches Bing and the engines built on top of it, which includes ChatGPT web search. Google is not covered: it doesn’t support this protocol. Use it after a batch of fixes, not before.
The structured-data (schema.org) check only fetches a limited number of product pages per run. Beyond those, the check reads "non vérifié" — treat that as unknown, not as a failure. Run the audit again to cover more pages.