GEO Mediapilote, a brand visibility audit for AI search

People increasingly ask AI assistants their questions instead of a search engine: that is the ground covered by GEO, optimising for AI search. Mediapilote wanted to offer clients an audit of that visibility: is the brand mentioned, is its website cited, does a competitor come first?

The project started in October 2025 as a WordPress plugin with a React interface, before being rewritten as a standalone Next.js app in spring 2026. It now has a public page presenting the service.

An audit crosses dozens of queries with four assistants, each with its own quotas and response times. All of it had to run in the background, with progress tracking and the ability to resume an interrupted audit.

Scoring raised a reliability issue: an AI answer graded by another AI must get a stable, justified score, otherwise the report is worthless to the client. Recommendations had to rest on verifiable evidence rather than generic advice.

I wrote an audit orchestrator that queries the four assistants in parallel with concurrency tuned for each, then scores every answer on seven criteria (mention, link, position, accuracy, freshness, error and competitor penalties) using JSON Schema structured output. A technical audit (robots.txt, llms.txt, sitemap) and a check of the brand’s presence on 14 platforms complete it.

Recommendations are generated by a model that investigates the site with read-only tools and cites the pages it read. The client report is a GSAP-animated page that can be exported to PDF. The tool also connects to Search Console, GA4 and the internal SEO tool, and the actual cost of each audit is recorded.