Providers, modes and languages
The four assistants
Section titled “The four assistants”| Provider | Assistant | Model used (consumer default, verified 2026-07) |
|---|---|---|
| OpenAI | ChatGPT | gpt-5.6 |
| Anthropic | Claude | claude-sonnet-5 |
| Gemini | gemini-3.5-flash-lite | |
| xAI | Grok | grok-4.5 |
Every call goes through the provider’s official API, with the same model its consumer assistant uses by default — except Gemini, which runs on Flash-Lite, the lighter sibling of the Flash model behind the Gemini app, to keep daily tracking affordable. No scraping, no browser automation, no terms-of-service games — and the same question gives the same kind of answer a real user gets. Models are updated when the providers change their defaults, and every run stores the model that answered it, so a jump in the index can be traced to a model change rather than a market move.
Only the assistants the service has an API key for are asked. An assistant that is not configured is left out of the batch (and of the index) — it is never scored as “not recommending you”.
Two modes
Section titled “Two modes”- Plain — the model answers from what it learned in training. This is the slow-moving battle: it changes with model releases, and is influenced by how your brand is represented across the public web.
- Grounded (Starter and Pro) — the model runs a live web search (each provider’s own search capability) before answering. This is the fast battle: it changes with what the search index returns today, and the answer carries citations you can act on.
The two often disagree. A brand well established in training data can be absent from grounded answers dominated by a fresh “best of 2026” listicle, and vice-versa. On paid plans both are tracked and reported separately and the blended index averages them; the Free plan runs plain mode only.
Languages
Section titled “Languages”AI answers change with the language of the question — different sources, different brands, different order. Questions are therefore tracked natively per language (written in that language, not translated), and scores are sliced per language. Supported today: English, Spanish, Catalan, French, German, Italian, Portuguese, Dutch. Plans limit how many languages a project tracks (Plans).
Schedule
Section titled “Schedule”A batch is one full pass over every active prompt × assistant × mode (the plan’s) × language. Batches run daily on paid plans and weekly on Free; Run now triggers one on demand. Every answer, extraction and score is stored, so history is never recomputed from memory.