Every number we show is reproducible. This page documents exactly what we run, how we score it, and where the limits are — so a dealer can independently verify any claim we make on their result page.
We send 6 fixed shopper-style questions to a large language model on the Lovable AI Gateway. The questions cover Research, Cross-Shopping, Smart Money, Service, Financing, and Hyper-Local intent — the same intent categories used in the paid InventoryPilot audit. The 6 templates never change between checks.
Google gemini-3-flash-preview via the Lovable AI Gateway. The model name and the exact UTC timestamp are stamped on every result page.
Each of the 6 questions runs twice. We only count a question as "mentioned" when both runs agree your dealership appears in the AI's top-5 recommendations. If the two runs disagree, the question is flagged inconclusive — it does not count for you and does not count against you.
We compare the AI's listed dealership names against your input using a normalized, token-based match: punctuation/location words are stripped, generic brand words (Ford, Chevy, Toyota…) are ignored, and we require every distinctive token from your store name to appear in the candidate. This is the same matcher used in the paid product.
When you are mentioned, the AI lists 5 dealerships in ranked order. Fixed position scores: #1 = 100, #2 = 80, #3 = 60, #4 = 40, #5 = 20. We report the better of your two runs' positions.
We do not claim this is your "Google rank," your "SEO score," or a prediction of future AI behavior. It is a snapshot of how one model answered 6 specific questions at one moment in time. Different models, different cities, and different phrasings will produce different results — which is exactly why the full paid product runs 30+ queries across multiple intents on a recurring schedule.
One check per dealership per 30 days (cached result returned on repeat). Maximum 3 unique checks per IP per day.