HOW IT WORKS
How a reading is made.
What happens between your photo and the result on screen — how a lookup travels, how uncertainty is shown, what the tool can’t do, and how the guides are kept honest.
The path a lookup takes
Selecting an image only creates a preview in your browser — nothing is sent yet. When you choose “Identify this car,” the browser resizes the image to at most 1,800 pixels on its longest edge, re-encodes it as a JPEG, and sends it with your optional note to this site’s Worker. The Worker validates the request, applies a per-visitor rate limit, and forwards it to Qomza’s analysis gateway, where an AI model produces a structured result. The Worker checks that result against a strict response schema before the page displays it. A name lookup takes the same path without an image, and you can cancel while a lookup runs.
AI writes the reading
Every part of a result — the suggested identity, observations, specifications, model topics, similar cars, and any market estimate — is generated by an AI model, and no human checks an individual reading before you see it. AI output can be incomplete, outdated, or wrong, so treat a result as a starting point for your own checks, never the last word about a specific car.
How uncertainty is shown
Each reading carries one of four states. An identification suggests a make, model, and sometimes a year or trim. An uncertain result names a candidate and says so — “This is a candidate, not a confirmed identification.” When there isn’t enough to go on, the result asks for a closer view or another clue, and a “not a car” result says the input didn’t appear to be a vehicle. When a candidate is shown, a qualitative label appears with it: “Good match,” “Possible match,” or “Early clue.” These are judgements about how well the evidence supports the answer, not calibrated probabilities — we deliberately don’t show a percentage. The guide to how AI car identification works walks through what the model reads and where it fails.
What a lookup can’t do
A photo or name can’t establish an individual vehicle’s condition, safety, roadworthiness, ownership, originality, history, or exact specification. The tool doesn’t decode VINs, search history records or recall databases, compare safety ratings, or inspect anything. “Common model topics” are model-wide research prompts, not findings about your car. A market estimate appears only when a model is clearly identified and always shows its currency, region, and assumptions — it’s an AI-generated guide, not a quote, appraisal, or verified listing range.
How the guides are made
Each guide starts from a reader task and is researched against named, authoritative sources — government registers, safety programmes, and consumer-protection agencies — linked beside the claims they support. A “Sources checked” date on each buying guide shows when they were last verified. Hero images are AI-generated illustrations, labelled as such on the page and listed with their licences on Image credits. Because official services change, the guides are reviewed periodically.
Corrections
If a guide is wrong, say so — reports are read, and a confirmed error is corrected on the page; where the guide shows a sources-checked date, that is updated too. The support page explains how to report a problem.