How PromptGraph measures AI visibility for dealerships
Every number we publish comes from a method you can reproduce. Here is the whole thing, including what we do not measure.
AI visibility is easy to claim and hard to prove. A dealer asks "am I in ChatGPT?" and gets a screenshot back. That is not measurement. This page is the method behind every score PromptGraph reports, so a GM can check our work and a competitor can argue with it on the facts.
Two scores, because there are two questions
Readiness answers "can an AI agent read and act on this store?" It is a fixed set of checks against the website. It is deterministic: run it twice, get the same answer.
Visibility answers "does AI actually name this store?" It is a prompt-based test across engines, locations, and repetitions. It moves day to day, so we report it with the run count and the date.
Most vendors publish one number and let you guess which question it answers. We publish both, separately, because readiness is the cause and visibility is the effect. When the second number is low, the first one tells you why.
Readiness: the checks and their weights
Every rooftop is fetched the way a browser would fetch it: homepage, robots.txt, llms.txt, sitemap.xml, and a short list of well-known paths for agent endpoints. Scores are absolute. A 100 means every check passed. Nobody is graded on a curve, and the leader in a market is not set to 100 by definition.
| Check | Points | Why it matters |
|---|---|---|
| MCP endpoint | 20 | Lets an AI agent query the store directly |
| Machine-readable inventory | 15 | A JSON feed at a known path, so AI does not guess what is on the lot |
| llms.txt | 10 | Tells language models what the site is and where the important pages are |
| AI discovery file | 10 | A machine-readable description of the business for agents |
| AutoDealer schema | 10 | JSON-LD identifying the site as a dealership with address, hours, brands |
| GPTBot, ClaudeBot, Google-Extended allowed | 5 each | robots.txt does not block the crawlers that feed AI answers |
| XML sitemap | 5 | Declared in robots.txt or served at /sitemap.xml |
| Name and phone on the homepage | 5 | Lets AI match the site to the business listing |
| Any structured data | 4 | At least one JSON-LD block |
| Organization or LocalBusiness schema | 3 | Entity-level connection to parent and location |
| FAQ schema | 3 | Marked-up answers AI can quote directly |
The agent layer carries 55 of the 100 points on purpose. Schema and a sitemap are 2019 hygiene. What separates stores in 2026 is whether an agent can do anything after it finds you.
Across the first 4,287 franchise rooftops we scored, the average was 40.7 and the best store hit 62. You can browse every one of them on the Readiness Index.
Visibility: the prompt test
Prompts. Non-branded, in three buckets a shopper actually uses. Find: "best Toyota dealer near Plano." Buy: "RAV4 inventory near Plano," built from the top searched models in the market. Service: "where to service a Toyota near Plano." Branded prompts inflate a store's own number, so we do not count them toward visibility.
Engines. ChatGPT, Google AI Mode, Google AI Overview, Gemini, and Grok. Each is reported on its own. A blended number hides which engine is doing the work.
Location. Two lenses. Geo-pinned: the request exits from a residential connection inside the city named in the prompt, which is what a shopper on their phone gets. Geo-neutral: no location signal, which is what a relocating buyer or an out-of-market shopper gets. The gap between the two lenses is often the most useful thing in the report.
Repetitions. AI answers vary run to run. Every prompt is run at least three times on three different days. We report the spread, not just the mean. If two stores are two points apart and the spread is five, we say they are tied.
Metrics. Visibility is the share of answers naming the store. Share of voice is the store's mentions divided by all dealer mentions. Named first is the share of answers where the store is listed first. Citation share is which domains the engine cited, so you can see whether AI learned about the store from its own site, from DealerRater, or from a competitor's page.
Weighting. Cities inside a market are weighted by search volume for the brand and its top models, so a store's score follows where the demand is.
What we do not measure, and say so
Sentiment. Every "AI sentiment" score we have seen for dealers lands in a narrow band that never separates stores. We are not going to decorate a report with it.
Business outcomes. Visibility is not sales. When a client gives us GA4 access we report LLM referral sessions and leads next to the visibility number, clearly labeled as a separate measurement.
Corrections
Every readiness check is reproducible with a browser. Every visibility run is stored with its prompt, engine, location, timestamp, and raw answer. If a store believes a result is wrong, email Gray@PromptGraph.ai with the store and the check and we rescan within two business days.
How is this different from CarGenius's dealer AI visibility reports?
Their reports measure visibility only, on three engines, with relative grades where the market leader is always set to 100, and they do not disclose prompt counts or repetitions. We measure readiness and visibility separately, on five engines, on an absolute scale, and publish run counts and spread. Full comparison on the PromptGraph vs CarGenius page.
How often are scores refreshed?
Readiness is rescanned monthly for every store in the index. Visibility runs monthly for tracked stores and on demand for any new client.
Can a dealer dispute a score?
Yes. Email the store name and the specific check. We rescan and correct or explain within two business days.
See where your store stands today.
PromptGraph runs the readiness checks and the prompt-based visibility test on your rooftop, then deploys the fixes. Free scan first, no card.
Run the free AI visibility scan