How can an automotive brand track whether ChatGPT recommends its models when people ask which car to buy?
How an automotive brand can track whether ChatGPT recommends its models when people ask which car to buy, with prompt sets, citations and honest limits.
Car brands ask us a version of the same question: when someone asks ChatGPT which car to buy, does it recommend our models? The answer is trackable, but only if you set up the measurement properly. Here is how we approach it.
Start with the buyer's words
Car buyers rarely name a brand first. They describe a need: a family of five, a long commute, a first car, a small business that needs cargo room. So the prompt set starts there. We write discovery prompts with no brand named, comparison prompts that name two models, and objection prompts such as reliability or running costs. Then we split by persona (first-time buyers, families, fleet managers) and by country, since model ranges and availability differ by market.
We ask the client's sales and dealer teams what buyers actually say. Their phrasing beats anything we would invent.
Track brand and model separately
A brand can be named while the model being launched is not. We track the brand, each model line and the competing models. For each prompt we record whether the client appears, where in the answer, and in what tone.
The platform we run client programs on is Promptwatch. It runs prompt panels daily per model, country and language, with ChatGPT (Search) monitored through the real interface. We read Visibility Score, Share of Voice, Sentiment and Position in Answer, with competitor models in the same prompts. The Brand Book lets us define what counts as a mention, which helps when model names have variants.
Read the sources behind the answer
ChatGPT's car recommendations are shaped by what it can find. That tends to include review sites, forums, video reviews and news. Promptwatch's citation analytics shows which pages and domains are cited, and the Reddit and YouTube citation reports cover forum and video sources. The offsite mentions report finds third-party pages naming the brand without a link. Those become a PR and partnership list for the client's team.
Check that your own pages are readable
Model pages, spec sheets and comparison pages need to be reachable by bots like OAI-SearchBot, GPTBot and ChatGPT-User. Promptwatch's IP-verified crawler logs show whether they arrive and what errors they hit. Logs start on the Professional plan, so we tell clients up front that the Essential tier will not show this.
Local intent
Car shopping often has a local angle, like nearby dealers. State and city tracking in Promptwatch starts on Business, and local GEO and SEO targeting can pin a monitor to exact coordinates for ChatGPT, Google AI Overviews and Google AI Mode. We only recommend that tier when the client's market really needs it.
Connect to outcomes
Visitor analytics tracks AI-referred sessions and conversions. If the client has a configurator, test-drive form or dealer locator, we map those as conversions so we can report on action, not only on mentions.
How we report
Monthly, we show the client three things: share of voice against named competitor models, the sources shaping the answers, and AI-referred visits. We highlight prompts where a rival model is named and the client is not, and the pages we will fix first.
What we tell car brands honestly
We cannot make ChatGPT recommend a model. Answers vary by run, wording and country, and recommendations reflect third-party sources the brand does not control. Anyone promising a guaranteed recommendation is guessing. What we can offer is a clear baseline, a list of fixes backed by evidence, and measurement to show what changed.
For a brand that wants to try the method first, a free Explore account covers ChatGPT with 10 prompts, enough to see how the answers look before committing to a paid plan.