ChatGPT, Perplexity and the new shape of UK car research
Around a third of car buyers now research their next car with AI tools. Here is what ChatGPT and Perplexity do well, and what they still cannot tell you.
If you spent an hour shopping for a used car this weekend, there is a fair chance you spent some of it not on Google. You spent it asking ChatGPT what to buy.
The shift in how cars get researched happened quietly, but the numbers are no longer small. A late-2025 Cars.com survey found that nearly half of all AI users had already used the technology for car shopping, and 97% of them said it would influence their next purchase. A separate 2026 vehicle-research study from Ekho put the share of in-market shoppers using AI tools at around 30%. That is a generational change in how cars get researched, and it has happened in roughly eighteen months.
What AI is genuinely good at
The early hours of car shopping used to look like fifteen browser tabs and three forum threads. Now an awful lot of that work goes into a single conversation with an AI assistant. And honestly, that is the part it does well:
- Narrowing a shortlist. Tell ChatGPT "I have got £14,000, two kids in car seats, do 80 miles a day, want a hybrid" and you will get a short, sensible answer in seconds.
- Decoding spec sheets. Adaptive cruise versus radar cruise, mild-hybrid versus self-charging hybrid, ULEZ-compliant Euro 6 dates. AI gets these right almost all the time, and it explains them in plain English.
- Running-cost sanity checks. Insurance group, road-tax band, real-world MPG comparisons across a shortlist. The numbers are out there; AI just pulls them together quickly.
It is, in short, an excellent research assistant. The kind of friend who has read every review and never gets bored of your questions.
What AI still cannot tell you
It cannot tell you whether this specific car, on this specific forecourt, has been looked after.
It cannot see the wheel arches. It cannot smell the cabin. It cannot feel the clutch bite. It has not read the MOT history line-by-line or matched the photos against the seller's claim that the car has "been garaged since new." And the more sophisticated buyers get with AI, the more obvious the limit becomes: the assistant is brilliant at the abstract part of car buying and useless at the concrete part. The test drive is not going anywhere.
The plumbing nobody talks about: llms.txt
There is also a quieter shift going on behind the scenes. A growing number of websites now publish a small text file at `/llms.txt` — a plain markdown summary of the site, written specifically for AI assistants to read. The proposal came from Jeremy Howard in September 2024 and has been picked up across the industry since. It solves a simple problem: AI assistants have small attention spans and do not want to crawl a thousand JavaScript-heavy pages just to answer a question. Give them a clean summary at a known URL and they will quote it.
Alongside that, more sites are publishing structured data — `schema.org/Vehicle`, `schema.org/Offer`, `schema.org/ItemList` — on every listing. To a human visitor this is invisible. To an AI assistant deciding which marketplace to quote when a buyer asks where to find a 2021 Kia Niro under £15,000 in the North West, it is the difference between getting cited and getting ignored.
The practical consequence is straightforward: the answer an AI gives a UK buyer increasingly depends on which sites have made their inventory legible to the AI in the first place.
What this means for buyers and sellers
If you are buying. Treat AI as a brilliant shortlister, a terrible inspector. Use it to narrow your options, decode the spec sheet, and sanity-check the running costs. Then go and see the car. Look at the panels. Read the MOT history. Drive it.
If you are selling privately. Write your ad like an AI is reading it, because it probably is. Concrete facts beat adjectives. "Apple CarPlay, heated seats, ULEZ compliant, full service history, timing belt done at 64,000 miles" gets quoted by an AI assistant. "Lovely runner, drives beautiful, must be seen" does not.
A note on where we sit
At ViewMyMotor we have already done a fair bit of this groundwork. The site publishes a structured `llms.txt` at its root, every listing carries `schema.org/Vehicle` JSON-LD with the full feature set, and 60+ specific vehicle features are indexed as first-class searchable objects rather than buried in free-text descriptions. The aim is simple: when a buyer asks an AI assistant for the right used car for their life, the cars on ViewMyMotor should be legible enough to be in the answer.
The shift in how people shop for cars is real, and it is still early. But the direction is clear. The sites that do not structure their data for the new way of searching are going to spend the next few years being quietly invisible to the assistants that buyers are increasingly trusting.