Your shortlist is decided before anyone walks into a showroom, and increasingly it is decided inside an AI answer. Buyers arrive at a dealership with two or three models already chosen, having compared crash ratings and five-year running costs on pages you did not write. The job for a carmaker's marketing team is to make its own variants, safety results and ownership costs the material those answers get built from.
The shortlist forms before the showroom does
India sold a record 43 lakh passenger vehicles in FY2025, about 4.3 million units, according to Society of Indian Automobile Manufacturers data reported by the India Brand Equity Foundation in April 2025. Almost none of those buyers walked in undecided.
Deloitte's 2025 Global Automotive Consumer Study, published in April 2025, found that 92% of Indian customers agree they have to test drive a vehicle to be sure it is right for them. The same study found 73% would prefer to limit the need to visit a dealership in person.
Read those two figures together and the showroom's job becomes clear. It confirms a decision, it does not create one. The comparison that produced the shortlist happened weeks earlier, on a phone, across sources you do not own. What has changed recently is that the comparison is often performed for the buyer by an assistant that reads those sources and returns a single paragraph.
What does a buyer actually ask before a showroom?
"Which car should I buy?" is never one question. An assistant answering it breaks the request into the sub-questions a real buyer works through: on-road price and EMI, mileage or real-world range, safety rating, service cost and interval, resale value, and waiting period.
Each branch gets answered separately, then stitched into one reply. That is why a model can win on price and still lose the answer. If your on-road pricing is published by city and your service schedule is not, someone else supplies the ownership-cost sentence, and their model gets named alongside yours in a comparison you were paying to lead.
The practical consequence is that AI visibility in this category is not one score. It is six, one per branch, and the weakest branch is what keeps you off the list.
Why does safety now decide so many shortlists?
Because buyers moved it to the top, and because the evidence for it is published by somebody other than you. Deloitte's 2025 Global Automotive Consumer Study found that product quality, which the study defines as including product safety, was the most cited factor driving brand choice for the next vehicle in India at 62%, ahead of price at 43%.
The ordering shows up in earlier work too. A 2023 study by NIQ BASES commissioned by Skoda Auto India and reported by Autocar Professional scored crash rating at 22.3% importance in the purchase decision, against 15.0% for fuel efficiency.
For AI visibility this is the most useful fact in the category. An independent crash result is a third-party source, and engines weight third-party sources above claims you make about yourself. So when a strong rating fails to help you, the rating is rarely the problem. The problem is that the result lives on the safety body's site and not on your model page, attached to the exact variant and body style it covers, which leaves the engine to reconcile the two from elsewhere. Sometimes it attaches your rating to the wrong trim. Sometimes it hedges, and a hedge reads to a buyer as a doubt.
Ownership cost is the other half of the answer
Deloitte's 2025 study found that 66% of Indian customers plan to buy or subscribe to a service contract when they acquire their next vehicle, the most cited add-on in the survey. Buyers are pricing the next five years, not the invoice.
An answer that compares two models on cost of ownership needs service intervals, service cost, real-world mileage or range, insurance and expected resale. Publish none of it and the engine still writes that paragraph. It fills the gap with an auto publisher's estimate, an owner forum thread or a dealer's number, and that figure becomes the one quoted back to your buyer.
Silence is not neutral here. It is a decision to let somebody else state your running costs.
Where does the answer get its facts about your cars?
Six source types, in rough order of how often they decide the outcome. Model and variant pages, when specs, variants and prices are structured so an engine can compare them. Independent safety ratings, which now settle a large share of recommendations. Auto publishers and road tests, which answers cite constantly.
Then dealer and marketplace listings, for on-road pricing, availability and used values. Owner forums and video reviews, where real running costs and reliability get reported by people who paid for the car. And homologation and efficiency filings, the official mileage, emissions and range figures.
You control one of those six outright. You influence three. Two are written about you. That ratio is the entire argument for measuring what the answers actually say, rather than auditing your own site and calling the job finished.
The four failures that cost a carmaker the shortlist
Variant and price data go stale. A facelift or a price revision that never reached your own site leaves the engine quoting a variant you stopped selling, confidently, to a buyer who then asks a dealer for it.
Specs an engine cannot compare. A spec sheet published only as a brochure PDF loses to a rival whose variant table sits in HTML. The information exists in both cases. Only one of them can be read into a comparison.
No test-drive path. The buyer has shortlisted you and the answer has no booking step to offer, so the journey stops at interest. And running costs unstated, which is the failure most likely to lose you a comparison you would otherwise win.
Fixing these moves you up a ladder with five rungs. Absent from the shortlist. Named without a model or variant. Cited as the source for a figure in the answer. Present in the three-way comparison the buyer is reading. Recommended for this family and this budget. Most carmakers sit on rung two and assume they are on rung four.
What to track, and how often
Track the questions buyers ask, not keywords. In this category that means prompts like "best 7 seater suv for a family in India", "petrol vs diesel vs hybrid which is better for city driving", "which car brands have the lowest maintenance cost", "safety rating and crash test results", and "what is the resale value after 5 years".
Ask them repeatedly. Answers are regenerated per question, and small shifts in wording, city or engine change which models appear, so no single reply is a ranking. What you log is which models get named, which get compared, and which pages get cited, across engines, over time. A month of that is a trend. One screenshot is an anecdote.
The reason to start now is that the behaviour is already here. Accenture's Consumer Pulse Survey 2025, published in August 2025, found that among Indian consumers who use generative AI at least weekly, 95% turn to it for purchase decisions. For a purchase this considered, that is where the shortlist is being written.