Automotive: How AI Helps Shoppers Pick a Car

A buyer who asks ChatGPT "which electric car should I choose with a 35,000 € budget" gets a shortlist of two or three models, not a catalog. A French study published in September 2026 shows that shortlist changes sharply depending on which AI engine answers. For a carmaker or a dealer network, the question is no longer just ranking well on Google: it's knowing what each AI says about its lineup, its reliability and its price, to buyers who may never visit its own site before deciding.

This article covers how buyers already use AI in their car-shopping journey, what a study on electric vehicles reveals about the gap between ChatGPT and Gemini, which sources carry authority on reliability and road tests, and what structured spec sheets can (and cannot) change.

Are car buyers already using AI to choose a vehicle?

Yes, but how much depends on what gets measured. A Google-Kantar study run in India in May 2026, among 1,022 buyers who had purchased a vehicle in the previous six months, found that 93% of them used some form of AI during their research. The figure uses a broad definition, though: conversational assistants like Gemini, but also Google Search's AI Overviews and AI Mode. Buyers who used AI also consulted more sources, averaging 7.8 touchpoints versus 6.9 for buyers overall (Google-Kantar, May 2026).

The study was commissioned by Google, which has a direct commercial interest in showing its own AI features gaining ground; it also covers the Indian market, not France. We found no equivalent, independent study isolating the use of a dedicated chatbot (ChatGPT, for instance) for car buying in France specifically. What can be said with confidence: conversational assistants are now a documented step in the car-buying journey in several markets, and nothing suggests France is an exception.

What do ChatGPT and Gemini recommend for an electric car?

Not the same thing. That's the conclusion of a study published on September 14, 2026 by the agency Bubbling, in media partnership with Autoactu.com. The method: 2,016 simulated conversations and 8,064 questions, put to both engines between August 20 and 26, 2026, built from 36 questions and three buyer profiles (an affluent family, a modest household eligible for social leasing, a young urban buyer) (journalauto.com, September 14, 2026).

The results show clearly different recommendations depending on the engine:

• ChatGPT names Tesla in nearly a quarter of answers where the buyer doesn't specify a brand, and surfaces Chinese brands like MG and BYD far more often than Gemini does.
• Gemini puts Renault first, with 26% of citations, and favors French brands more broadly.

The authors are explicit about the limits of their own work: the exercise covers simulated journeys, with no measurement of real commercial impact, and no guarantee that the same questions produce the same answers at another point in time (Autoactu.com, En voiture Carine). Bubbling itself advises companies on their visibility inside AI conversations, which doesn't invalidate the method but does mean the result should be read as a snapshot, not a durable law. For a car brand, the operational takeaway fits in one sentence: "being visible in AI" isn't enough — you need to know in which engine, on which questions, and how far behind your competitors, which is exactly what an AI visibility audit measures by querying several engines rather than one.

Which sources do AI engines cite about a car?

Overwhelmingly third-party sites, not the manufacturer's own site. That's the general finding of our calibration study on the pages AI engines actually cite (methodology): between 79% and 89% of citations go to third-party sources, across sectors. For cars, those third-party sources take familiar shapes: specialist press road tests, comparison sheets, owner forums and used-car comparison sites.

In practice, a carmaker or dealer network directly controls only part of what an AI can read about its vehicle: its own product page, yes, but also what independent road tests say about it, what buyers publish on third-party review platforms, and how comparison pieces stack it against direct competitors. Ignoring those third-party sources, by betting only on your own site, means leaving most of the conversation to others. We cover this mechanism, which applies across every heavily-compared sector, in our article on how ChatGPT picks its sources.

Reviews published on third-party platforms play a particular role here: they're numerous, dated and often quantified, three traits that make them easy for an AI to pick up. We cover how to structure them and make them findable in our guide on customer reviews and AI visibility.

Reliability: which ranking do AI engines repeat?

Car reliability rankings don't all measure the same thing, and an AI giving a quick summary can blend different methodologies without saying so. The biennial survey from French consumer magazine Que Choisir (issue 654, February 2026), based on more than 53,000 testimonials from European drivers collected between May and July 2025, ranks Lexus first among 36 brands, followed by Toyota, Suzuki, Subaru, Honda, Tesla, Smart and Mazda. French brands rank far lower: Renault 26th, Peugeot 32nd, Citroën 33rd, DS Automobiles 34th (La Tribune Auto, covering the Que Choisir survey).

Other organizations, such as the US-based J.D. Power or Consumer Reports, publish rankings built on different panels and criteria, and don't always reach the same conclusions for the same brands. For a French brand sitting mid-table on the most-cited European survey, the right GEO response isn't to dispute the number, but to document what has changed since: extended warranties, fault rates by model generation, feedback from repair networks. A page that dates its own figures and links to the full survey is more likely to be cited with nuance than one that ignores the topic.

Structured spec sheets: useful, never decisive

Schema.org defines a Car type, with dedicated properties (engine power, torque, fuel capacity, wheelbase, fuel type) that let you describe a technical spec sheet in a structured way. Google also offers a structured-data format specific to vehicle listings, but that rich result is limited to verified dealerships and currently available only in English, in the US and its territories (Google developer documentation): a French site has no access to it, however good its markup.

That doesn't make Car markup useless: a well-structured, consistently dated spec sheet is still easier for a crawler to read than a table baked into an image or a PDF. But our calibration study is clear on this point, and it holds for cars just as it does for any other sector: structured data and files like llms.txt help a crawler understand a page, they are not a decisive citation factor. Google itself doesn't present them as necessary for its own AI results. What actually drives citation, we've already detailed for product pages in general, and the same principles (a short answer up front, quantified evidence, visible freshness) apply to a vehicle spec sheet: see our guide on Schema.org for GEO and our article on what AI engines look for in a product page.

Electric vehicles and incentives: information that ages fast

Few sectors change their figures as often as buying an electric vehicle in France. The incentive scheme, renamed "coup de pouce véhicules électriques" (electric-vehicle boost), is now funded through energy-saving certificates (CEE) rather than directly from the state budget: amounts, which depend on a household's reference tax income, have become indicative and fluctuate with the CEE market. In 2026, the lowest-income households can get up to €5,700, a top-up of €1,200 to €2,000 exists for vehicles assembled in Europe with a European-made battery, and plug-in hybrids remain excluded from the scheme since July 2025 (Capcar, 2026 ecological bonus).

For a brand or a comparison site, this is exactly the kind of figure an AI can quote with an outdated number if the source page isn't kept current: a bonus amount shown without a date, copied from a 2024 or 2025 article, misleads a buyer before they ever reach a dealership. Freshness isn't a cosmetic criterion here; it determines the accuracy of what the AI repeats. We describe a comparable mechanism, under a different regulatory framework, for energy prices in our article on electricity suppliers facing AI.

→Energy: How AI Recommends an Electricity Supplier

What should you actually do?

1. List your buyers' real questions. "Best electric car for a €X budget", "[brand] reliability", "[model A] or [model B]": test these exact phrasings, not your lineup's marketing names.
2. Query several engines, not one. The Bubbling study shows ChatGPT and Gemini don't recommend the same brands on the same topic. A diagnosis limited to one engine hides part of the problem.
3. Date every figure that moves. Ecological bonus, warranties, pricing: put the date of the figure directly in the text, not only in a side table.
4. Structure your spec sheets. Use Schema.org's Car type properties for your specifications, without expecting an automatic effect on citations.
5. Track what's said about you off your own site. Specialist press road tests, forums, comparisons: these, not your product page, concentrate most of the citations.
6. Measure over time. An AI's recommendations about a car brand can shift with a model update or a widely-shared new road test: a one-off measurement isn't enough.

That's precisely what Beeleven's AI visibility audit measures — Beeleven being the agency behind Detekia: it questions ChatGPT, Gemini, Claude and Perplexity on your buyers' real questions, identifies which models and competitors are cited instead of yours, and weights the result by each AI's actual usage in France.

Frequently asked questions

Do ChatGPT and Gemini recommend the same electric cars?

No. According to the September 2026 Bubbling study, ChatGPT more often cites Tesla and Chinese brands like MG or BYD, while Gemini favors Renault and French brands. The study covers simulated journeys and doesn't measure real commercial impact (source).

Can you trust a reliability ranking an AI cites?

First check which ranking it's citing: Que Choisir, J.D. Power and Consumer Reports measure with different panels and methods, and don't always reach the same conclusions. An answer that doesn't name its source should be checked directly against the cited study.

Is Schema.org's Car markup enough to get cited more?

No. It's an aid to reading a spec sheet, not a decisive citation factor according to our calibration study. Google's "vehicle listing" rich result is also limited to verified dealerships in the US, inaccessible to a French site.

Why is it risky to cite the ecological bonus without a date?

Because it's no longer set by the state but funded through energy-saving certificates, which makes the amounts indicative and variable. An undated figure, repeated by an AI, can mislead a buyer about the aid they're actually entitled to.

Sources

• Journal de l'Automobile, "ChatGPT vs Gemini: which AI recommends your electric cars?", September 14, 2026.
• Autoactu.com, "What AI engines tell electric car buyers".
• En voiture Carine, "Electric car advice from AI: Renault, Tesla on top".
• Google-Kantar via Autocar Professional, "93% Of Car Buyers Use AI In Research", 2026.
• La Tribune Auto, covering the Que Choisir reliability survey no. 654, February 2026.
• Capcar, "Ecological bonus 2026: amounts, conditions and eligible vehicles".
• Schema.org, Car type.
• Google Search Central, "Vehicle listing (Car) structured data".

→How AI Engines Choose the Products They Recommend

→Competitors Cited by ChatGPT: How to Know Who's Taking Your Place