In August 2026, Bubbling asked ChatGPT and Gemini 8,064 questions across 2,016 complete conversations with simulated electric-car buyers. The result: conversational engines already have their champions, their brand narratives and their recommendation criteria, and none of them match registration figures, or what brands believe they are saying. Here are the six key findings from our study, published in an exclusive media partnership with Autoactu.com.
The 30-second version
- The podium depends on which engine you ask: Renault leads on Gemini (≈ 26% spontaneous share of voice), Tesla on ChatGPT (≈ 24%). There is no such thing as "the" AI ranking.
- AI recommends less and less as the purchase gets closer: 27% of answers contain an explicit recommendation at the discovery stage, 17% at the decision stage.
- Public incentives (60%) and price (56%) are the criteria most associated with the final recommendation: not technology, not ecology (present in only 15% of exchanges).
- Hyundai and Kia are introduced by the engines themselves in 99% of cases where they appear: nearly 29% of the engines' spontaneous initiative, far beyond their market share.
- Every brand is reduced to a signature phrase: the engines recycle it over and over, and some brands have none.
- 56% of conversations discuss financing: and not a single bank is ever named.
Which electric car do ChatGPT and Gemini recommend first?
It depends on the engine, and that is the first finding. Across the study's 1,008 spontaneous conversations (no brand named in the question), Tesla, Renault and Peugeot share the lead at around 21% of voice each. But this aggregate ranking hides two parallel markets.
| Gemini | ChatGPT | |
|---|---|---|
| Spontaneous #1 | Renault (≈ 26%) | Tesla (≈ 24%) |
| Answer style | web-cited, pedagogy first | synthetic, names brands upfront |
| Dominant sources | French: automobile-propre.com, caradisiac.com, carmaker sites, ecologie.gouv.fr | international editorial consensus |
| Brands favoured | Renault, Peugeot | Tesla, Volkswagen, BYD |
We call pre-priming the measurement of what an engine proposes on its own, before the buyer mentions any brand. Under pre-priming, the ranking shifts again: Renault and Hyundai tie for first place (≈ 16% each), and the Korean brands together account for nearly 29% of the engines' spontaneous initiative, a massive gap with their French registration figures. The engines do not recite the market: they recite the consensus of the international motoring press, with tests from ADAC and Consumer Reports among the most frequently echoed references in the corpus.
"A brand can be carried by the engines, carried by the buyer, or carried by no one. Knowing which side you are on is the first GEO decision to make."
At which stage of the purchase journey does AI recommend?
Less and less as the purchase approaches, the study's most counter-intuitive result. The explicit recommendation rate drops from roughly 27% of answers at the discovery stage to 17% at the decision stage, while conditional answers ("it depends on your usage") climb from 52% to 63%. We call this the inverted funnel: AI is a brilliant teacher and a poor salesperson. It explains superbly, compares willingly, then hands the decision back to the buyer at the very moment of closing.
The championship replays at every stage: Tesla dominates discovery (recommended in 5.7% of the stage's answers) then drops to 1.1% at decision, where Peugeot and Renault take the lead. The reason lies in the criteria: at decision time, public incentives are associated with 60% of recommendations and price with 56%. The engines know the French schemes in detail: the social leasing scheme appears in 719 of the 2,016 conversations, and ecologie.gouv.fr and service-public.fr rank among their most consulted sources. In France, documenting incentive eligibility is the number one lever of the final recommendation.
What do the engines say about each brand?
Every brand lives inside the engines through one dominant praise and one dominant criticism, recycled in a loop, what we call its signature phrase. A few measured examples.
| Brand | Signature phrase (praise) | Dominant criticism |
|---|---|---|
| Tesla | the Superchargers, a major asset | price, build quality, service |
| Peugeot (e-3008) | 701 km, the family-SUV benchmark | software maturity |
| Renault (R5) | acclaimed design, queen of social leasing | teething problems |
| Hyundai / Kia | reliability, a near-monopoly on the criterion | almost none |
| Volkswagen | no salient praise | software bugs |
Volkswagen illustrates what we call the extra risk: neither praise nor criticism, a brand confined to list mentions, its range reduced to the ID.4 in the engines' memory. These narratives enter through identifiable channels: criticism comes in through third-party test organisations, numbers through carmakers' official pages, which the engines re-quote with amounts, conditions and campaign dates. A carmaker's offers page has become a media channel in its own right.
How was the study conducted?
The study rests on 36 distinct initial questions, extended into complete four-exchange conversations by simulated buyers embodying three profiles (family, young urban professional, modest household eligible for France's social leasing scheme), replayed daily on ChatGPT and Gemini from 20 to 26 August 2026: 2,016 conversations and 8,064 questions in total.
Rankings are debiased through a four-level scale down to the pre-priming measure, uncertainties are handled through cluster resampling, and the limits are declared: the study measures the visibility and narratives produced by the engines on simulated journeys, not the preferences of real buyers. Roughly one French person in five already uses generative AI to research information before a purchase (Baromètre du numérique, Arcep/CREDOC).
FAQ
Which electric car does ChatGPT recommend most in 2026?
In our August 2026 corpus, Tesla holds the highest spontaneous share of voice on ChatGPT (≈ 24%), ahead of Renault and Peugeot. But at the end of the purchase journey, when the criteria shift to incentives and price, Peugeot and Renault take the lead in recommendations.
Do Gemini and ChatGPT recommend the same cars?
No. Gemini, drawing on mostly French sources, puts Renault first; ChatGPT, reflecting an international consensus, puts Tesla first. Any AI visibility measurement must be multi-engine.
What is GEO (Generative Engine Optimization)?
GEO is the practice of optimising a brand's visibility inside generative AI answers, the equivalent of SEO for conversational engines. Our GEO vs SEO article covers the difference.
How can I know what AI says about my brand?
By measuring it: presence, recommendations, narratives and criteria associated with your brand inside engine conversations, continuously and across several engines. That is what the Bubbling platform does.
The takeaway
Conversational engines are not waiting for brands: they already have a ranking (different on each engine), a disengagement point (the decision stage), king criteria (incentives and price) and a signature phrase per brand. All of it can be measured, this study is the first French baseline, and all of it can be worked on: the narratives have identified entry channels, the target gaps are quantified, the vacant territories such as financing are mapped.
The full study (22 pages: debiased rankings, recommendation funnel, narrative map, 11 brand scorecards, statistical annexes) is available from Bubbling, including a 90-minute debrief and the data tables. A dedicated brand audit starts at 1,500 euros excl. VAT per month.
A Bubbling study, exclusive media partnership with Autoactu.com. Data collected 20 to 26 August 2026; engines evolve without notice, results describe this period. Full methodology and limitations in the report.