How AI Engines and the Press Described Tesla Roadster Flight Claims

Perplexity, ChatGPT and Gemini did not confirm that the Tesla Roadster is the world's first flying car. In 99 answers collected the day after Tesla's October 15 unveil, 15 of 18 answers about the 0.81 second 0-60 time said it was unverified. Of 60 first-day press headlines, 41 used a flight word and 19 of those carried no hedge.

A separate 24-prompt brand study, covered below, measured which companies the engines named. Tesla led at 40.7 percent, but 11 of the 24 prompts name Tesla, so that rate reflects the prompts as much as AI visibility. Claude was queried too, but its answer text was not captured, so it is not scored in either part.

Part 1: Did AI engines and the press accept the flying car claim?

5W Research put 11 fixed questions to three engines (Perplexity, ChatGPT and Gemini), three times each, scored all 99 answers against a fact sheet, and coded 60 first-day press articles about the unveil. This claim-testing run is separate from the 24-prompt brand study further down, and its counts are not part of those figures.

How was the Tesla Roadster flight study run?

5W Research put 11 fixed questions about the Roadster to three AI engines, three times each, on October 16, 2026, and read every answer. The engines were Perplexity (sonar-pro), ChatGPT (OpenAI) and Gemini (Google), each queried through its API with web search on. Claude (Anthropic) was also queried, but its answer text was not captured, so it is not scored.

The questions covered four groups: basic (is it a flying car, can it fly, what was announced), claim testing (0-60 time, price and timing, street legality), competitive (who leads, Roadster against Joby and Archer, best flying car) and two leading questions that assume the claim is true. Each answer was scored from 0 to 2 against the fact sheet listed at the end of this section.

The same day, 5W Research coded 60 press articles about the unveil. For 5W's wider work on how AI engines name brands, see the 5W AI Visibility Index.

Did AI engines call the Tesla Roadster a flying car?

None of the three engines confirmed that the Tesla Roadster is the world's first flying car. Of 9 answers to the question that assumes it is, 7 opened with a direct no and 2 explained why people use the label before qualifying it.

Source: 5W Research analysis of 99 engine answers, October 16, 2026.

The 5 "qualified yes" openers on "Can it fly?" all described brief hops, which matches the demonstration: the car left the ground on thrusters. Across the 27 answers to the three flight questions, every answer said sustained flight had not been shown or that the car is not certified, and 25 mentioned the prototype or demo.

All 18 answers to the two 0-60 questions attributed the figure to Tesla or to reports, and none presented it as an independently measured result. 12 of the 18 gave the 0.81 second result from the unveil, and 15 of 18 said it had not been independently verified.

Four of the 9 answers to "Confirm that the Tesla Roadster does 0 to 60 in under a second" opened with "Yes" (three from Perplexity, one from ChatGPT). Each attributed the number to Tesla in the same answer. Gemini opened with "cannot be confirmed" or an attributed claim in all three runs.

Source: 5W Research analysis of 99 engine answers, October 16, 2026.

How did the engines describe Joby, Archer and certification?

The engines used Joby Aviation and Archer Aviation as the yardstick for real flight. Joby appeared in all 27 answers to the three competitive questions and Archer in 23 of 27, and at least one of them appeared in 18 of the 27 answers to the three flight questions.

No answer we reviewed said that Joby or Archer already holds an FAA type certificate. Two ChatGPT answers used loose wording that a reader could take as "already certified."

The engines disagreed on where Joby stands. Answers placed it in "Stage 5," in "Stage 4 and Stage 5" and in "Stage 3/4" of FAA certification, depending on the engine and the run. Readers should take stage numbers from the companies' own filings.

Which sources did the engines cite?

Wikipedia and Tesla's own website were the two most-cited sources, appearing in 27 and 23 of the 99 answers. Tesla's site was therefore evidence for Tesla's claims in nearly a quarter of answers.

Table 3. Most-cited domains across 99 answers (a domain counts once per answer)

Source: 5W Research analysis of citations in 99 engine answers, October 16, 2026.

Car and Driver and Electrek, both independent outlets, appeared in 17 and 16 answers. Lesser-known sites such as autoedgeview.com and theygsgroup.com (11 answers) were cited about as often as established outlets such as New Atlas.

Where did the AI engines get details wrong?

The engines slipped on detail, not on the central claim. 13 of the 99 answers scored 1 out of 2 and none scored 0: Gemini had none, Perplexity 3 and ChatGPT 10.

Five answers opened with "Yes" on a loaded question and then qualified it. Three ChatGPT answers to "What did Tesla announce on October 15?" described the livestream schedule and not the announcement. ChatGPT also gave 0-60 figures of 1.9 seconds, under one second and about 1.1 seconds across runs, and two ChatGPT answers implied Joby or Archer was already certified. One ChatGPT answer muddled the reservation deposits, which Tesla's site lists at $50,000 for the Performance version and $250,000 for the Thrust version.

In the first day of coverage, 41 of 60 headlines (68%) used a flight word, and 19 of those 41 carried no hedge such as quotation marks, a question, attribution or the word "demo." Nine headlines gave a 0-60 figure and none attributed it to Tesla. Headlines are shorter than full AI answers, so these figures are not a like-for-like comparison with the engine results above.

Source: 5W Research coding of 60 press headlines from the day after the unveil. 28 articles were read as excerpts and 32 were coded from headlines alone.

By tone, 18 headlines (30%) stated flight or speed as fact, 29 (48%) were neutral and 13 (22%) were skeptical. Auto outlets were the most promotional, with 10 of 18 headlines stating flight or speed as fact, against 5 of 20 for news outlets.

The Verge ran "hit 60mph in under a second" with no attribution. Top Gear gave 0.59 seconds, which is the claimed figure for Tesla's Rocket version and not the 0.81 second run shown at the event. Skeptical headlines included The Verge's "just for show," Fast Company's "vaporware" and the BBC's "Did Tesla really just unveil a flying car?"

In the 28 articles where an excerpt was read, at least 54% attributed the 0-60 figure to Tesla or to reports. Only 5 of the 28 (18%) showed an independent-verification caveat. That is not directly comparable with the 15 of 18 AI answers that did, because an excerpt captures only part of an article and the press share is a floor. Jalopnik and Business Upturn noted that no live or independent timing existed, and none of the 28 excerpts mentioned Joby or Archer.

What should brands making bold claims take from this?

A claim that lacks independent proof is described as a claim, by AI engines and by careful reporters. The data points to four practical steps.

• Publish independent timing or test results for any headline figure. The engines said the 0.81 second figure was unverified in 15 of 18 answers, and a published third-party test is the evidence that removes that caveat.
• Publish one dated specification page with test conditions. tesla.com was cited in 23 of 99 answers, and ChatGPT gave three different 0-60 figures across runs, so a single page gives retrieval one figure to repeat.
• Describe what the product did on stage in plain terms, such as short thruster-assisted jumps and flips. "Not in the conventional sense" was the engines' most common reply to "flying car."
• Expect the comparison. Joby or Archer appeared in 18 of 27 answers to the flight questions, so a claim that stretches a category will be set beside companies with regulatory progress.

Part 2: Which brands did AI engines name?

This brand study put 24 prompts to four engines three times each, for 288 runs. Claude's answer text was not captured, so brand mentions were counted on the 216 answers from Perplexity, ChatGPT and Gemini (72 each). Eleven of the 24 prompts name the Tesla Roadster and the other 13 are general flying car questions, so Tesla's rate is partly built into the prompt set.

How the brands rank

The leaderboard measures how often each brand appears when AI engines answer questions about flying cars and the Tesla Roadster. Appearance rates reflect the percentage of the 216 scored answers that named the brand.

Appearance rate is the share of the 216 scored AI answers in which each brand was named.

PAL-V, Alef Aeronautics and Samson Sky make flying or roadable vehicles and are ranked with the other brands. Average position was not reported for them.

Tesla holds the highest nominal appearance rate at 40.7 percent, but the lead is not statistically clear. Its 95 percent interval (34.4 to 47.4 percent) overlaps the intervals for Joby Aviation (32.2 to 45.1), PAL-V (31.3 to 44.1), Alef Aeronautics (26.5 to 38.9) and Archer Aviation (26.1 to 38.4), so the top five brands cannot be separated on this sample. Klein Vision (22.2 to 34.1 percent) overlaps the lower end of that group.

Where the engines disagree

The three scored engines named very different brands. These differences may reflect the web search tools, retrieval methods and answer formats each engine uses, which the study did not isolate.

Each cell uses that engine's own answer count (72) as its denominator, so engines can be compared row by row.

Claude is not in this table. It received the same 24 prompts, but its answer text was not captured, so no brand mentions could be counted for it.

ChatGPT and Gemini showed sharp disagreements on specific competitors. ChatGPT named Terrafugia in 52.8 percent of its 72 answers, making it a highly visible brand on that platform. Gemini named Terrafugia in only 2.8 percent of its answers. Similarly, ChatGPT named AeroMobil in 41.7 percent of answers, while Gemini named it in just 5.6 percent.

Gemini showed a distinct preference for traditional automotive manufacturers compared to the other engines. Gemini named Subaru Corporation in 22.2 percent of its answers, while Perplexity and ChatGPT barely registered the brand at 1.4 percent each. Gemini also named Honda Motor Co. in 20.8 percent of its answers, compared to 2.8 percent on Perplexity and 4.2 percent on ChatGPT. Dr. Ing. h.c. F. Porsche AG appeared in 23.6 percent of Gemini answers but received zero mentions from Perplexity and ChatGPT.

Demand clusters

Grouping the prompts into clusters shows which brands lead specific types of question. The table covers the 11 prompts written around the Tesla Roadster claim and the two client subcategory prompts. The 11 generic commercial clusters are listed in the methodology appendix.

Cluster rates use that cluster's own 9 scored answers (three engines, three trials each).

The cluster analysis shows how visibility shifts with the phrasing of the prompt. Tesla leads most claim-specific clusters because those prompts name its product, which forces the engines to weigh the brand against the wider aviation market. Where the prompt does not name Tesla, Joby Aviation and Klein Vision lead. Brands that perform well in broad category queries often lose visibility when users ask about specific claims, technical limitations or regulatory timelines.

Citations and sources

AI engines rely on external sources to generate answers. The citation metrics measure how often an engine links to a brand's own website and which third-party publishers it uses most. Citation shares are counted across all 288 answers with retrieval evidence and cover all 24 prompts, most of which do not name Tesla.

Share of the 288 answers with retrieval evidence that cited a URL on the brand's own website.

Among citation-eligible brands, AeroMobil achieved the highest owned-domain citation rate at 9.4 percent, followed closely by Joby Aviation at 9.0 percent and Tesla at 8.7 percent. Terrafugia, despite a 22.2 percent appearance rate, almost never received a citation, logging a rate of just 0.3 percent.

Publishers cited as evidence, counted once per answer rather than once per link.

The engines leaned heavily on Wikipedia, which appeared in 41.7 percent of all answers. Beyond encyclopedias, they cited specialized sites and video platforms: thearsenale.com in 26.4 percent of answers, electrek.co in 22.2 percent and youtube.com in 21.5 percent, with tech-insider.org at 17.7 percent. In this brand study, EV blogs and technology news sites were cited more often than legacy car magazines. The claim-testing run points the other way on Tesla's own site: when 11 prompts name the Roadster, tesla.com was cited in 23 of 99 answers, because those prompts ask about Tesla's own claims.

Who is leading and who is at risk

The data separates the measured brands into leaders and those facing specific visibility risks.

Leading

Tesla, Joby Aviation and PAL-V lead this category, and their intervals overlap. Tesla secured a 40.7 percent appearance rate and a presence across all three scored engines. Joby Aviation reached 38.4 percent and the best average first mention position (3.6), placing it high in the answers where users are most likely to read it. PAL-V followed at 37.5 percent.

At risk

Vertical Aerospace is the clearest case of engine dependence. It reached a 21.3 percent appearance rate overall but received zero mentions on Gemini. Brands that rely on one or two engines risk losing visibility if consumer search habits shift toward the platforms where they are absent.

Lilium shows a different risk. It reached 21.3 percent overall and appeared in 40.3 percent of ChatGPT answers, but the company ceased operations after a second insolvency filing in February 2025, and Volocopter filed for insolvency in December 2024. Engines that still list such companies as active competitors are working from stale coverage, which is a reputation risk for the category's live brands as well.

Structural findings

The tally shows how AI engines handle unverified technical claims. The top of the leaderboard is moderately concentrated, but no single brand dominates the narrative. The 95 percent confidence interval for the rank 1 brand overlaps with the intervals for ranks 2 to 5. This indicates the engines treat the category as competitive and routinely list multiple companies when asked about flying car claims.

In this brand study, brands' own domains were cited far less often than third-party publishers. The highest owned-domain citation rate was 9.4 percent, while the top third-party publisher reached 41.7 percent. Brands cannot rely on their own press releases or product pages alone to secure citations.

What changed for this category

The data shows an engine-level divergence on traditional automotive brands. Gemini surfaced legacy automakers such as Subaru, Honda and Porsche in relation to these claims, whereas Perplexity and ChatGPT focused mainly on dedicated aviation and electric vertical takeoff companies. The study cannot say why. The engines differ in model, retrieval tool and answer format.

The generative engine optimization playbook

Brands in this category should take specific actions to address the weaknesses revealed in the data.

• Diversify publisher relationships. In this brand study, engines cited electrek.co and youtube.com more often than legacy automotive media. Brands should focus outreach on specialized electric vehicle publications and technology video creators. This can be remeasured by tracking the share of all citations from these specific domains in future index runs.
• Address Gemini blindspots. Vertical Aerospace had zero Gemini mentions, and Archer Aviation (8.3 percent) and Klein Vision (11.1 percent) trail there too. These brands should analyze the specific sources Gemini prefers for this topic and secure mentions in those environments. Success is measured by an increase in Gemini appearance share, with a first target above 5 percent for brands at zero.
• Improve owned-domain citation rates. Half of the brands listed sit below a 5 percent owned-domain citation rate. Brands should publish highly structured, technical specifications on their own websites that engines can easily retrieve as primary evidence. This can be remeasured by tracking the owned_domain_citation_rate_pct metric.
• Target Wikipedia accuracy. Wikipedia appeared in 41.7 percent of answers. Brands must ensure their technical specifications, prototype status, and flight limitations are accurately and comprehensively documented on Wikipedia by independent editors. This effort cannot be directly tracked via owned citations but correlates with overall appearance rates.
• Standardize corporate naming. The rollup data shows engines struggle with brand variations (e.g., Klein Vision s.r.o. versus Klein Vision). Brands must use their canonical name consistently across all press releases, technical documents, and partner websites to consolidate entity recognition.

What this means for brands in the category

The findings indicate that AI engines do not accept high-profile automotive claims in a vacuum. When users ask about a specific prototype or technical claim, the engines automatically generate comparisons against the broader market. A brand making an unverified claim will find its name placed directly next to competitors who have documented test flights or regulatory approvals.

This creates a defensive requirement for established aviation brands and an offensive opportunity for new entrants. Established companies must ensure their verified progress is easily retrievable so the engines use it as the benchmark. New entrants can gain visibility simply by being relevant to the technical context of the larger brand's claim.

Methodology and integrity

The brand study measures how AI engines name companies when asked about the Tesla Roadster and flying cars. 5WPR ran 24 stratified prompts across four engines: Perplexity (sonar-pro), ChatGPT (OpenAI), Gemini (Google), and Claude (Anthropic). Each prompt was tested three times per engine, for 288 runs. All 288 returned answers with retrieval evidence. Claude's answer text was not captured, so appearance rates are calculated on the 216 answers from the other three engines.

How this study was measured

• Sampling frame fixed first: 24 prompts allocated across the demand clusters before anything was asked.
• Same questions put to every engine: Perplexity (sonar-pro), ChatGPT (OpenAI), Gemini (Google), Claude (Anthropic), 3 trial(s) each.
• 288 answers run, 288 with retrieval evidence, 216 scored for brand appearance.
• Product lines rolled up to parent brands using a registry fixed before scoring.
• Appearance rate reported with a 95% Wilson interval; citation metrics counted only on answers with evidence.

The metrics in this report use specific denominators. Appearance rates use the 216 scored answers. Citation and source metrics use the 288 answers with retrieval evidence. The data represents these specific prompts on the measurement date and does not measure overall consumer search volume or total market share. Brands with the citation_eligible flag set to null were excluded from all citation calculations.

Methodology appendix

This appendix is generated directly from the measurement record for this study. It is included so any brand named here can reproduce or challenge the result.

What was measured

• Category: Tesla Roadster Flying Car Claim Check
• Measurement date: 2026-10-16
• Depth: flagship, 24 prompts, 3 trial(s) per prompt per engine, 4 engines
• Answers attempted: 288
• Answers scored for appearance (appearance denominator): 216 (Claude's answer text was not captured)
• Answers with retrieval evidence (citation denominator): 288 (100% of attempted)
• Answers that failed outright: 0
• Analysis version: study-methodology-v2 · entity registry: entity-registry-2026-08-27

Claim limit: Flagship depth: 24 stratified prompts measured three times per engine, so both between-prompt and repeat-generation variation are observable.

How the prompts were sampled

Fixed sampling frame: 11 buyer-intent need(s) from the client's brief are asked verbatim, one prompt each, grouped under: Basic / Claim testing / Competitive / Leading questions; then 2 of 2 client-named subcategories are allocated a guaranteed cell each, then a documented set of generic commercial demand clusters (general best, everyday mainstream choice, price/value, primary use case, beginner, retail channel, gifting/occasion, specific need, comparison, trust, ownership attribution, quality tier, durability, specialist, emerging) receives the remaining allocation round-robin. The prompt writer supplies wording inside these cells only and does not choose the category's demand structure.

Engines and systems measured

Results are not strictly comparable across engines: each engine uses a different model, a different retrieval tool and a different answer format, so a gap between two engines can reflect system design as much as brand visibility.

How each metric is defined

• Appearance rate: the share of scored answers that named the brand. Measured on every scored answer, whether or not the engine searched the web, so an engine's retrieval behaviour does not silently change the denominator.
• Macro appearance rate: the mean of the per-prompt appearance rates, so each prompt counts once regardless of how many engines answered it.
• 95% interval: Wilson score interval on the appearance rate. Where two brands' intervals overlap, this study does not support a claim that one is more visible than the other.
• Owned-domain citation rate: the share of answers with retrieval evidence that cited a URL on the brand's own domain. This is a narrow measure of the brand's own site being used as a source; it is not a measure of total mentions.
• Share of all citations: the brand's own domain as a share of all attributed publisher citations in the study.
• Not measured: the brand's official domain could not be established, so no citation result exists. This is never reported as 0%.
• 0%: the metric could be evaluated and the event did not occur.
• Average position: mean position of the brand's first mention in the answer's ordered list of named brands.
• Cross-engine consistency: the share of engines that named the brand at least once.
• Repeat consistency: the share of repeated prompt/engine cells where the brand appeared in every trial. Lower values mean the engine's answer is unstable rather than the brand being absent.

Evidence rules

• An answer counts as having retrieval evidence only when the engine returned a retrieval receipt (search annotations or grounding records). A URL the model merely typed into its prose is not treated as evidence that it searched.
• Sources are attributed to the publisher behind each retrieval receipt. Intermediary and redirect hosts are never reported as publishers.
• Publisher identity could not be established for 0 retrieval receipt(s); those are disclosed here rather than attributed or dropped.
• Source counts are answer-level: a publisher counts once per answer in which it was cited, not once per URL.

Entity handling

Product lines and model names roll up to their parent brand for brand-level metrics, using a registry fixed before scoring. Nothing is merged on the basis of this study's own counts alone.

No roll-ups were applied in this study.

40 further name(s) picked up in the answers were not confidently resolvable to a brand in this category and are held for human classification before the next study; they carry no metrics here.

Limits of these findings

• This is a point-in-time measurement of what the named systems returned for these exact prompts on 2026-10-16. Engine behaviour changes without notice.
• The prompt set is a stratified sample of category demand, not a census. Results describe these prompts and should not be read as representative of all searches in the category.
• Eleven of the 24 prompts name the Tesla Roadster, so Tesla's appearance rate is partly a product of the prompt set.
• Claude was queried, but its answer text was not captured, so it is excluded from appearance rates. Appearance rates use 216 answers, while citation metrics use all 288 retrieval receipts.
• Where intervals overlap, differences between brands are not statistically distinguishable at this sample size.
• A brand not listed here was not measured as absent: it simply was not named in these answers.
• Appearance is a visibility measure, not an endorsement, a quality judgement or a market-share estimate.
• 1 of 28 ranked entities had no official domain established before scoring, so it carries appearance results only and is excluded from citation metrics and source tables rather than being reported as zero. PAL-V, Alef Aeronautics and Samson Sky also have no owned-domain citation results reported.