Why Marketplaces Now Out-Cite Car Brands in AI Answers

An AI engine answering a question about which car to buy in Indonesia is now more likely to cite a marketplace listing than the car brand's own website. That is not a hunch. It is a measured, repeated finding from the only Indonesia-specific automotive AI-citation study run twice, a year apart, by the same team.

Maverick Indonesia, working with GridOto and Dataxet, published AI Visibility of the Indonesian Automotive Industry in 2025 and a follow-up, AI Citation Sources in Indonesia's Automotive Industry, in 2026. The two editions together span 2,100 AI-generated responses and more than 22,800 citations across ChatGPT, Perplexity and Google AI Overview. Read side by side, they document a genuine year-on-year shift, not a snapshot.

What the Study Actually Found

In the 2025 edition, news outlets such as Kompas.com, GridOto.com and Detik.com accounted for the largest single share of citations, with marketplace platforms close behind and official brand websites trailing at roughly 16.69%, cited mostly to confirm a warranty term or a safety spec rather than to be recommended outright. By the 2026 edition, that order had flipped. Marketplace platforms rose from 26% to 31.5% of all citations, overtaking news media, which slipped from 33.8% to 29.7%. The combined share going to earned sources, news plus marketplace, climbed from 86.2% to 90.7%. Brand-owned content is a small and shrinking slice of what an AI actually reads before it answers.

The single sharpest mover was YouTube, up from 0.75% to 9.53% of citations in one year, a twelvefold increase and the largest jump of any category tracked. A video review with an honest verdict and a clear specification rundown is no longer a nice-to-have distribution channel. It is a realistic, and growing, path into the answer itself.

Why This Is Happening, Not Just That It Is

The underlying mechanic is not unique to Indonesia or to cars. Aggarwal et al.'s foundational GEO study, presented at KDD 2024, found that generative engines favour sources with credible statistics, direct quotations and named citations over sources that merely rank well, and that keyword-stuffed or thin content shows zero or negative benefit. A marketplace listing is, almost by construction, a dense block of exactly the kind of structured, extractable fact an engine wants: price, mileage, location, seller, condition, all in one place. A brand's marketing page, by comparison, is often longer on persuasion than on the specific, sourced number an AI needs to complete an answer.

BrightEdge's analysis of automotive prompts found something that reinforces the same point from a different angle: AI systems recommend a car brand in 97% of responses even when the triggering prompt names no brand at all. The engine is not waiting to be asked "which Toyota model." It is inferring a shortlist from budget, seating and use case, and it needs a well-structured, verifiable source to pull that shortlist from. Increasingly, in Indonesia, that source is a marketplace page or a video, not a manufacturer's brochure copy.

Why Manufacturer Sites Still Matter, Just Differently

None of this makes brand-owned content pointless. It changes its job. At roughly 16.69% of citations in the 2025 baseline, official sites were cited overwhelmingly as verification sources, the place an engine checks a warranty term or a safety rating it already found elsewhere. That is a real function. An engine that cannot verify a fact against an authoritative source is less confident citing it at all, from any source. Losing that verification layer would not just cost the brand's own citations. It would likely reduce the accuracy, and possibly the frequency, of citations to the marketplace and video sources that now carry the volume.

The practical shift is in where the effort goes. A manufacturer or dealer optimising only their own domain is optimising the smallest and slowest-growing layer of the category's citation mix. The larger, faster-growing opportunity sits in making sure the marketplace listings and video content already citing your models are accurate, current and well-structured, because that is where an increasing majority of the answer is actually being assembled.

A Three-Layer Governance Problem, Not a Content Problem

This reframes automotive GEO as a governance question before it is a writing question. A manufacturer's official page, a dealer's local listing, and a marketplace entry for the same model need to agree on price, variant and specification, because generative engines cross-reference across the web to resolve inconsistencies, and a mismatch between any two of the three tends to reduce confidence in all three rather than simply picking a winner.

"Zero-Click Influence": Why the Battle Is Won Before the Search Even Feels Like Shopping

The Maverick Indonesia research team has a specific name for what happens when a household types "mobil keluarga terbaik di bawah 300 juta rupiah" into an AI: zero-click influence. The AI hands back a synthesized, confidence-projecting shortlist, and that shortlist pre-screens the brand consideration set before the household has consciously begun comparing anything. No dealership visit happened. No marketplace listing was clicked. The competitive outcome was already substantially decided in that one exchange.

Indonesia's underlying AI adoption rates make this a structurally important moment for automotive specifically, not a fringe behaviour. Indonesia ranks first globally in Statista's measure of AI adoption for daily activities, at 60% of respondents, ahead of Saudi Arabia at 56%, Malaysia at 55% and India at 52%. Katadata Insight Center's 2024 national survey found that 65% of Indonesian internet users had already used AI tools, with 81% of that group using AI primarily to search for information, and Snapcart separately found that 71% of Indonesian AI users default specifically to ChatGPT. A 2026 study by researcher Dedy Budiman at Universitas Prasetiya Mulya reported that 74.6% of Indonesian consumers now use AI for product research, with 52.3% saying AI brand mentions influence their purchase consideration; that particular figure comes from a single, self-published academic study rather than a peer-reviewed, repeated one, and is worth treating as suggestive rather than definitive until corroborated elsewhere.

Ekho's behavioural research adds a comparison that puts the shift in perspective: 30% of in-market vehicle shoppers globally used an AI-powered tool during their buying journey, more than double the 12.7% who used online marketplaces, and ahead of the 15.8% who visited an OEM or dealer site directly. AI was not a niche supplement to the traditional research path. For a meaningful share of shoppers, it was the primary front door, with the dealer website relegated to the point of transaction execution rather than discovery.

This Is Not Only an Indonesian Story, Which Is Exactly Why It Matters

None of the citation-share findings above happen in isolation. A cluster of independently-run, named US studies through late 2025 and early 2026 confirms the same underlying behavioural shift is happening at scale in a completely different market, which strengthens rather than weakens the case that Indonesia's numbers reflect a real structural change and not a local statistical artefact. Cars.com's AI in Car Shopping Consumer Survey, fielded 4 to 10 November 2025 across 936 respondents, found 44% had used AI-powered search tools while shopping for a vehicle, and 97% of those users said the technology would influence their future purchase decisions; 73% called AI a time-saver, and 59% treated it as a starting point for research rather than a final answer. CarGurus' eighth annual Consumer Insights Report, covering more than 3,000 recent buyers and sellers and released 3 December 2025, found 80% open to using AI and 26% already doing so, with 53% of consumers now considering three or more brands, up from 43% the year before, comparing vehicles and finding listings as the two leading AI use cases.

Cox Automotive's sixteenth annual Car Buyer Journey Study, fielded fall 2025 across 2,300 recent buyers, found 19% of all buyers and 25% of new-vehicle buyers had used an AI website or AI-generated overview, with AI users reporting 84% overall shopping satisfaction against 71% for non-users, a genuinely large gap. CarEdge's own 500-shopper 2025 survey landed in a similar range: one in four buyers already using AI tools, 88% of them finding it helpful, and 40% of remaining shoppers planning to adopt AI for their next purchase. None of these figures should be imported directly into an Indonesian forecast, since consumer behaviour, platform preference and market structure differ meaningfully between the two countries. What they do establish is that the shift documented locally by Maverick Indonesia is part of a genuine, multi-market pattern, not a one-off finding specific to a single study design.

Where in the Page an AI Overview Actually Pulls Its Citation From

A separate, detailed US study adds a layer the Indonesia-specific research does not cover: which part of a page actually earns the citation once an AI Overview decides to reference a dealership at all. C-4 Analytics tracked 151 US dealership domains across ten website platforms and 33 states in July 2025 and found 85% of domains earned at least one AI Overview citation, with 76.59% of the triggering queries being purely informational rather than transactional. Within those citations, the specific page type mattered enormously: general informational landing pages accounted for 37.86% of citations, model and trim pages 20.58%, service and parts pages 12.82%, model comparison pages 8.22%, and vehicle detail or listing pages only 7.63%. Service and parts content outperforming model comparison content in this breakdown is a striking, underappreciated result, and it directly reinforces the case, covered in depth elsewhere in this series, that fixed-ops content is both underserved and disproportionately citable relative to how little competitive attention it receives.

The Four Phases of AI-Assisted Car Shopping, and Where Citations Cluster

Ekho's behavioural research into AI-assisted vehicle shopping identified four distinct phases a buyer moves through: open-ended exploration, deep model-specific research, pricing and reliability validation, and local dealer discovery. That sequence matters for citation strategy because each phase pulls from a different layer of the citation mix documented above, and treating "automotive GEO" as one undifferentiated effort misses where the real advantage actually sits.

Open-ended exploration, the "what's a good family car under a certain budget" stage, is exactly where the 97% brand-recommendation-without-a-brand-name finding applies, and it is disproportionately answered from marketplace and editorial sources with broad category coverage, not a single manufacturer's page. Deep model-specific research leans harder on comparison content and increasingly on video, which is consistent with YouTube's twelvefold citation-share jump. Pricing and reliability validation is where an official brand page earns its keep as a verification source, the 16.69% layer confirming a number the buyer already found elsewhere. Local dealer discovery is the odd one out: a separate, US-based study of 151 dealership domains found that only 0.78% of AI-Overview-triggering automotive queries were primarily local or geotargeted, with 76.6% informational instead. Local, "near me" intent still routes mostly through the traditional Map Pack and Google Business Profile, not the generative answer.

The practical takeaway is that a single citation-share number, averaged across all four phases, tells you less than breaking it down by phase. A brand doing well on pricing-validation citations but invisible during open-ended exploration has a different problem than a brand well-cited early but never confirmed later. The Maverick Indonesia figures describe the aggregate; a brand's own prompt panel needs to test each phase separately to know where its specific gap sits.

What This Means If You Only Have Budget for One Thing

If a manufacturer or dealer can fund exactly one automotive GEO action this quarter, auditing marketplace listing accuracy against official data outranks writing new brand-owned content. The citation math has already moved past owned content as the primary lever, and that shift shows no sign of reversing between the two editions of the same study. Owned content still earns its place as the verification layer an engine checks once it has already found a marketplace or video source worth citing. It is no longer the layer doing most of the work.

Frequently Asked Questions

Does this mean manufacturer websites are becoming irrelevant?

No. They are becoming a smaller share of a larger citation pool, with a different job: verification rather than primary recommendation. An engine that cannot confirm a marketplace-sourced fact against an official page is less likely to cite it confidently at all.

Is the Maverick Indonesia study the only source of this kind of data?

It is the only named, repeated, Indonesia-specific automotive AI-citation study we found running the same methodology across two consecutive years. Most other automotive GEO citation research is US-focused and should be treated as directional for an Indonesian market, not applied directly.

Should a dealer network stop investing in its own website?

No. A dealer's own site is still the authoritative source for local inventory, current price and test-drive booking, none of which a marketplace listing captures reliably. The point is sequencing: verify and correct the higher-citation-volume layers before assuming a new owned page will move the needle on its own.

How often should the citation mix be re-checked?

Quarterly is a reasonable floor, given that the tracked shift between the 2025 and 2026 editions of the same study was large enough to reorder the top citation source entirely. A category moving this fast does not reward a set-and-forget audit cadence.

For a deeper look at how citation share and recommendation share should be tracked as separate metrics, see our piece on GEO for Automotive, and for the fuller platform-by-platform citation mechanics behind this shift, Tessar Napitupulu's Cited or Silent covers the per-engine playbook this data feeds into.

Want the full RoGEO-based citation audit for your own models? Download the first chapters of Cited or Silent free, or see how this connects to the rest of the stack on our SEO for Automotive page.