For website providers, marketplaces and inventory syndicators
Car buying is moving somewhere your platform does not publish yet.
You already have the two assets this requires: the dealer relationship and the inventory feed. VIN Index turns them into a channel the assistants can read and act on: sold under your name, reported through a rail your dealers can check, and running on infrastructure that already exists.
The shift
Discovery relocates about once a generation.
The internet moved car buying off the lot and onto the listing page. Nearly every company in automotive retail software, including yours, and including the marketplaces, was built in the decade that followed that one move. It is happening again, and the second one is further along than it looks.
The query stops being a filter set
A shopper who used to select a make, a model, a price ceiling and a radius now describes a situation: what they drive for work, what they owe on the trade, what their credit looks like, what they need it to do by Friday. The interpretation of that sentence happens before anyone reaches a search results page, and whatever does the interpreting decides what inventory is even eligible to be shown.
The result page collapses into an answer
Two hundred listings become three named sources. That is a different competition from the one this industry spent twenty years learning to win. Being in the index stops being the same thing as being in the answer, and the properties that decide which sources get named (structure, machine-readability, freshness, entity clarity, whether an assistant can query you directly) are not the properties anyone optimized for.
The buyer arrives already decided
The comparison now happens in a session nobody in the chain can see. By the time a shopper reaches a dealer, the shortlist is set. Every measurement system in this industry was built to watch a funnel whose first two steps have quietly moved somewhere else.
None of this means the listing model disappears this year. It means the layer above it is being assembled right now, and the sources it learns to trust are being chosen right now, by systems that are being trained on what exists today. Relocations like this settle once. The companies that were already standing there when it settled did not get there by being early to notice. They got there by being the easiest thing to read.
The question reaches your account manager first.
A dealer comes back from a twenty-minute conference session about AI search and asks the company that runs their website what is being done about it. That call is already happening. The only variable is whether the answer is a roadmap sentence or something the dealer can see working on their own inventory next week.
Website providers
You own the surface it has to run on
The pages, the feed plumbing and the dealer relationship are already yours. The work that makes inventory legible to an assistant has to happen on that infrastructure or not at all, which puts you first in line for it, and makes it awkward if someone else does it to your dealers.
Marketplaces
A product for the dealers you already have
This sells into your existing base, on inventory those dealers already carry, and it sits alongside what you sell today rather than in place of it. Nothing about it asks you to change what the marketplace does or where your traffic goes.
Inventory syndicators
Stop being a pipe
Distribution is becoming a commodity precisely as the value moves to how well the data is understood at the other end. The same feed, structured and measurable, is a different product from the same feed delivered on time.
There is a renewal argument underneath this that is worth being plain about. A platform that can show a dealer what the assistants did with their inventory last month is having a conversation about capability. A platform that cannot is having one about price. That gap widens every quarter this goes unanswered.
Already running
The unglamorous half is the half that takes the year.
Everything below is live infrastructure, not a roadmap. It is listed in detail on purpose: the pitch is not that this is clever, it is that it is finished, and that almost none of it is the part anyone expects to be hard when they scope it.
Feed ingest
One parser per provider format, normalized into a single canonical vehicle record. A content hash short-circuits unchanged rows, so polling a feed that has not moved costs nothing.
VIN resolution
NHTSA vPIC validation plus a commercial decode, cached. The feed stays the source of truth; the decode fills the holes it leaves, which is most of the trim and options data.
Server-rendered pages
Every vehicle page is complete HTML on first byte. A large share of AI crawlers never execute JavaScript, so a client-rendered VDP is, to them, an empty page.
Machine-readable twins
Vehicle JSON-LD, stable entity identity across surfaces, and a plain-text twin of every page for the retrieval paths that prefer text to markup.
A live agent endpoint
MCP over HTTP plus in-browser tool declarations, so an assistant can query inventory and act on it directly instead of scraping a page and guessing.
Outbound catalogs
The shopping-feed formats the AI surfaces actually ingest, generated and refreshed on a schedule rather than exported by hand when someone remembers.
Crawler identification
Signed-agent verification plus a user-agent list refreshed nightly. That list changes every month; a hardcoded one is wrong by the next quarter and silently under-reports.
Confirmed-event attribution
Crawls, human visits and agent tool calls are counted as three separate things and never blended. No estimated-total multiplier, because no defensible one exists.
Citation measurement
A standing evaluation that asks ChatGPT, Perplexity, Gemini, Claude and Google AI Mode the questions a shopper would, and records what came back.
Payment math and disclosure
Per-credit-band amortization with the disclosure rules attached. An advertised payment that is wrong is a regulatory event, not a bug, and it differs by market.
Cost control
Per-tenant budgets, a global spend ceiling and a circuit breaker that trips on its own. Anything that calls a model on a schedule needs this before it needs features.
Lead delivery
Standards-based lead formats and signed webhooks into the CRM the dealer already runs, carrying the attribution metadata with them.
Read that list as a scoping document rather than a feature list. Structured pages are a sprint. Knowing which crawler is real, what it took, whether an answer changed because of it, and what a payment figure is allowed to say in Ontario versus Texas: that is the year, and it is the part that has to keep working after someone ships it.
Building it once is the easy half.
The target moves every month
Engines change how they crawl, how they identify themselves, how they cite, and what they will and will not read. A crawler list is wrong within a quarter. A markup assumption is wrong within two. The durable asset is not the pages. It is the measurement rails that tell you what changed and what it did to your dealers, and those only become useful after they have been running for a while.
It is a measurement problem wearing a content problem's clothes
Most attempts start by generating better pages. That is the visible half. The half that decides whether any of it worked is the ability to separate a crawl from a human visit from an agent tool call, tie a lead back to an AI-mediated session, and ask the engines directly whether the answer changed. Without that, a team ships content and has no way to know it is losing.
A single platform only ever sees its own slice
What makes the measurement worth anything is watching the same behaviour across many dealers, many feed formats and many markets at once, including dealers whose sites are somebody else’s. A platform building this internally gets a view bounded by its own footprint, which is exactly the view that cannot tell you whether the number moved because of the work or because an engine changed something that week.
The compliance surface is not forgiving
The moment a page states a payment, it enters advertising law: trigger terms, disclosure, all-in pricing, per-market variation across the US and Canada. It is the least interesting work in the stack and the only piece where being wrong at scale is a legal problem rather than a bug report.
And it never actually ships
There is no version of this that is done. It is a permanent line on a roadmap competing with everything your dealers ask for by name, and it is a line that has to be staffed even in the quarters when it produces nothing visible.
You could build all of it. The question worth putting on the call is what that team would have shipped instead this year, and how much of the window closes while it is being built.
What you can show them
The numbers your dealers already grade you on.
Dealers judge a platform on a report. Right now the fastest-growing part of their discovery arrives in that report as direct traffic, or as nothing at all. The rail closes that gap with confirmed events, and confirmed events only.
Which assistants read the inventory, and when. Which vehicles they read. When an agent called a tool rather than loading a page. Which inquiries came out of an AI-mediated session, carried through to the CRM with the attribution attached. Three categories that most reporting blends into one, kept apart, because the moment they are blended the number stops being checkable.
What it deliberately does not include is an estimate of total AI traffic. There is no defensible multiplier for that, and a number a dealer cannot verify is a liability the first time they ask how it was calculated.
Practically, it becomes a section in the reporting your dealers already read, under your brand, next to the metrics they already argue with you about. It is the difference between telling a dealer you are doing something about AI and showing them the row.
Counted separately, always
- CrawlA verified AI crawler fetched the page. Identity checked, not assumed.
- VisitA person arrived from an AI surface. Referrer and session, not inference.
- Tool callAn assistant queried the endpoint directly. The strongest signal, and the one nothing else can see.
- LeadAn inquiry that carries its AI-mediated origin into the CRM.
Each one is a thing that happened, with a timestamp. That is the whole design constraint: everything on the report has to survive a dealer asking where it came from.
The commercial shape
Four ways it becomes a line on your invoice.
These are not alternatives so much as a starting position. Most partnerships end up as one primary model with a second layered on once the first cohort renews.
Model 01
Resell it as yours
You price it, you bill it, it appears on your invoice next to the website. We take a share; the majority stays with you, because you own the relationship that makes it sell.
Model 02
Uplift the plan they already have
No new SKU and no new sales motion. The capability becomes the reason the current tier holds its price, and the reporting section below is what makes that defensible in the room.
Model 03
Wholesale per rooftop
A flat rate per rooftop under whatever you charge for it, at any lot size. Your margin scales with the size of your base rather than with anyone’s inventory count.
Model 04
Price it on the outcome
For partners who would rather buy proof than a subscription: priced against attributed outcomes, measured on the same confirmed-event rail your dealers see. Never per lead.
Splits, floors and terms get set on the call, because they depend on how many rooftops come with you and who does the selling. Two things do not move: there are no per-lead charges in any model, and no dealer can pay to be ranked above another one. Those are the reasons the endpoint stays worth querying, which is the reason any of this is worth reselling.
Three moving parts on your side.
The integration is deliberately shallow. If standing this up required a platform migration, the argument on this page would not survive contact with your engineering roadmap.
Step 01
Send the feed you already send
Same file, same schedule, same transport. We write the adapter for your format once and it covers every dealer on your platform after that.
Step 02
We stand the surfaces up
Structured vehicle pages, the agent endpoint, the outbound catalogs and the measurement, under your dealers’ own brands and hosts, not ours.
Step 03
Events and leads come back
Into the CRM your dealers already run and the reporting they already read, with the attribution attached rather than reconstructed later.
The usual first step is not a contract. It is a handful of rooftops, one feed, a fixed window, and a report at the end showing what the engines actually did with them, which is also the fastest way to find out whether any of the above is true.
Schedule a partnership call.
Thirty minutes, with the person who built it, not a rep. Bring your feed format, a rough rooftop count and the commercial shape you would want, and you will leave knowing whether this is worth a pilot.
- What your dealers’ inventory looks like to an assistant today
- Which of the four commercial models fits your base
- What a first cohort would cost you to run, and what it would show
Would rather just email? trevor@vinindex.ai reaches us directly.
Request a call