Technical guide

How VIN Index actually works.

Written for the people who have to wire it up: website providers, feed vendors, dealer-group IT, and anyone doing technical diligence. If you are a dealer wondering what this does for your store, start with how it works instead.

Inventory feed requirements

We read one authorized inventory feed per rooftop. In almost every case it is the same export already going to the dealer’s website or marketplace vendors, so there is nothing new to produce.

Transport: SFTP is the default, pull preferred, so no push infrastructure has to be stood up on the provider side. HTTP pull is supported where a provider offers it.
Direct CRM sync: Dealers on BKD CRM can connect in one click. That mints a read-only inventory key against the CRM’s export rather than wiring a file transfer.
Format: Whatever the provider already emits. Each source runs through its own feed-format adapter, one parser per provider (EDealer, D2C, Convertus and so on), which maps the file into a single canonical vehicle schema.
Refresh: One-hour freshness target from feed change to live page. Unchanged payloads short-circuit before any downstream work.
Photos: Re-hosted at ingest and converted to WebP, so vehicle imagery is served from our own storage rather than hot-linked from the provider.
Scraping: None, ever. Inventory comes only from the feed the dealer authorizes. There is no crawl fallback and no scraped inventory path in the system.

The one place we read a dealer’s public website is at signup, to pre-fill dealership profile information such as hours, address, and logo. That path never touches inventory.

Data and validation sources

The feed is the primary source of truth for a vehicle. Decode sources normalize it and fill gaps. Nothing about a specific vehicle is invented.

VIN validation: NHTSA vPIC, the free US federal VIN decoder, for spec correctness on every VIN.
Gap fill: A commercial-grade VIN decode fills fields the feed omits. Results are cached, so the same VIN is never decoded twice needlessly.
Market context: Live market data across comparable listings in the region, used for factual positioning such as “priced below the regional median for this year, make, and model.”
Enrichment: One prompt-cached model call per vehicle produces the condition summary, grounded buyer answers, and illustrative financing scenarios.
Re-run policy: Enrichment is keyed on the VIN, a content hash of the payload, and the prompt version. It only re-runs when the description, options, or photos actually change. A price or mileage edit never triggers a rewrite.
Never generated: Per-vehicle accident history, ownership history, or service history. If the facts do not support a claim, the claim is omitted rather than filled in.

Market context is published when it strengthens the vehicle’s position. We do not publish negative deal ratings and we do not place a dealer’s vehicle into a public price-ranking system. Any financing figure shown is an illustrative sample calculation with disclaimers attached, never an advertised rate, term, or approval.

What we publish

Every validated vehicle is published as a retrievable document and as a queryable record. Both are designed on the assumption that the client reading them cannot execute JavaScript.

Listing pages: Server-rendered HTML with all indexable content in the initial response. Cached with tag invalidation on ingest and served with CDN cache headers.
Structured data: Vehicle JSON-LD in the initial HTML, plus dealer entity markup with stable identifiers, so an engine can resolve the vehicle and the seller as entities rather than as text.
Plain-text twin: A .md representation of each vehicle for retrieval clients that prefer text over markup.
Agent endpoint: A server MCP route over HTTP (Model Context Protocol) plus browser WebMCP tool declarations. Read tools include search_inventory and get_vehicle. Signed agents are verified with web-bot-auth.
Discovery files: Per-host sitemap.xml, robots.txt, and llms.txt. IndexNow pings fire on change, which reaches Bing, Yandex, Naver, Seznam and Yep. It does not reach Google or Brave, which discover on their own schedule.
Hosting model: Subdomain per dealer by default. Opt-in custom domain with automatic SSL. The canonical host is decided once per dealer at onboarding and never migrated, because a later switch redirects away accrued authority.

Sold vehicles wind down on a staged schedule rather than disappearing: the page stays live and marked sold, then leaves the sitemap and redirects to a lightweight record, then returns gone, then not found. A sold vehicle page is never redirected to the dealer homepage, which is the pattern that poisons crawl hygiene.

AI platform integrations

Listing pages and the agent endpoint wait to be read. Syndication does not: it formats inventory the way each shopping platform wants it and delivers it ahead of any search.

Microsoft: UCP JSON catalog feed for Copilot shopping.
OpenAI: JSONL product feed for ChatGPT shopping.
Google: Merchant-style product feed for AI shopping placements.
Refresh: Feed files regenerate roughly every four hours and are served from object storage under /feeds/*.
Availability: Several of these programs are still rolling out and each platform controls its own intake. We publish to each channel as it opens to feeds. We cannot guarantee acceptance or placement.
Plan gating: Syndication is a paid-tier capability. Free listings are published and discoverable through the network agent endpoint but are not syndicated outbound.

These catalogs are shared by every seller by construction, so syndication is the one surface where a dealer’s vehicle can appear near another dealer’s. Even there the result carries the dealer’s own link and attribution and the inquiry routes to them. The dealer’s own listing pages and agent endpoint carry no competing inventory at all.

CRM lead delivery and tracking

Leads are delivered to the dealer’s existing CRM with attribution metadata attached. We are not a lead broker and there is no per-lead charge at any tier.

Wire formats: ADF/XML over email, the format most dealer CRMs already ingest, or a signed JSON webhook. Free-tier dealers receive email delivery.
Attribution payload: Each lead carries the AI source and the vehicle it came from, so a lead can be traced back to the channel that produced it.
Delivery guarantees: Delivery is attempted inline and swept by a short-interval retry job. Duplicate submissions are suppressed within a 24-hour window.
Analytics axes: First-party server-side events on two axes, hit type by platform. A crawl, a human visit, and an agent tool call are never merged into one number.
Agent verification: Signed agent traffic is verified with web-bot-auth rather than inferred from user-agent strings alone.
Not reported: Any “estimated total AI traffic” figure. No defensible volume multiplier exists, so we report confirmed events only.

What we control, and what we do not

Worth stating plainly, because a lot of this category is sold on implication. The line below is the honest one.

VIN Index controls

How inventory is formatted and structured
Whether a machine can read it without running JavaScript
How fresh the published data is
That every vehicle has its own dealer-branded page
That feeds are delivered to each channel that accepts them
That visits, tool calls, and leads are attributed
That the lead routes to the dealer and nowhere else

The AI platforms control

Whether they crawl a given page, and how often
Whether they accept a syndication feed at all
Which vehicle they recommend to a shopper
Which source they cite in the answer
Ranking and placement, everywhere

Which is why the product promise is eligibility, not rank. We build inventory to be found and cited. Nobody who is honest can sell you the citation itself.

Questions about wiring up a feed?

Most integrations are a one-time credential handoff and take about an hour. If you manage feeds for multiple rooftops, we can wire them together.

Start an integration Read the FAQ