Your inventory feed needs a ten-VIN test
A practical check for dealers evaluating whether an AI shopping connection can answer from the cars they actually have.
Alex YankouskiAuthor · AI-assisted preparation
A shopper asks for a used SUV below a price limit, with a particular feature, near a particular town. The dealership has a suitable car. The assistant does not return it.
Where did the opportunity disappear?
Possibly the feed did not include that rooftop. Possibly the feature arrived as free text that the connection could not interpret. Possibly the location was missing. Or the assistant never used the inventory connection at all.
“We need more AI visibility” is too broad to diagnose any of those problems. A small, repeatable inventory test gives the dealership and its vendors something concrete to work on.
Choose ten vehicles that can expose a mistake
Ten is a manageable starting sample, not an industry benchmark or a statistically representative study. The aim is to find defects before expanding distribution.
Select examples that exercise different parts of your inventory:
- An ordinary, complete listing that should be easy to find.
- A vehicle with a recent price change.
- A vehicle with equipment buyers regularly request.
- A new arrival and a recently sold unit.
- Two similar vehicles at different rooftops, if you operate a group.
- A record with a known missing field.
- Two further examples from different body styles or feed paths.
The sold unit is a negative test. Its job is to demonstrate that stale stock is removed or handled accurately. The incomplete listing tests whether the system admits what it does not know.
For each example, record the VIN or stock identifier, store, source-system status, expected price, equipment evidence and time checked. Use your authorized operational records as the reference. A stale website page is not a reliable substitute for the source of truth.
Check three separate surfaces
First, inspect the incoming inventory. Did the expected record reach the connection, with the correct rooftop and fields?
Second, test the connection directly. Can it retrieve the expected vehicle and return the facts supplied by the dealer?
Third, test the shopper experience in the assistant. Does a suitable request cause the assistant to use the connection, show the relevant result and explain missing information appropriately?
These stages answer different questions. A working inventory response does not prove that an assistant will choose to call it. An assistant mentioning your business does not prove that it has current stock. A search-engine citation is a fourth surface and should have its own measurement.
Record which stage failed. Otherwise the feed provider, website provider and AI vendor can all be discussing different problems while looking at the same screenshot.
Use a worksheet with a place for failure
| Test | Evidence to record | What should trigger investigation |
|---|---|---|
| Identity | VIN/stock and rooftop match | Another vehicle or store appears |
| Price | Source value, returned value, timestamps | Unexplained mismatch or lost price conditions |
| Availability | Active/sold status and update times | A sold unit is presented as available |
| Location | Store address and location used in search | Wrong rooftop or an unexplained distance result |
| Equipment | Requested feature and source evidence | Missing evidence becomes a confident claim |
| Retrieval | Request, returned identifiers and outcome | A qualifying stocked vehicle is omitted |
| No-match behavior | Deliberately impossible request | Unsupported alternatives presented as exact matches |
Keep the original request and the returned evidence. “It looked wrong” starts a discussion; identifiers and timestamps allow someone to reproduce it.
Report coverage with an honest denominator
Suppose eight of ten expected records can be retrieved. That is 8/10 for this deliberately selected sample. It is not 80% of your entire inventory and not an estimate of the percentage of buyers who will find you.
Separate record retrieval from field completeness. All ten cars might be present while equipment data is usable for only six. Combining those outcomes into one score would hide the work that matters to feature-based searches.
For a larger audit, define the eligible inventory population, deduplicate physical offers appropriately, document selection and repeat the same method. Break group results down by rooftop. A strong result at the flagship store can mask a completely disconnected smaller location.
This is a proposed testing method. No measurements in this article are presented as results from a completed Dealer Handshake customer test.
Give every defect an owner
A feed issue needs a different fix from a lead-routing issue. Assign the next action when the defect is recorded:
- The inventory owner corrects missing or wrong source information.
- The integration owner investigates mapping, ingestion or update delays.
- The assistant-experience owner checks tool use and how results are presented.
- The store verifies the corrected vehicle facts before the case is closed.
Repeat the affected case after the fix. Do not widen the project into a full technology replacement because one column is poorly mapped.
The useful operating question is simple: when the dealership changes a vehicle fact, can the relevant shopping experience show the right information within the freshness expectation you have agreed?
Where Dealer Handshake fits
Disclosure: I work on Dealer Handshake.
Dealer Handshake helps dealerships make their vehicles discoverable in compatible AI shopping tools and receive inquiries from interested buyers. Information for dealer groups and vendors.
The free website scan can start a visibility discussion. It should not replace the source-to-result checks above. A website scan, a feed test and a real shopper interaction reveal different things.
Bring ten vehicles, one inventory owner and one person responsible for the shopper experience to the same review. Leave with a short defect list and a repeatable test. That is a much stronger starting point than another meeting about being “AI ready.”
Start with your store: run the free scan or review the group and vendor connection.