EKOM Catalogs in the Wild · Automotive Edition · 2026
Catalogs in the Wild — Automotive Edition

The Catalog Is Where the Money Leaks Out

Car buying stopped being a browsing problem and became a matching problem — decided by machines reading structured product data most dealerships have never had a way to read at depth. The catalog behind the listings has quietly become the fault line between the shoppers a dealership wins and the ones it never knew it lost.
01

The Industry Moved. The Catalog Didn't.

The storefront became the dealership — before anyone had a way to read it the way the machines now do.

For most of the last two decades, a dealership's online listing was a rough draft. A photo, a price, a phone number — and a salesperson on the other end of a test drive to smooth over whatever the listing got wrong. A garbled feature list or an inconsistent color name was an inconvenience, not a disqualifier, because a person was always the last step before a decision.

That's no longer where the decision happens. Buyers now spend roughly 14 hours shopping for a vehicle, about seven of them online, and they touch an average of 4.6 websites before they ever walk onto a lot. The digital storefront isn't a brochure that supports the sale anymore — for most shoppers, it is the sale, right up until the paperwork.

Dealers have chased the shopper there with their wallets, spending a record $9.96 billion on advertising in 2025 — roughly $739 a vehicle, nearly three-quarters of it digital. Every one of those dollars lands a shopper on a vehicle detail page and hands the outcome to whatever data is sitting behind it. On an average new vehicle now selling for $48,205, that listing is the highest-leverage asset a dealership owns — and the one almost no one has had a way to read at depth. The money is spent to get shoppers to the catalog; the catalog is where it either converts or quietly leaks away.

02

The Reader Changed — And It Doesn't Ask Questions

A human who hits a confusing listing picks up the phone. An answer engine drops the record.

The shopper reading the catalog is increasingly not a person at all. AI-assisted discovery has moved from novelty to default in a single buying cycle, and two independent 2025 studies now agree it's inside the funnel, not adjacent to it: 44% of car shoppers have used an AI-powered search tool, and 97% of those users say it will shape what they buy. Cox Automotive, measuring separately, found one in four new-vehicle buyers leaned on AI in their most recent purchase — the first year it even thought to ask the question.

What the shopper used to see
PHOTO · SPECS · PRICE
2026 GLE 450 Coupe
$79,900

A person reads around a bad field, cross-checks, and calls the store to confirm. The listing is a starting point for a conversation.

→ Ambiguity gets a phone call
What the machine reads
body_styleSUV ✗
fuel_type— (blank)
drivetrain4MATIC
price79900

An answer engine reads only the structured fields. One contradicts the vehicle; one is empty. It can't ask.

→ Ambiguity gets the record dropped

Same vehicle, two readers. The one that now decides what gets shown is the one that never picks up the phone.

This is the shift that changes the stakes on every field in a catalog. A human shopper who hits a confusing listing can still call and ask. An AI answer engine can't, and doesn't. It reads structured attributes, not intent — so when a field is ambiguous, blank, or contradicts the one next to it, the system doesn't guess in the shopper's favor. It drops the record, or answers with whatever survives after the bad field is discarded. A defect that a person would have shrugged off now removes the vehicle from the answer entirely.

And most dealerships can't even be cleanly read in the first place. In one 2026 audit of dealership websites, fewer than half were fully open to AI-style readers — the rest partially or entirely blocked, most often by bot-protection the store never chose to switch on. The machines that increasingly decide which cars get shown are hitting a wall before they reach the inventory.

Where they do get through, the tolerance for error has collapsed. Google disapproves any vehicle listing whose feed price doesn't exactly match the price on the landing page — and repeated mismatches can suspend the entire account. Marketplace deal-rating tools depend on clean trim and option data to price a car fairly; strip that data on ingestion and an accurately priced vehicle can be flagged a bad deal before a shopper ever sees the real number. The catalog isn't being read more forgivingly than before. It's being read literally, at machine speed, by systems that treat every inconsistency as a reason to move on.

03

The Margin for Error Got Thinner

A bad field used to get erased by the sale. Now it sits exposed for months, on inventory earning less per unit.

None of this would matter as much if margins were fat and inventory turned overnight. Neither is true anymore. Front-end gross profit on a new vehicle has fallen to roughly $3,284 (Q2 2025) — down from the $5,000-plus peaks of the pandemic, after three straight years of compression. And the cars aren't moving fast enough to bury a mistake: new-vehicle days' supply has run between 76 and 98 days through 2025 and into 2026, which means a vehicle — and every flaw in its listing — sits exposed for weeks or months before it turns.

Put those together and a single bad field changes shape. A wrong value used to get corrected by turnover — the car sold, the mistake became irrelevant. Now it sits exposed for weeks, on a unit already earning a fraction of what it did three years ago, in front of every marketplace and answer engine that re-reads the feed on its own terms. The thinner the margin and the slower the turn, the longer a data defect has to cost you — and the less room there is to absorb it.

04

What's Actually Hiding in a Catalog

Real records, exactly as found. Dealers withheld — the errors aren't. Brands named, because the defects are brand-specific.
Every example below is a live listing surfaced by an unattended read — no hand-built rules, no cherry-picking.
A twin-turbo V8 super-SUV, listed as electric / Mercedes-BenzCritical
Field on the recordValue as foundSignal
EngineElectric Motor✗ conflict
TrimAMG G 63 · performance V8✓ agrees — V8
Cylinders8, twin-turbo✓ agrees — V8
Two of the record's own fields — the trim and the cylinder count — agree it's a V8; the engine field is the lone outlier. The truth is determined against the evidence on the row, not the loudest field.
Not a typo a shopper would catch — a fact a shopper would trust, on one of the highest-margin vehicles in the building. And an EV filter that now surfaces a gas V8.
Fuel economy stored in the engine field / ChevroletCritical
Live now
The engine field literally reads 17 CITY / 20 HWY. On another unit it reads Charlotte, NC — a location where an engine spec should be.
What the read caught
The field is unusable for either purpose: engine search returns nothing, and the structured MPG data is discarded instead of displayed.
The filter looks fine until a shopper tries to use it — then the car simply isn't there.
New cars advertising a used-car inspection / Chevrolet · HondaCritical
Live now
New-vehicle descriptions carry pre-owned boilerplate — a "rigorous inspection process" and checks made prior to making any pre-owned vehicle available for sale. At one Honda store it's on every new car in the catalog.
What the read caught
Used-car copy on a new-car row isn't a formatting nit — it's a condition-and-disclosure problem with the buyer's name on it.
The kind of defect that's invisible in a spreadsheet and very visible in a compliance review.
4WD trucks telling buyers they're AWD / ChevroletHigh
Live now
Four-wheel-drive Silverados carry an AWD tag in their features — across a large share of the trucks in a single catalog. Silverado trucks don't offer AWD at all.
What the read caught
A false mechanical claim on the vehicle's own spec sheet — and a drivetrain filter that now routes trucks into the wrong bucket.
A truck buyer filtering for 4WD loses trucks that are sitting right there.
One car underpriced by ~$30,000 / VolkswagenCritical
Live now
A Taos SE AWD carries a dealer price roughly $30,000 below its siblings — "the cheapest vehicle in the catalog by a factor of ten." A Ford Raptor at another store runs $10,700 over MSRP against an otherwise consistent pattern.
What the read caught
Both are feed artifacts, not real prices — one a bait-looking listing headed for ad disapproval, one a margin-and-trust problem.
Price integrity is exactly what the marketplaces now check automatically, and exactly what fails silently.
Coupes filed under "SUVs" / Mercedes-BenzHigh
Live now
GLE Coupe units carry body_style = SUVs despite their own descriptions — and VIN prefixes — confirming Coupe.
What the read caught
The structured field is what every filter, marketplace, and agent actually reads. Every coupe shopper's search loses a coupe.
The catalog disagreeing with its own sales copy — and the machine trusting the wrong half.
Listings publishing their own template code / Volkswagen · InfinitiCritical
Live now
Vehicle descriptions render raw merge tags a shopper can read: {YMM} {SERIES} in {EXTERIORCOLOR}, and on other units {excited} and {features}. The template engine never filled them before the page went live.
What the read caught
Unsubstituted tokens aren't a typo — they're the listing telling every shopper and crawler that the page is broken, on the copy meant to sell the car.
The one defect a buyer needs no expertise to spot — and the machines read it as missing content.
Two different prices on the same car / JeepCritical
Live now
A Compass Latitude's description advertises $30,484 while its structured price field reads $31,231 — the page states one number and the system quotes another.
What the read caught
An advertised-price mismatch inside a single record — the exact discrepancy Google disapproves automatically, and the kind an advertised-price rule is written around.
Not a feed artifact this time — two conflicting prices a shopper can screenshot, on the live page.
05

The Pattern Is the Point

The most important finding isn't any single defect. It's that the same defects recur everywhere — unprompted.

So we read the inventories of 13 real dealerships directly — seven brands, two of the largest dealership website platforms, single-brand stores through high-volume metro rooftops — with nothing supplied: no schema, no configuration, no one telling the system what to look for. The same defect classes surfaced in catalog after catalog, and more than four in five findings were high-severity — errors that break commerce or search, not cosmetic nits.

The striking result is a meta-finding: read a set of independently-operated dealerships and the same classes of defect keep appearing, without anyone being told what to look for. That's not a skill gap at any one store. It's structural — the signature of a catalog that has more surface area than any team can review by hand.

The same defects, across independent dealerships
Share of the 13 catalogs where each defect appeared
Feature lists packed into one field shoppers can't filter77%
Broken or placeholder stock numbers77%
Missing prices — no savings shown77%
Interior color mixed up with seat material69%
Wrong data sitting in the engine field62%
Vehicles filed under the wrong body style54%
Entire catalogs with no photos at all46%
Entire catalogs with no fuel type listed31%

These aren't cosmetic. A blank fuel-type field makes a diesel or an EV invisible to the one filter built to find it — and in the stores where it occurred, it hit more than half of every vehicle on the lot, including electric models shoppers filter for first. Missing images — roughly a third of all the vehicles we read — get listings suppressed on every major marketplace, since a photo-less vehicle is the fastest thing to down-rank. Each defect maps straight to lost syndication, broken search, or eroded trust on the very page a shopper is standing on.

The findings, by severity
Surfaced unprompted, on catalogs never seen before
32%
54%
12%
Critical 32%   High 54%   Medium 12%   Low 2%  —  more than four in five break commerce or search directly, not style
The part worth sitting with

No one supplied a schema. No one mapped a field. No one told it what to look for.

Every one of these catalogs was read cold — nothing supplied, nothing configured. If a catalog no one has ever seen can be read this completely, and the same defects surface store after store, then the reason those defects are still live on the storefront isn't effort or attention. It's that no one has had a way to read a catalog this closely before — the constraint is scale, not skill.

After this study, we ran the same cold pass on a further seven-rooftop metro group — luxury imports and domestic trucks the first thirteen didn't include: Porsche, Jaguar, Maserati, Audi, Acura, Jeep, and Ram. The same defect classes surfaced again, on brands and segments the study hadn't touched — plus a few the first set never showed: unrendered template code on live pages, an advertised price that contradicts the price in the same record, entire used lots with no mileage field at all, and used vehicles that should read Certified. Twenty rooftops now, and the pattern only hardens: the wider the read, the more clearly the constraint is scale, not any one store's skill.

06

The Catalog Is the Starting Point, Not the Ceiling

The new-vehicle listing is where dirty data is visible. It isn't where it's most expensive.

The inventory catalog is the most-scrutinized data a dealership manages — every shopper, marketplace, and agent checks it first. That's exactly why it's the way in, not the whole story. The specific defects above don't travel into the parts catalog or the customer record. But the condition that produces them does: data populated by manual, uncoordinated entry, at a scale no team can review by hand. That same condition runs the two systems where the real money lives — and resolving the catalog is how a dealership starts to reach them.

Unlock 01 · Fixed Operations

Parts and service is ~13% of dealership revenue — and roughly half of gross profit.

Franchised dealers wrote more than 270 million repair orders and over $156 billion in service and parts sales in 2024. It's the highest-margin part of the store, and it runs on data with the same failure modes as inventory: fitment and supersession catalogs (governed by ACES/PIES standards) full of gaps, wrong attributes, and inconsistent updates. In auto-parts e-commerce, inaccurate fitment data is a leading driver of returns — the exact "field that disagrees with the vehicle" problem, applied to the most profitable counter in the building.

And the customer relationship those service dollars depend on is itself leaking. Selling-dealer service retention on late-model vehicles fell to 54% in 2025, down from 72% just two years earlier — even as the average vehicle on the road reached a record 12.8 years, meaning the work is growing while the dealer's share of it shrinks. That gap reaches far past the service drive: 74% of owners who service where they bought are likely to buy their next vehicle there too. The repeat sale rides on the service relationship, and the service relationship rides on the customer record.

Unlock 02 · The Customer Record

The relationship that drives repeat revenue rides on a customer record no one owns.

Service retention isn't falling because of price — dealer repair costs actually run below independents. It's falling at the point where the customer relationship is managed. Dealer customer data is siloed across CRM, DMS, and service systems and entered manually by different teams — exactly how duplicate and fragmented records form, the same manual-entry failure mode as the inventory catalog, on the data that decides whether a buyer ever comes back. It surfaces first in the catalog because that's the data the outside world reads. Resolve that, and the same discipline — reconcile, determine what's true, keep it current — reaches the parts data and the customer record next.

None of this means every dealer's parts or customer data is broken — that's a separate read, on separate data. What the catalog proves is that a scale-of-data problem rarely stays contained to the one system a customer happens to see. The inventory feed is simply the first place it becomes undeniable.

07

The Fix Isn't More People. It's a Different Kind of Read.

Most of what's wrong with a catalog isn't missing information — it's information that disagrees with itself. A trim field says one thing; the description says another. An engine spec is blank, but three sibling records with the identical drivetrain all agree on the correct value. The fix isn't guessing which value sounds right. It's determining, field by field, which value is actually true — against the evidence sitting in the same catalog, against the manufacturer spec, against the record that got it right.

That distinction is why generation alone makes this worse, not better. A model can produce fluent, confident copy about a vehicle with no way to know whether the underlying attributes are correct — and when the source data is wrong, generation doesn't catch the error. It describes the error more convincingly, and publishes it everywhere the catalog feeds. Resolving a catalog means reconciling the conflicting values first, citing what justified each correction, and only then generating anything — because by then there's something true to generate from.

The work runs as one continuous loop — normalize what's already there against its own evidence, enrich what's missing, distribute the result in the exact shape each channel requires, then keep watching those channels, because a marketplace changing its rules is when a resolved catalog starts to drift again. Every correction carries the evidence behind it, so nothing is fixed on a guess. That's the resolution layer — the read every catalog in this paper went through, and the same read EKOM would run on any dealership's.

Start with a catalog analysis

Request a catalog analysis. See what the machines see in your own inventory.

Share a representative slice of your inventory — the kind of export your systems already produce. You'll get back a clear read of where your product data carries error today, what it's costing across your channels, and how much of it resolves on the first pass — documented against your own records, with the evidence behind every finding. No commitment, and no services engagement required. Every catalog in this paper looked fine from the front end. The only question is whether you'd rather know.

Jonah Santo
Chief Commercial Officer, EKOM · [email protected] · ekom.ai/analysis
Sources
  1. NADA, 2025 Annual Financial Profile of America's Franchised New-Car Dealerships (ad spend $9.96B, $739/vehicle, ~75% digital; average new price $48,205; service & parts $156B / 270M+ repair orders, 2024).
  2. Cox Automotive, 2025 Car Buyer Journey Study (~14 hrs shopping / ~7 online; 4.6 websites; 1 in 4 new-vehicle buyers used AI).
  3. Cars.com survey, November 2025 (44% of shoppers used AI-powered search tools; 97% of AI users say it will impact their purchase).
  4. Savvy Dealer, 2026 dealer-website AI-visibility audit (~44% of sampled dealer sites fully open to AI-style readers).
  5. Google Merchant Center / Vehicle Ads policy, 2025–26 (feed price must match landing-page price; mismatches disapproved).
  6. Haig Partners, Q2 2025 Report (front-end gross ~$3,284 per new vehicle, down from $5,000+ pandemic peaks).
  7. Cox Automotive new-vehicle inventory reports, 2025–26 (days' supply ~76–98).
  8. Cox Automotive Service Industry Study, November 2025 (selling-dealer service retention 54%, down from 72% in 2023; 74% repeat-purchase likelihood; dealer service visits down 12% since 2018).
  9. S&P Global Mobility, 2025 (average U.S. vehicle age 12.8 years across ~289M vehicles).
  10. Auto Care Association, ACES/PIES parts-data standards.
  11. CDK Global, "Clean CRM Data Is the Key to More Car Sales," July 2026 (siloed, manually-entered customer records create duplicates).
  12. EKOM catalog analysis, 13 live dealership inventories, July 2026; extended to a further seven-rooftop metro group (twenty rooftops total), July 2026.
EKOM · The Resolution Layer for Product Data© 2026 EKOM AI · Catalogs in the Wild