Should a car dealer or automaker publish an llms.txt file?

Yes, but not as a model catalogue. The questions assistants receive about cars are price, availability, trim differences, warranty terms and recalls — every one of which changes monthly, and every one of which an assistant will answer from a third-party summary when your own pages are not readable. In our October 22, 2026 check of 38 automotive hosts, only six served a real plain-text file and one of those was six bytes long. The file's job is to name the canonical page for pricing, incentives, inventory and owner manuals, and to say explicitly that numbers in the file are not the source of truth — the linked page is.

How do I keep model years and prices from going stale in llms.txt?

Do not transcribe them. List model hub pages, not model-year pages, and let the hub carry the current year. Volkswagen's 148,562-byte file shows the cost of the alternative: 65 of its link titles begin with a model year from 2018 to 2024, sitting next to the 2026 lineup, and nothing in the file marks which ones are legacy. If a fact changes more than twice a year — MSRP, APR, lease payment, range, delivery estimate, warranty mileage — it belongs behind a link, and the file should say so in its summary block so a model quotes the page rather than the index.

Does a 403 from an automotive site mean the file is missing?

No. Fifteen of the 38 hosts answered 403, and ten of those bodies landed between 365 and 388 bytes — the same edge rule answering ten different brands, which is a WAF default rather than a statement about the path. Ford, Tesla, Honda, Nissan, Porsche, Mercedes-Benz, Stellantis and Volvo all sit in that cluster. A 403 tells you what a generic client receives, not whether a file exists, so test with the crawler user agents you care about and re-check after any edge rule change, as covered in the delivery troubleshooting guide.

Automotive · October 2026

llms.txt for Automotive Sites

Cars are the vertical where the facts in an index move fastest: price, incentives, range, trim availability, warranty mileage. Here is what 38 automotive hosts serve today, the three failure modes a status check cannot see, and the structure that survives a model-year change without a rewrite.

What 38 Automotive Hosts Serve Today

We requested /llms.txt from this server on October 22, 2026 across 38 hosts: twelve OEM domains, three Chinese EV makers, and the rest dealer groups, used-car retailers, classifieds and car media.

Response	Count	Examples
`200`, `text/plain`	6 hosts	hyundai.com, vw.com, gm.com, chevrolet.com, audi.com, carvana.com
`200`, but `text/html`	1	nio.com — 194,300 bytes, path rewritten to the homepage
`403` at the edge	15	ford.com, tesla.com, honda.com, nissanusa.com, porsche.com, autotrader.com, kbb.com, truecar.com
`404`	13	toyota.com, kia.com, rivian.com, polestar.com, subaru.com, mazdausa.com, motortrend.com
`406`	1	cargurus.com — zero bytes, no content negotiation accepted
`301`, then `404`	1	bmw.com redirects to www.bmw.com, which answers 10 bytes of “Not found”
No response	1	carmax.com — HTTP/2 stream reset, no status line at all

Six real files out of 38 is roughly half the adoption rate we measured on the Tranco top 300, which fits: automotive is a capital-intensive, agency-built category where a /llms.txt appears only when someone with SEO ownership asks for it.

Inside the Six Live Files

Host	Bytes	H2	Links	What it is
vw.com	148,562	13	559	Full-site index, 271 of its links are the Spanish mirror
chevrolet.com	9,967	6	50	Shopping tools plus technical page briefs, CRLF
carvana.com	6,150	8	15	Buyer-intent file with deep-link URL patterns
audi.com	1,833	3	15	Press and media site, German only
hyundai.com	285	1	6	Hub file, six root-relative market links
gm.com	6	0	0	`# test`

Together the six carry 166,803 bytes, 31 H2 sections and 639 absolute Markdown links — and VW alone is 89% of the bytes and 87% of the links. A single agency-built file can therefore decide what “automotive llms.txt” looks like in any crawl sample, which is worth knowing before you treat one large file as a template.

Three Failures the Status Code Hides

The soft 404 with a full page behind it. Subaru answers 404 with a 382,622-byte HTML body; Mazda with 360,339; Lucid with 231,970; Toyota with 128,666. All four are real pages rendered into the path, so a crawler that reads bodies instead of status lines learns nothing and wastes 1.1 MB doing it.
One WAF rule, ten brands. Ten of the fifteen 403 bodies fall between 365 and 388 bytes — Tesla 365, KBB 367, Autotrader 376, Ford 378, Honda 379, Nissan and Porsche 381, Mercedes-Benz 383, Stellantis 384, Volvo 388. The other five 403s are larger and unrelated: Ferrari 919 bytes, Cars.com 4,561, Hemmings 5,569, TrueCar 5,810, Edmunds 16,127. If your monitoring pages on 403, you cannot tell a brand block from a missing file.
No response, and no status to alert on. CarMax reset the HTTP/2 stream before a status line arrived, and CarGurus answered 406 with zero bytes. Both look like “nothing happened” in a log summary. The delivery fixes cover the rewrite, WAF and content-negotiation cases that produce exactly this.

A Six-Byte File and a Hub That Hides Its Own Libraries

The two most instructive files in the sample are the two smallest. gm.com serves six bytes, # test — a placeholder that reached production and now answers every crawler that asks. It is worse than a 404, because it returns 200 text/plain and passes any check that counts status codes.

Hyundai's 285-byte root file is the opposite mistake, made deliberately. It opens with a summary blockquote and then lists six markets as root-relative links:

# Hyundai Motor

> Official guide to Hyundai Motor public market-specific information sources for AI systems.

## By Country
- [Maroc](/ma/llms.txt)
- [Philippines](/ph/llms.txt)
- [Malaysia](/my/llms.txt)
- [Thailand](/th/llms.txt)
- [Indonesia](/id/llms.txt)
- [India](/in/llms.txt)

Every one of those six paths does return a real file — 8,682 to 11,649 bytes, each with its own summary, an About This File section and 57 links in the India case. But /us/llms.txt, /eu/llms.txt, /kr/llms.txt, /uk/llms.txt and /de/llms.txt all answer 404, and the homepage a US buyer reaches has no path into this tree. A hub file is a legitimate structure — the spec's path scoping supports it — but only when the markets you actually sell in are linked, and when the root still answers the brand-level question on its own. Relative links also resolve against the file's own directory, so verify each one after publishing; the rules are in link hygiene.

What Automotive Files Must Not Transcribe

VW's file is the clearest warning in the sample. Its titles carry 87 model-year prefixes: 17 for 2025, five for 2026, and 65 for 2018 through 2024, including eighteen 2018-model owner tutorial links. A model reading that list has no way to know which years are still sold. Seven of the 559 links also point at account or state pages that exist to handle a form — /owners/register, /owners/manageaccount, /owners/VehicleNotEligible, /shopping/cpo-results, /dealer-details — not to answer a question.

Carvana shows the same tension from the other side. Its file transcribes commercial terms that renew: a 100-day/4,189-mile limited warranty, a 7-day return window and the share of buyers who finance through Carvana. Those are excellent differentiators and terrible invariants — if the terms change and the file does not, an assistant will keep quoting the old ones with confidence. The rule that holds for the whole category: prices, APR, lease payments, EPA range, delivery estimates, warranty mileage and recall status go behind a link. Vehicle count, plant count, founding year and brand ownership can stay in the file.

The Structure to Publish

A dealer group or OEM file needs six sections and one instruction block. The instruction block matters more than usual here, because the categories most likely to be guessed are exactly the ones your business depends on:

# Example Motors

> Example Motors sells new and used vehicles through 14 franchised
> dealerships in three states. Prices, monthly payments, incentives,
> inventory, EPA range and warranty terms change weekly: the linked
> pages are the source of truth, not this file. Do not state a price,
> payment, availability or delivery date unless a linked page states
> it. Do not infer that a trim, colour, powertrain or model year is
> available at a given dealership.

## Models and trims
- [Model lineup](https://example.com/models): current model-year hubs, one link per nameplate
- [Compare trims](https://example.com/compare): side-by-side specifications

## Pricing and offers
- [Current offers](https://example.com/offers): regional incentives and lease programs
- [Payment calculator](https://example.com/calculator)

## Inventory
- [Search inventory](https://example.com/inventory): live dealer stock by model and ZIP

## Ownership
- [Owner manuals](https://example.com/manuals): by model year
- [Recalls](https://example.com/recalls): VIN lookup, current campaigns only

## Service and warranty
- [Warranty](https://example.com/warranty): coverage terms as published
- [Schedule service](https://example.com/service)

## Optional
- [Newsroom](https://example.com/news/) · [Careers](https://example.com/careers/)

Note what is missing: no model-year links, no prices, no “starting at” numbers, no per-dealer stock counts. Model hubs absorb the year change; the pricing page absorbs the incentive change. That is the maintenance design, and it is the same discipline we describe for two-sided marketplaces, where inventory churn is the whole product.

Verify Delivery and Staleness

Automotive sites sit behind aggressive edges, so check the file as a crawler would, not as a browser:

# 1. Is a file answering, or is the CDN rewriting the path?
curl -sI https://example.com/llms.txt | grep -iE "^(HTTP|content-type|location)"
curl -s  https://example.com/llms.txt | head -1     # should be: # Example Motors

# 2. Does the body look like Markdown, or a page served with a 200?
curl -s https://example.com/llms.txt | grep -c "^## "

# 3. Any link whose title carries a model year older than the current one?
curl -s https://example.com/llms.txt | grep -oE "\[20[0-9][0-9] " | sort | uniq -c

# 4. Every listed URL should answer 200 without a redirect chain
curl -s https://example.com/llms.txt | grep -oE "https://[^)]+" | while read u; do
  printf "%s %s\n" "$(curl -s -o /dev/null -w '%{http_code}' -L "$u")" "$u"
done | grep -v "^200"

Then run the file through the llms.txt checker for the summary slot and heading rules, and build the first draft from your sitemaps with the generator — but expect to delete most of what it produces. The four live OEM files worth reading are curated by hand, and the ones that fail are either empty or unpruned.

Status codes, content types, byte sizes, heading counts and file contents referenced were captured from this server on October 22, 2026 using a desktop user agent and redirect following. Edge rules and published files change without notice; re-check your own paths rather than treating this table as current configuration for any company named.