How should eco brands publish llms.txt so assistants cite them?
Quick answer
Publish a claim-safe /llms.txt at the origin: one H1, a blockquote that matches the PDP, then markdown links to pages you want cited (materials proof, FAQ, comparison, policy). Optional /llms-full.txt expands the same URLs with on-page quick answers. It is not a second green essay. Do not invent carbon-neutral or plastic-free in the file. Strategy lives in AEO vs SEO. HTML FAQs live in FAQ citations. Measuring citations lives in the AI search audit.
Vs other AEO guides
| Guide | Job |
|---|---|
| AEO vs SEO | When to invest in answer engines vs classic search. Not a file spec |
| FAQ for AI citations | On-page question HTML and FAQPage schema |
| AI search audit | Whether ChatGPT, Perplexity, or AI Overviews already cite you |
| Materials proof page | The HTML hub this file should point to |
| Comparison pages | Brand-vs tables on the site, not in the index file |
| LLM ad-copy claims | ChatGPT writing ads. Different desk, same kit |
| This guide | Root llms.txt / optional llms-full.txt so assistants find cite-worthy URLs |
What each file is for
| Asset | Job |
|---|---|
/llms.txt | Curated CommonMark index: H1, optional blockquote, H2 lists of absolute HTTPS links |
/llms-full.txt | Optional longer index. Same URLs plus the quick answers already on those pages. Skip it until you can generate it |
| HTML proof pages | What should actually get quoted. Build the materials hub first if it is missing |
| robots.txt | Allow both files. A sitemap is for search crawlers, not a substitute |
The 2026 llmstxt.org revision still treats the file as a map agents search, then follow. Homepage rel="describedby" can point at it. None of that replaces on-page proof.
Steal this publish kit
- H1 = the brand name buyers search, not a slogan.
- Blockquote = 1 to 3 sentences copied from the allowed-claims kit. Same unit, same cert scope as the PDP. Run shaky numbers through the green claims calculator.
- Link 5 to 12 live URLs: materials/sourcing hub, FAQ, comparison (if claim-safe), shipping/returns, company cert/About, one hero collection. Absolute
https://only. - One-line notes after links must be facts already on the page (“GOTS cotton, license on /materials”), not new adjectives.
- Refresh when the substantiation folder changes. Stale files teach models last year’s claim.
What never goes in the file
| Leave out | Why |
|---|---|
| Unscoped eco-friendly, plastic-free, carbon neutral, net zero, 100% sustainable | The blockquote will get quoted as current fact |
| 2030 goals written as present tense | Aspirational copy belongs on a dated About FAQ, not the index |
| Staging, UTM, affiliate, or draft collections | Broken or tracked URLs poison the map |
| Every SKU PDP | Conflicting metafields become “the brand said X and Y” |
| Raw LCA PDFs with no HTML explanation | Models misread scope; put the table on the proof page first |
| Ad drafts and waitlist copy | That is the ad-copy gate, not GEO |
Ship it (Shopify included)
curl -I https://yourdomain.com/llms.txtreturns 200, UTF-8,text/plainortext/markdown, no login wall.- Shopify Online Store 2.0 cannot place a file next to
theme.liquidand get that root URL. Serve it from Hydrogenpublic/(see Hydrogen), a CDN worker in front of the shop, or an app that owns/llms.txt. - Allow the paths in robots.txt. Do not Disallow
/*.txtglobally. - Optional: homepage
rel="describedby"(or an HTTP Link header) pointing at the file.
Go / no-go and how to measure
| Call | When |
|---|---|
| No-go | The summary would be greener than the PDP, or you have no proof URL to list |
| Go | Kit-true summary, 5 to 12 live links, root 200, robots Allow, a human signed the inventory |
| Partial go | US URLs only until EU copy is scoped; or llms.txt now and skip llms-full until it is generated |
Re-run the visibility audit on “what is [brand] made of” and “is [brand] carbon neutral.” Winning looks like a citation of /materials or FAQ, not a hallucinated percent. Use the AEO readiness calculator for the rest of the stack.
The file is a map
If the pages behind the links are vague, the index just helps models quote the vagueness faster.
Frequently asked questions
How should eco brands publish llms.txt so assistants cite them?
Put a kit-true markdown index at the origin. Link live proof pages. Keep llms-full optional and generated. Do not invent green claims in the summary. Then audit whether assistants cite those URLs.
Does publishing llms.txt make ChatGPT cite us?
No. It is a discovery map, not a ranking factor. Citations still depend on crawlable, specific pages. The file helps if those pages already exist.
Do we need llms-full.txt?
Only if you can generate it from the same kit and keep it fresh. The proposal does not require it. A stale full file is worse than none.
Can the summary say carbon neutral if that is our 2030 goal?
No. Present-tense environmental claims in the file get quoted as current fact. Put the dated goal on About or FAQ, and link that page.
How is this different from the FAQ or AEO guides?
Those are pages and strategy. This is the root file that points at them. Build the pages first, then the map.