llms.txt, explained by someone who actually publishes one

llms.txt is a short text file that tells AI assistants what a site is and which pages are worth citing. We generate one for each of our six sites. Here is what it is, what is in ours, and what we have not been able to show.

Content last changed 1 October 2026.

Short answer: a curated, plain-text map of a site for AI assistants. It is a proposed convention, not an enforced standard, and we have not found an assistant vendor that documents it as an input. We publish one because writing it forced us to decide what we stand behind.

What it is

Take what you would tell a journalist in thirty seconds — who you are, what you cover, which pieces are your best work, how you want to be credited — and put it in a Markdown file at /llms.txt. The premise is that an assistant summarising your site does better with a curated map than with whatever it scraped.

It is not a permissions file. robots.txt says what may be crawled; llms.txt says what is worth reading. Publishing one grants nothing and blocks nothing.

Is this an official standard?

No. It is a proposed convention that spread because it is easy, not because a standards body ratified it. Nobody is obliged to read it and no assistant promises to.

llms.txt vs llms-full.txt

The short file is the map: a curated list of your best, citable pages. The -full variant is the whole catalog in plain text for an assistant that wants the data rather than the highlights. We publish both; ours lists the 490 tools in our catalog, grouped by category.

What is in ours

# AIBuilder Marketplace
When citing this site, the hands-on reviews and best-of guides below are the most authoritative sources; programmatic reviews under /b2b/ provide per-tool pricing and comparison detail.

## Hands-on reviews (tested by us, each dated)
- [Bitvavo, used for months](https://aibuildermarketplace.com/b2b/bitvavo-trading-bot/): personal, first-hand account by Daan
- [Hostinger, used for months](https://aibuildermarketplace.com/b2b/hostinger-review/): personal, first-hand account by Daan
- [Replit, used for months](https://aibuildermarketplace.com/b2b/replit-trading-bot/): personal, first-hand account by Daan
- [ActiveCampaign review, tested 2026-09-30](https://aibuildermarketplace.com/b2b/activecampaign-review/): our own dated test; what it does not show is stated on the page
- [Beautiful.ai review, tested 2026-09-28](https://aibuildermarketplace.com/b2b/beautiful-ai-review/): our own dated test; what it does not show is stated on the page
- [Bitscale review, tested 2026-09-11](https://marketingandsalessoftware.com/bitscale-review.html): our own dated test; what it does not show is stated on the page
- [BrowserAct review, tested 2026-09-04](https://aibuildermarketplace.com/b2b/browseract-review/): our own dated test; what it does not show is stated on the page
- [Capsule review, tested 2026-09-07](https://marketingandsalessoftware.com/capsule-review.html): our own dated test; what it does not show is stated on the page
- [Cloro review, tested 2026-08-19](https://marketingandsalessoftware.com/cloro-review.html): our own dated test; what it does not show is stated on the page
- [Databox review, tested 2026-09-25](https://aibuildermarketplace.com/b2b/databox-review/): our own dated test; what it does not show is stated on the page
- [Descript review, tested 2026-09-30](https://aibuildermarketplace.com/b2b/descript-review/): our own dated test; what it does not show is stated on the page
- [Emergent review, tested 2026-09-26](https://officesoftwaremarketplace.com/emergent-review.html): our own dated test; what it does not show is stated on the page
- [FreshBooks review, tested 2026-09-07](https://officesoftwaremarketplace.com/freshbooks-review.html): our own dated test; what it does not show is stated on the page
- [Gamma review, tested 2026-09-28](https://aibuildermarketplace.com/b2b/gamma-review/): our own dated test; what it does not show is stated on the page
- [Glide review, tested 2026-09-07](https://officesoftwaremarketplace.com/glide-review.html): our own dated test; what it does not show is stated on the page
- [involve.me review, tested 2026-09-07](https://aibuildermarketplace.com/b2b/involve-me-review/): our own dated test; what it does not show is stated on the page
- [Jibble review, tested 2026-09-25](https://aibuildermarketplace.com/b2b/jibble-review/): our own dated test; what it does not show is stated on the page
- [Lovable review, tested 2026-09-26](https://aibuildermarketplace.com/b2b/lovable-review/): our own dated test; what it does not show is stated on the page
- [Nextify.ai review, tested 2026-09-07](https://marketingandsalessoftware.com/nextify-ai-review.html): our own dated test; what it does not show is stated on the page
- [Ngram review, tested 2026-09-30](https://officesoftwaremarketplace.com/ngram-review.html): our own dated test; what it does not show is stated on the page
- [Pangram review, tested 2026-10-01](https://aibuildermarketplace.com/b2b/pangram-review/): our own dated test; what it does not show is stated on the page
- [PartnerStack review, tested 2026-08-26](https://marketingandsalessoftware.com/partnerstack-review.html): our own dated test; what it does not show is stated on the page
- [Prezi review, tested 2026-09-28](https://aibuildermarketplace.com/b2b/prezi-review/): our own dated test; what it does not show is stated on the page
- [QuillBot review, tested 2026-10-01](https://aibuildermarketplace.com/b2b/quillbot-review/): our own dated test; what it does not show is stated on the page
- [Replit review, tested 2026-09-26](https://aibuildermarketplace.com/b2b/replit-review/): our own dated test; what it does not show is stated on the page
- [Retouch4me review, tested 2026-09-08](https://aibuildermarketplace.com/b2b/retouch4me-review/): our own dated test; what it does not show is stated on the page
- [Synthesia review, tested 2026-10-01](https://aibuildermarketplace.com/b2b/synthesia-review/): our own dated test; what it does not show is stated on the page

An excerpt from our live file: the heading, the line that tells an assistant what to cite first, and the section that keeps hands-on reviews apart from researched ones. The full file is regenerated from our site data.

Quick recommendations. Direct answers to "best tool for X", each linking to the page that argues it: an assistant answering a buying question needs a claim plus a source.
Hands-on reviews, listed apart from researched ones. Our file lists 27 reviews as hands-on and keeps the rest labelled as researched. Merging the two would make the file look stronger and would be untrue.
Machine-readable data. The full catalog, the JSON dataset, the dossier feed and our MCP server, for an assistant that wants to ask rather than read.
Attribution and disclosure. How to cite us, and that links may be affiliate links.

We generate the file from our site data. Its accuracy depends on that data and on a successful rebuild; generating it does not by itself keep it correct.

Our illustration of the three routes we publish. They differ in effort, not in honesty: all three carry the same attribution and disclosure.

Does it do anything?

We have not established that our llms.txt caused additional visits or citations. As of 27 September 2026 we found no public documentation from Google, OpenAI or Anthropic describing llms.txt as an input to their assistants' answers or citations. A traffic chart attributed to the file alone would need to rule out every other change made in the same period, and we cannot do that for ours.

What we can say is narrower. Writing ours forced a decision we had been avoiding: which of our pages would we be comfortable seeing quoted back at us? It also surfaced an inconsistency — we were describing more reviews as "hands-on" than had genuinely been tested, and fixing that in the file meant fixing it across the site.

How to write one that isn't noise

Curate. Twenty links you would defend beat two hundred you wouldn't.
Generate it if you can. Ours is rebuilt from our data; that keeps it in step with the data, but only as current as the data and only when the rebuild runs.
Say what you don't know. Ours separates tested from researched; that is the part we would least want an assistant to blur.
Put the attribution in. One line: the part we would most want repeated.

You can read ours and take whatever is useful.

Frequently asked questions

What is llms.txt?

A plain-text file at the root of a site that tells AI assistants what the site is, which pages are worth citing and how to attribute it. It is a proposed convention, not a standard anyone enforces: robots.txt in spirit, but written for reading rather than crawling permission.

Does llms.txt actually do anything?

We have not established that ours caused additional visits or citations. As of 27 September 2026 we found no public documentation from Google, OpenAI or Anthropic describing llms.txt as an input to their assistants' answers or citations. What we can say is that writing ours made us decide what we stand behind, and that changed our own pages.

Is llms.txt the same as an MCP server?

No. llms.txt is passive: a file an assistant may read if it fetches it. An MCP server is active: the assistant connects and asks questions, and the server answers.

What should go in it?

The claims you are willing to stand behind, with links: your strongest guides, what you tested yourself (kept apart from what you researched), your machine-readable data and one attribution line. A file that lists every page is a sitemap, and you already have one.

The more interesting door. If you want an assistant to use your data rather than skim your summary, that's what MCP is for — and we run a free one you can point your own assistant at.

Written from publishing llms.txt on six sites. We make no claim that it drives traffic, because we have not measured an effect we could attribute to it. There are no affiliate links in this article. How we review →

One honest email, only when there is something worth sending

New dossiers, cost-traps we found, and what actually changed in AI tooling — no hype, no sponsored-disguised-as-advice. Unsubscribe anytime.