AI.txt is a proposed convention — a small text file at yourdomain.com/ai.txt where you declare how your content was created and what your policies are on AI training and AI-assisted production. Let's be upfront: unlike robots.txt or ads.txt, ai.txt is not yet a formal internet standard. No W3C spec, no IETF RFC, no major crawler officially committed to reading it. What it does offer is a clean, machine-readable way to state your position publicly — and as AI transparency rules tighten worldwide, having that declaration in place is cheap insurance.
Why declare anything at all?
Three pressures are converging on publishers. First, readers increasingly want to know whether what they're reading was written by a person, drafted by AI, or something in between — especially for advice in finance, health, and legal topics where trust is everything. Second, regulators are moving: the EU's AI Act includes transparency obligations for AI-generated content, and similar disclosure expectations are spreading. Third, AI companies themselves are building training-data filters and publisher preference systems, and a machine-readable declaration is exactly the kind of signal those systems can consume.
An ai.txt file addresses all three at once: one public, structured statement of your content origin and AI policies, readable by humans and machines alike.
The proposed ai.txt format
Since no standards body has ratified ai.txt, what's below is the format gaining traction among publishers and tooling — a simple key-value file. Treat it as a convention to adopt, not a spec to cite:
# ai.txt — Content Origin & AI Policy for yourdomain.com
# This is a proposed convention, not a formal standard
content-origin: human
training-permission: blocked
ai-training-contact: licensing@yourdomain.com
editorial-policy: https://yourdomain.com/editorial-policy
fact-checking: manual
ai-generated-labeling: labeled
last-updated: 2026-10-07
version: 1.0What each field means
- content-origin — how your content is primarily created. Sensible values:
human(written entirely by human authors),ai-assisted(human-led, with AI used for editing, research, or drafting support), orai-generated(primarily produced by AI models). Be honest — this field only helps if it's true. - training-permission — your stance on AI training use of your content:
allowed,blocked, orlicensed-only. This is a declaration of intent, not an enforcement mechanism — pair it with actual robots.txt rules for training crawlers if you meanblocked. - ai-training-contact — an email where AI companies can reach you about licensing. If you ever want to be paid for training use instead of just opting out, this is how they find you.
- editorial-policy — a URL to your human-readable editorial standards page. This is the field with real SEO-adjacent value: it gives both readers and machines a verifiable statement of your process, which supports the experience and trustworthiness signals search quality guidelines look for.
- fact-checking — your verification process:
manual,editorial-review,automated, ornone. - ai-generated-labeling — whether AI-involved content on your site is labeled for readers:
labeledorunlabeled. - last-updated / version — keep these current so anyone reading the file knows it's maintained, not abandoned.
How ai.txt differs from robots.txt and llms.txt
These three files answer different questions, and confusing them is the most common mistake:
- robots.txt answers "which crawlers may request which URLs?" — it's an access-control signal with 30 years of history and RFC 9309 behind it. This is where you actually block training bots.
- llms.txt answers "what should an AI read first to understand my site?" — a proposed markdown format (H1 title, blockquote summary, H2 link lists, per the llms.txt spec by Jeremy Howard) that helps language models navigate your content efficiently. It's about guidance, not permission.
- ai.txt answers "how was this content made, and what's your AI policy?" — a declaration of origin and intent. It's about transparency, not access control.
A complete setup uses all three: robots.txt to enforce, llms.txt to guide (see the llms.txt generator), and ai.txt to declare. Generate your declaration with an ai.txt generator so the format stays consistent.
Setting it up in 10 minutes
- Decide your honest values. Don't write
content-origin: humanif AI drafts half your posts. Pick the value that's actually true — the field's only value is credibility. - Write your editorial policy page first. The
editorial-policyURL should point to a real page describing who writes your content, how it's reviewed, and how errors are corrected. Without it, the ai.txt is just claims. - Create the file. Plain text, UTF-8, at
yourdomain.com/ai.txt. Keep a copy in version control so changes are tracked. - Link it from your footer or about page. A machine-readable file nobody knows about helps nobody. Mention it alongside your privacy policy and terms.
- Match it with enforcement. If your file says
training-permission: blocked, your robots.txt should actually block the training crawlers (GPTBot, ClaudeBot, CCBot). A declaration without enforcement is just a wish. - Review it yearly. Update
last-updatedand revisit your values as your production process changes.
What to put on your editorial policy page
The editorial-policy field is only as strong as the page it points to. A two-sentence "we write great content" page undermines the whole declaration. A credible editorial policy page covers: who writes for the site (staff roles or contributor criteria), how topics are chosen, the review process before publication (who edits, who fact-checks), how you handle corrections and updates to published articles, whether and how AI tools are used in the workflow, and how readers can report errors. You don't need legalese — a few honest paragraphs beat a polished but empty page. This page does double duty: it backs your ai.txt declaration and gives human quality raters exactly the transparency signals they look for.
The honest limitations
Don't oversell this file — to yourself or your readers. Today, no major AI crawler is documented as reading ai.txt, so it doesn't currently change what any bot does. It carries no legal weight on its own; it's a statement of position, not a license or a contract. And because there's no ratified spec, two sites' ai.txt files may use slightly different fields — that's fine for now, but it means you should keep your fields simple and self-explanatory.
Why bother, then? Because conventions become standards when enough of the web adopts them — robots.txt itself started as an informal convention in 1994 before becoming RFC 9309 nearly thirty years later. Early adoption costs you ten minutes and positions you ahead of disclosure requirements that are clearly coming. Just don't mistake the declaration for the enforcement.
FAQ
Is ai.txt an official standard?
No. As of 2026 there is no W3C, IETF, or other formal specification for ai.txt. It's an emerging convention proposed by publishers and tooling vendors. Use it as a public declaration, not as a compliance mechanism.
Do any AI crawlers actually read ai.txt?
None of the major AI companies document ai.txt support today. The file's value right now is transparency for readers, regulators, and future tooling — plus a timestamped public record of your stated policies.
Should I still use robots.txt if I have ai.txt?
Absolutely — they do different jobs. Robots.txt is the enforcement layer that actually stops compliant training crawlers from fetching your pages. Ai.txt is the declaration layer that states your policy. Use both: declare in ai.txt, enforce in robots.txt.
What should I put for content-origin if I use AI for editing?
ai-assisted — that's exactly what the value is for. It means humans lead the process and AI helps with editing, research, or drafting support. Reserve human for fully human-written content and ai-generated for content primarily produced by models.
Does ai.txt help with SEO or E-E-A-T?
Indirectly. Google's quality guidelines value demonstrated experience, expertise, and trustworthiness. A public editorial policy page linked from your ai.txt gives both human reviewers and automated systems a verifiable statement of your process. It's a supporting signal, not a ranking factor — the content itself still has to be good.
Where should the file live?
At your domain root: yourdomain.com/ai.txt, served as plain text — same placement convention as robots.txt and ads.txt. Link to it from your footer or about page so humans can find it too.
Pairing ai.txt with llms.txt
If you're adopting ai.txt, consider llms.txt alongside it — they reinforce each other. While ai.txt declares how your content was made, llms.txt tells AI systems what's worth reading on your site: a markdown file with your key pages organized for efficient model consumption. Together they send a coherent message: "here's what we publish, here's how we made it, and here's our policy on AI use." One declares provenance, the other improves discoverability — and both cost minutes to implement.
The takeaway
Ai.txt won't stop a single bot today — and anyone telling you otherwise is selling something. What it does is stake your position publicly, in a format machines can read, at a moment when AI transparency is moving from nice-to-have to expected. Ten minutes to write it, one editorial policy page to back it up, and robots.txt rules to enforce what you declare. That's the honest version of AI content transparency in 2026. Start today: one file, one honest declaration, zero regrets later.
