alltio.
← Part of How it works
For rights holders

How AI companies know your work is registered

Discovery works in the opposite direction from what most people expect. AI companies don't find a mark on your file — they ask the registry.

You've registered your work with Alltio — or you're considering it. A fair question follows: your file never left your device, so how does an AI company ever find out that your work is on the record, that it has an owner, and that it has terms?

The answer is that discovery works in the opposite direction from what most people expect. AI companies don't stumble across a mark on your file. They ask the registry.

The fingerprint does the work

When you register, your browser computes a cryptographic fingerprint of your file — a SHA-256 hash. Think of it as the file's unique signature: an exact copy of a registered file gives the same fingerprint, and for text the published rule gives the same fingerprint across platforms, because the rule reduces every version of the text to the same characters. Your registration binds that fingerprint to three things: who you are, what your terms are, and when you said so.

Here is why that matters. Responsible AI companies run every file through a pipeline before using it. Computing a fingerprint is a trivial step in that pipeline — it costs almost nothing. So before a file enters a training set or a retrieval index, the AI company's systems fingerprint it and ask Alltio one question: is this on the record?

The registry gives one of three answers:

  1. Registered, available for licensing — here are the terms. The work can be licensed through the standard Alltio contract, and usage settles back to you.
  2. Registered, not available for AI use — you chose protection only. The AI company now knows, with a timestamp, that this work is off the table. Using it anyway is no longer an accident they can claim.
  3. Not registered — no record exists.

Notice what this means: your work doesn't need to carry any mark at all. An exact copy of your file, renamed or moved to someone else's server, produces the same fingerprint, and the fingerprint leads back to your record. A copy that has been edited, re-encoded, or reformatted produces a different fingerprint. The registration does not travel inside the file. Anyone holding an exact copy can compute its fingerprint and look it up.

And every time an AI company checks, that check itself becomes part of the audit trail. Under rules like the EU AI Act, AI companies must document how they handled rights. Querying the registry is how they prove they looked. That is why they have a reason to ask — checking is cheaper than defending.

One honest caveat: an exact fingerprint matches exact copies. If a work is heavily modified — a cropped image, a re-encoded video — the match can break. And this applies to your own revisions too: edit one sentence of a registered essay and the revised file has a different fingerprint. Matching for altered copies of media is on our roadmap; for your own revisions, the answer exists today — version deliberately, below.

Version deliberately

When you revise a registered work, that revision is a new registration — new fingerprint, new record, new timestamp. On the registration form, give the new version a label (“v2”) and paste the record ID of the version it replaces into the “Supersedes” field. The registry then does two things: the new record carries a verifiable back-link to the old one, and the old record is marked as superseded — visible to anyone who checks it, but never edited and never deleted.

This matters because silent edits break your own chain of proof. A registered record whose fingerprint no longer matches the published file looks, to a verifier, like a claim about a different work. A version chain does the opposite: it shows continuous, dated custody of the work through every revision — which is exactly what a dispute, a diligence check, or a licensing negotiation wants to see.

Telling crawlers before they knock: your site declaration

The fingerprint lookup finds a registered work wherever an exact copy of the file, or the same words, appear. But if you publish on your own website, you can go one step further and declare your registrations at the front door — so AI crawlers know the terms before they fetch a single file.

This is a small machine-readable file you place on your site (the same way robots.txt has worked for thirty years, and aligned with the emerging RSL standard for site-level licensing signals). It says, in effect:

The works on this site are registered with Alltio. Here are their registry IDs. Here is where the terms live.

This file is not available yet. The format is drafted, and the registry does not yet generate it for you. The fingerprint lookup and the page embed work without it.

Marking the page itself: the page embed

The site declaration, once available, will cover your domain at the front door. The page embed marks the individual page where a work is published — the level at which AI crawlers actually read content. It has two parts, and both are generated for you when you register:

  • The Alltio mark, a small image linked to the public record. Readers, editors, and opposing counsel can see the claim and open the record in one click.
  • A machine-readable block (JSON-LD, the same structured-data format search engines have parsed for a decade) carrying the record ID, the fingerprint, the version, and the link to the superseded record. Crawlers don't have to guess — the registration data is on the page, in the format they already read.

You paste it once per page, and replace it when you register a new version. The embed is an accelerant, not the source of truth — the registry record is what actually binds your claim — but it moves discovery to the exact place the work lives. How to mark your pages →

A public record you can point to

Every registered asset also gets a permanent public record page on alltio.com. It shows your ownership claim, your terms, and the timestamp — never the work itself, which we don't have and never store.

This gives you something concrete to cite today, before licensing revenue exists:

  • In a takedown or dispute: link to the record as timestamped evidence of your claim and your terms.
  • On your website: display a “Registered with Alltio” badge that links to your records — a visible signal, like the ® symbol, that your work has an owner who put it on the record.
  • In licensing conversations: send the link instead of an affidavit.

The four layers, together

LayerWhat it coversWho acts
Fingerprint lookupYour work anywhere — copied, renamed, re-hostedThe AI company checks before using
Page embedThe page where the work is publishedYou (paste once per page)
Site declaration (not yet available)Your whole website, at the front doorYou (one file, once)
Public record & badgeDisputes, takedowns, deals, deterrenceYou cite it; anyone can verify it

You do the first once (registering), the second once per page, and the third once per site when it becomes available. The fourth is simply there when you need it.

The pattern is a century old. Songwriters never labeled every radio broadcast of their music; they registered once with a collecting society, and the infrastructure did the matching, metering, and paying. Alltio is that infrastructure for the AI economy — the registry answers, so your work doesn't have to.

The only step that matters is the first one: the record has to exist before anyone can check it.

Technical readers can review the Registry Lookup API reference.

Put your knowledge on the record.

Free. Read and fingerprinted in your browser, not stored.

Create your record