machinewitness

Services & fees

AI crawling record

Two halves of the same day from one archive: what your domain told crawlers, and what the crawler operators and AI providers published on that same day.

If this is the first product page you have opened: this archive has recorded, every day since 22 July 2026, what around 128,000 EU domains serve to machines, sealing each day so nobody can alter it afterwards. This product puts two halves of the same day side by side, what a domain reserved, and what the crawler operators themselves declared, from one neutral source instead of two disputed ones.

Where this is used

Case 1A reservation, and what the other side published on the same day

A picture agency has carried a reservation against text and data mining for months. Its images turn up in the answers of an AI system. The agency can show its own file, and it will be told that the file is its own account of events; and it has nothing at all about what the other side was declaring at the time.

This archive observes both sides of that day, for everyone alike: the reservation as the domain served it, and the crawler documentation, address lists and terms that operators and providers published themselves. The extract sets the two next to each other, day by day, and draws no conclusion.

Copyright · civil proceedings · rights-holder

Case 2A course of events over months, not a single date

A single date invites the answer that it was an exception, a misconfiguration, a version that was corrected the next day. A sequence of days does not, and it is the sequence that is almost never preserved at the time.

Where the domain has been under observation for a while, the extract can cover up to twelve months: for each day the reservation as served, and for each day the published state on the other side. What follows from the pattern is not our subject.

Copyright · pre-litigation · rights-holder

Case 3The other way round: an operator shows what was declared when it crawled

The same file serves the other side of the table. An operator or model provider that is accused of having ignored a reservation has the same interest in a dated record: what did that domain actually declare on the day in question, and what did we ourselves publish about our crawler at the time?

The answer comes from the same archive, at the same published fee, with no need to explain why. Both sides receive the same document, and neither is told that the other asked.

Copyright and IT law · defence · operator or provider

Two halves of the same day: what your domain told crawlers, and what the crawler operators published while it did. YOUR SIDE What your domain declared robots.txt, ai.txt, TDM reservation, the header that carries it. THEIR SIDE What operators published Their identities, their ranges, the terms they declared that day. SAME NIGHT Both sides in one seal Sealed together, so the two can be read against each other later. YOU HOLD Both halves, same day Priced from the published fees of the parts it is built from.
Two halves of the same day: what your domain told crawlers, and what the crawler operators published while it did.

What you can use it for

  1. Domain observed dailyYour domain is retrieved and sealed by both witnesses, every day.
  2. Operators and providers observed tooTheir own published statements are observed on that same day.
  3. Sealed into one rootBoth records are sealed into the same day's root.
  4. Set side by sideThe extract sets them next to each other, and draws no conclusion.

Two halves of the same day, sealed into the same root and set side by side in the extract.

  • What did our reservation say on that day, word for word, in the files a machine reads?
  • What did the crawler operators and AI providers publish on that same day, in their own documentation and terms?
  • For how long was each of those two states the case?
  • What is not in this, and where would it have to come from?

Typical use: filed as an annex together with the extract of the domain itself; where the method is disputed rather than the content, the technical procedure statement follows.

What is witnessed

Two records, both already part of what this archive stores, drawn together for the same day:

  • Your own domain's reservation, the daily observation described under observation on request, sealed and anchored like every other domain in the core.
  • The crawler operators' and AI providers' own published statements for that same day: their crawler documentation, address lists and terms. See what we store for the full field schema.

What this does not contain

There is a third half to this question, and we do not have it: whether a given crawler actually came. This archive records what was published, on both sides. It does not record visits to your server. Where you need that, there are two paths, and neither is this product: an access record at a declared witness host, which is available on request, or your own server logs, which are your own records and remain your own case to make.

We do not establish an infringement, we do not name any company as an infringer, and we notify no one. The operator and provider statements are an object of observation here, nothing else; they are listed, with their number and their dates, on what we store (13 crawler operators, 12 AI providers as at 17 September 2026).

Diagram: a captured file is hashed (SHA-256), the hash is placed on a Merkle path to that day's root, the root is checked against the public log, and outside anchors confirm the date, independently of MachineWitness.
The chain behind “sealed,” explained in the block below.

What sealed means here

This block is the same on every page of this site, and it is repeated on purpose: it is the part you need in order to judge everything else.

  1. One fingerprint for the whole day. Every observation made that day, yours among hundreds of thousands, is reduced to a single hash through a Merkle tree. One changed byte anywhere in that day, and the fingerprint no longer matches. There is no version that could be quietly corrected.
  2. Published where anyone can see it. The fingerprint goes into the public log the same night, under a fixed, citable URL, together with the instructions for recomputing it.
  3. Handed out of the house three times on the same day. An RFC 3161 time-stamp service, a decentralised OpenTimestamps anchor in the Bitcoin blockchain, and a qualified eIDAS time-stamp from GLOBALTRUST (e-commerce monitoring GmbH, Austria), a qualified trust service provider listed on the EU Trusted List. The third of these is paid for and supervised, and it is worth saying so plainly: only the qualified time-stamp carries the presumption laid down in Article 41(2) eIDAS. A free anchor establishes that the data existed and has not changed, but it carries no presumption laid down by law.
  4. Twice over, by two witnesses that cannot write to each other. Two machines at two providers in two countries, with separate keys. Each seals its own day and takes its own anchors.

In plain words, two sentences. We cannot change a byte afterwards, because the day's fingerprint would no longer match. And we cannot backdate one, because that fingerprint has been in other people's hands since the night it was made. No one has to believe us: every step can be repeated with standard tools.

Further: how it works · what we store · glossary.

Does this stand up in court?

We cannot promise that, and nobody can promise it honestly: what a court accepts is for the court to decide. What we can tell you is what the document is made of, and each of the four facts below can be checked before you buy anything.

  1. It comes from a third party, not from you. Not your screenshot, not your server log. The bytes were fetched and stored by a third party that did not know your matter existed, on a day chosen by the calendar and not by the case. That is the difference between a record and an account of events.
  2. A presumption laid down by law. Every sealed day since 31 July 2026 carries a qualified electronic time-stamp from a qualified trust service provider on the EU Trusted List. Under Article 41(2) of Regulation (EU) No 910/2014 (eIDAS), such a time-stamp enjoys a presumption of the accuracy of the date and time it indicates and of the integrity of the data it is linked to. Article 41(1) says something much narrower, namely that a time-stamp may not be denied legal effect merely because it is electronic. The two paragraphs are routinely confused; the one that matters here is the second.
  3. Two independent witnesses. Two machines, two providers, two countries, separate keys. Each one seals its own day and anchors it externally on its own. Neither can write to the other, so neither can be corrected to match the other after the fact.
  4. Verifiable without us. An appointed expert repeats every step with standard tools: recompute the hash of the file, rebuild the path from that hash to the day's root, check the root against the public log and against the external anchors. We do not have to be believed, and that is the point of the whole construction.

What follows from this in your particular matter is for your lawyer to say. We do not advise, do not rate and do not take a side, and the other side can order the same document on the same published terms. That is not a weakness of the document. It is the reason it is worth anything.

How to order

  1. By e-mail to contact@machinewitness.eu, or through the form on this site. Name the domain, and your billing address with a VAT number if you have one. We do not ask what the matter is.
  2. We confirm in writing what we received and name the fee for your request. Nothing is charged before you have that in writing.
  3. You receive a payment link.
  4. Delivery normally within five working days of a complete request; if a court deadline is running, say so when you order and we handle it within 48–72 hours instead, ahead of the normal queue.
  5. As a combination of two already published services: the domain's admission is entered, with its date, in the public record as with a standing observation, and each extract you order afterwards is delivered as PDF and ZIP as with an evidence extract, with the operators' and providers' state for that day filed as one further kind of file.

We answer in German, English or Spanish.

Fee

This is a combination of services that are already published, not a separate item: the standing observation of your domain (150 €, one-off), the evidence extract (450 €), and each further kind of file in the same request (150 €), where the operators' and providers' state for that day counts as one further kind of file. Each of these amounts is in the fee table; we quote them individually and you will not find a package price, because there is none.

What you receive

The same evidence extract as for any other request: a PDF with a ZIP alongside it, with the operators' and providers' published state for that day filed as one further kind of file next to your own domain's day.

The method sheet is included. Every extract comes with the general method sheet at no extra charge, including the one covering a single day. It describes how this archive observes, seals and anchors, and it carries a version and a date. It is not written for your matter, and it does not have to be: it is the same for everyone, which is precisely what makes it checkable.

The technical procedure statement is a different document, written for this particular observation, signed and addressed to a court or an appointed expert. You do not need it in order to file the extract. It becomes relevant when the other side disputes the method rather than the content. See the fee table.

What becomes public, and what does not

Visible to anyone

  • that the domain or URL is observed, and from which date
  • the dated entry in the public record of admissions
  • the daily roots and anchors, as for every other observation

Never published

  • who applied, in no document
  • why: we do not ask what the matter is
  • what the files said: content is issued only as an extract, at the published fee, to anyone

The other side can see that the URL is observed, and since when. That is the price of a witness that belongs to no party, and it belongs here, before the purchase, not in the small print. Where a look-alike domain is observed as a precaution, an opponent may infer that someone is preparing. Whoever does not want that buys a capture of a single day instead of a standing observation.

Neutrality. This archive records; it does not rate, rank or advise. Three conditions hold for everything on this page: it is visible to everyone in the same way; the fee is published and depends neither on who asks nor on how a matter ends; and whoever pays receives nothing a third party would not also receive, which means no notification, no mention as the applicant, no priority, and no content without an extract that anyone else could order too.

Further: how it works · evidence extract · what we store · terms.