What AI needs, and how Apparat provides it.

Apparat answers a need that’s more and more vital for organizations: control the input AI agents use to act, rather than try to influence the output they give.

An AI engine answering a question about an organization works with whatever it can find. If the information is scattered, unstructured, or unattributed, it has no way to tell a good answer from a bad one, a correct figure and an invented one arrive looking the same. Apparat creates a data layer that allows machines to confidently fetch and read information structured for their consumption. The complete process is both simple and thorough.

AApparat uses the organization’s public data. Results releases, filings, registrations, product and programme documents. The information exists; it is just not easily readable by machines.
BApparat renders that data machine-readable. Each one drawn from a named document, quoted word for word, dated, and sealed so any later change is detectable.
CThe data points are gathered in a register, published under the organization’s own domain. Static files served under its own name, with nothing installed and nothing running behind them.
DAI agents find the register, read it, and can verify that the information has not been modified. Structured for machines to consume, and built so anyone can confirm a fact against its source.

The rest of this page sets out each of those steps, and what a register does and does not claim.

1 · What’s a machine-readable fact?

Apparat translates dense corporate documents (administrative statements, financial reports, product presentations, programmes, press releases and so on), where information is presented in prose, into registers, with one line per fact or figure.

Source documents
6
the company’s own filings, releases and official registrations
Register
426
sourced, sealed facts

Dense documents, read source by source, become a complete set of precise, individually sourced facts. This is one of them, taken from a real register.

In the source

An official record, the company’s BODACC registration filing, stating, in French: “Forme : Société par actions simplifiée (à associé unique).”

In the register

  "label" "Forme juridique de Believe"
  "value" "Société par actions simplifiée…"
  "time_period" "2025-12-10"
  "context" "Forme : Société par actions…"
  "source_url" "…bodacc.fr/…/BODACC_B_…_01435.pdf"
  "type" "REGULATORY_IDENTITY"
  "seal" "e169afd399…"
Label
"label"
Names what the fact is, so a machine reads it directly rather than inferring it from surrounding prose.
Value
"value"
The fact itself, carried in the language of its source, so what is sealed is identical to what was filed.
Context
"context" · "time_period"
The exact passage it was drawn from, and the date it holds for.
Source
"source_url"
The authoritative document it came from. For a company’s own registration, the filing is the fact of record; there is no truer version elsewhere.
Seal
"seal"
A fingerprint anyone can recompute to confirm the fact is unchanged from its source, against a public log, without asking us.

↑ Top

2 · What’s a register?

Data is consolidated into a Data Fidelity Register (DFR), an artifact designed to provide AI agents with the three things they need to fetch and properly read contents on the web. To use a fact well, a machine needs three things of it, and it needs all three at once.

Readable

The fact is a structured, typed statement, published in the forms machines are trained to consume, not a figure buried in a sentence.

Findable

A sitemap and index tell a machine what is there; the register is served on the organization’s own domain and bound to it, so the machine finds it and can see whose it is. Discoverable, and provably the organization’s own.

Checkable

Every fact is sealed, so anyone can confirm it is unchanged from its source, without asking Apparat. What the seal proves is fidelity to the source; for a company’s own filed facts the source is the fact of record, so that is as much truth as exists.

Structure alone, discoverability alone, verification alone, each exists already, in separate tools. A register is the only thing that puts all three on the same fact.

↑ Top

3 · The benefits of Apparat

The authoritative version exists, and it is reachable.

Every fact that defines an organization, structured and precise, on its own domain, in the form machines read. Not a version assembled from stale pages, the one the organization actually published.

The record is curated, not scraped.

Each fact is drawn from a named source and verified against it. What a source does not support is left out. A register is a considered account of what an organization has published, not a dump of everything it ever wrote.

Anyone can check it, without asking.

Each fact is sealed, so that what was published, where it came from, and whether it has stayed unchanged can be confirmed by anyone, at any time, against a public log.

↑ Top

4 · How to implement Apparat

Putting a register in place is light, and most of the work is Apparat’s. On the organisation’s side it is one technical step, done once, by whoever manages the website: attaching the register to a web address that belongs to the organisation.

Nothing is installed in the organization’s systems, nothing has to keep running, and there is no dependency on Apparat. A register is files, not software.

↑ Top

5 · Results

The major AI crawlers, operated by OpenAI, Anthropic, Google and Microsoft, already fetch these registers, and the volume is growing.

Among them are the agents that read pages live to answer a user’s question, not only the crawlers that gather data in the background. The register is reaching the systems it was built for.

What we observe is that when an organization’s facts are structured and sourced this way, engines answer questions about it more accurately. Apparat does not, and cannot, control what an AI says, no one can. What it can do is make the authoritative version exist, reachable and machine-readable, so the right answer is there to be found.

We measure the difference with a controlled benchmark, and we run it on an organization’s own facts.

↑ Top

6 · What a register does, and does not

A register takes the facts an organization has published about itself, binds each one to the exact source it came from, quoting the passage verbatim, and seals it cryptographically so that any later change is detectable. The register is published on the organization’s own domain, as static files it controls, and anyone can confirm, without having to trust Apparat or anyone else, that a fact is unchanged since it was published and is genuinely the organization’s own.

What a register does not do is claim that those facts are true. Whether a figure is correct, whether a company really did what it says it did, is something only the organization itself and its auditors or regulators can establish. Apparat does not pretend otherwise. Its role is narrower, and more honest: to preserve exactly what was published, to prove it has not been altered since, and to show you the source and the exact quoted passage so that you can check the fact for yourself. We do not ask you to take our word for anything. We make it as easy as possible to check us, and to catch us if we are wrong.

For the facts that define a company, its legal form, its registration number, its filed results, the authoritative document is the fact of record: there is no truer version held somewhere else that an AI could find by looking harder. For what a company merely asserts about itself, the register shows you precisely what was claimed, by whom, and when, and leaves the judgment where it belongs, with you.

Three things, kept separate on purpose. The cryptography secures the record: it proves the fact is unchanged and provably the organization’s own. The source link establishes fidelity: it lets anyone confirm the fact faithfully reflects the document it was drawn from. The curation is deciding what goes into the register, and checking each fact in the document it came from before it is sealed. None of the three pretends to be the others, and none of them claims to be a proof of truth. That distinction is deliberate, and it is what makes the guarantee an honest one.

↑ Top

Apparat builds and operates Data Fidelity Registers for organizations. The place to start is a short conversation about what a register would cover.

Start a conversation