How it works

Ask an AI, get an answer. Ask OwnedBy, get a record.

A language model on its own produces a plausible sentence with nothing behind it. It was trained months ago, it cannot cite anything, and it sounds identical whether it knows or is guessing.

OwnedBy uses language models the way a newsroom uses researchers: they orient, they plan, and they read. But nothing is published without an admissible public source attached to it. Every link in the chain carries its receipt.

An example trail

AU
A brandBrand or productStart of the trail. Australia.
operated by
AU
Its operating companyPrivate companyLikely link, not yet backed by a document. Australia.
owned by
BR
The listed parent above itPublic companyLikely link, not yet backed by a document. Brazil.

Countries

AU → AU → BR · 1 border crossing

Every link starts stippled — a reading, not a proof — and sets into an engraved rule as a public source backs it. The bistre tick marks a border being crossed.

01 · Two speeds

Two speeds, honestly labelled

You get a preliminary reading within seconds. It is drawn in pencil: the chain is stippled, the verdict is capped, and nothing on screen pretends it is verified. It is a reading, not a proof, and it looks like one.

Behind it, the evidence lane works primary sources — legal-entity identifiers, corporate registries, regulatory filings, companies’ own published documents. As each link is supported it sets: the rule converts from stippled to engraved, top down, one segment at a time.

The interface never claims more certainty than the record holds. The confidence ladder is a product guarantee, not decoration.

02 · Accuracy

Never confidently wrong

A comparison between a chatbot answer and an OwnedBy record.
Aspect	A chatbot answer	An OwnedBy record
Sources cited	A chatbot answerNone. The answer arrives with nothing behind it.	An OwnedBy recordEvery link carries its own, one tap away.
Freshness	A chatbot answerFixed at the model’s training cutoff.	An OwnedBy recordResearched on demand, and again when it is asked for again.
When it is unsure	A chatbot answerAnswers anyway, in the same confident voice.	An OwnedBy recordSays so. A withheld answer is still an answer.
When the world changes	A chatbot answerNothing changes until the next training run.	An OwnedBy recordThe record is revised, and the earlier version is kept.
The same answer for everyone	A chatbot answerTwo people can get two different answers.	An OwnedBy recordOne shared record, with its provenance attached.
Accountability	A chatbot answerNothing to flag, and nobody to flag it to.	An OwnedBy recordEvery record carries a way to flag an error on it.

Every release is gated on a blind evaluation corpus before it ships, with zero tolerance for two categories of error: a resolution that is wrong, and an identity that is wrong. Those are the two ways a record can be actively harmful rather than merely incomplete.

When the engine cannot verify something, it abstains and says so. We take precision over recall: a withheld answer is still an answer, and “ownership isn’t publicly disclosed for this one” is a fact about the world rather than a failure.

Records are append-only and bitemporal: we record both when a fact was true in the world and when we learned it. A record therefore cannot silently forget what it said yesterday.

03 · Scale

A registry that researches itself

Tracing one brand’s ownership properly takes a human analyst hours, sometimes days: registries, filings, cross-referencing, dead ends. Multiply that by every brand on every shelf and you get careers of analyst time that nobody will ever fund.

The pipeline turns that work into a compounding public asset. A brand is researched once, then served to everyone who looks it up afterwards, and revised when the world changes. Every lookup leaves something behind.

Language models made research cheap. Provenance is what makes it trustworthy — verification is the scarce resource now, and that is the part we industrialised.

A note from the founder

Why I built this

I’m a Canadian living abroad. Watching the conversation about trade from a distance, I kept noticing that it runs on feelings and flags while the material facts — who actually owns what, where the profit finally lands — are out of reach for the person standing at the shelf.

Where profits flow is a view consumers were never given. It isn’t the whole story of what a company means to a place — manufacturing, jobs, identity and community all count, and a foreign-owned plant can carry a town. But ownership was the missing piece, the one nobody could check. Data like this is a form of empowerment: not a verdict on what to buy, just a fact you were owed.

What changed is that the work finally became possible. The registry reading, the filing cross-checks, the verification that takes an analyst days per brand — agentic AI can now do it for very nearly any brand anyone cares to ask about. So I built it.

04 · What’s next

Ownership as infrastructure

What comes next is a public API: the same records, the same provenance, available programmatically. Everything described above exists today; this section is direction, and we are not putting a date on it.

The point of an API is what other people do with it. Organisations plugging the records straight into their own work: research, tools that hand customers the same fact a shopper gets, business intelligence, and an honest read on how people perceive a brand’s ownership as against who actually owns it.

If you are building something that rests on ownership transparency, that is what we want to make possible.

Look up a brand.

Open search