Your business is a row in someone else's dataset
A fifty-member consortium is assembling an open base of places to ground AI. Reddit and Yelp book revenue from licensing what sits on their platforms. Your business is not paid for any of it - but entering those corpora costs nothing.
50 members
in the Overture Maps Foundation as of 13 July 2026, assembling an open base of places to ground AI
In this article
Short answer: what the assistant knows about you does not come from you. It comes from bases other people hold, sell or pool. You are not the customer of that layer, you are its raw material. That is unpleasant to read and it is verifiable in public filings. The good news sits in the same place: those bases accept your corrections, for free, and most businesses have never made one.
43 million dollars
of Reddit other revenue in the second quarter of 2026, up 24%
plus 98%
growth in Yelp's other revenue, the line that carries its licensing, in the same quarter
Your address, your opening hours and your reviews have become a product. It is traded between platforms, over your head, and nobody asks you.
A consortium formed to ground AI in the physical world
On 13 July 2026 the Overture Maps Foundation announced it had reached 50 members, nearly double its 2024 count. It was founded by Amazon Web Services, Meta, Microsoft and TomTom. Its object is an open base of places, and its executive director states the reason plainly: AI cannot reason about the physical world without an open, verifiable foundation beneath it.
The technical detail carries the consequence. The foundation gives every place a stable identifier, so that pipelines can anchor their outputs to real-world entities instead of probabilistic guesses. And the material comes from the members themselves: Meta, TomTom, Tripadvisor and Uber contribute live signals - pickups and drop-offs, ratings and reviews, foot traffic, interactions at physical locations. The June 2026 release refreshed the operating-status field of US points of interest from those aggregated, anonymised signals.
Read that sentence from behind your counter: a foundation of companies decides, from signals they collect, whether your business is still open. You signed nothing and nobody consulted you.
What that material is worth, and who books it
The price is in the accounts. In the second quarter of 2026 Reddit reported revenue of 805 million dollars, of which 762 million in advertising and 43 million of other revenue, up 24%. Its quarterly report notes that the company also earns revenue from content licensing and products sold directly to users. The release does not break that line down and names no buyer: we report what is written there, not what the press says about it.
At Yelp, in the same quarter, revenue grew 1% to 376 million dollars, while the other-revenue line, which carries licensing, jumped 98% to 33 million. There, the release does name the buyer: Yelp's content is powering local discovery for ChatGPT and other AI partners.
Two companies, two accounting lines, one movement: text that people wrote and listings that shopkeepers filled in have become a tradable asset. That market exists, it is quantified, and the local business it describes takes none of it.
Entry costs nothing, and it is machine-readable
We wanted to check what a machine can learn about a business without ever contacting it. On 20 September 2026, with a single script and no money, we read 3,461 distinct US businesses out of OpenStreetMap: name, city, trade and website. What took us an afternoon, any system does continuously.
That is the point that turns the reading around. The corpora deciding your digital existence are open to read and open to correct, and entry is free. A claimed business profile, correct opening hours, one address everywhere, a site readable without JavaScript: these are the only levers you hold, and none of them is billed.
| Corpus | Who holds it | What entry costs | What it feeds |
|---|---|---|---|
| OpenStreetMap | A community of contributors, under an open licence | Nothing | Maps, apps, and anything able to read an open base |
| Overture Places | A 50-member foundation, founded by AWS, Meta, Microsoft and TomTom | Nothing to read it | Pipelines that want to anchor an answer to a real place |
| Google Business Profile | Nothing, but you must claim it | Search, maps, and the answers built on them | |
| Apple Business | Apple | Nothing, but you must claim it | Maps, Siri and Apple's other surfaces |
| Yelp | Yelp | Nothing for a listing | Its own audience, and under licence ChatGPT's local discovery |
Why this matters now
Because the machine no longer ranks links, it says names. In our US study of 13 September 2026, 89% of the 872 ChatGPT and Gemini answers name at least one business, 5.2 names on average. An answer that names is an answer that read something. The question is no longer how to rank, it is how to be accurate and legible in what was read.
What this reading does not prove
That assistants read these particular corpora.
None publishes its source list. We show a market in place data being built and billed, not a pipe whose blueprint we have seen.
That fixing your listing will get you named.
We did not measure that link. A correct listing is a condition of legibility, not a promise of recommendation.
That Reddit's other revenue is AI licensing.
The release does not say so, and the quarterly report does not break the line down. We stay with the words in the documents.
That free entry means free work.
Claiming, checking and maintaining four or five listings takes time. It is not an expense, it is attention, and for a sole trader that has a real cost.
What it changes for a local business
Claim what is already written about you.
As long as nobody claims a listing, third-party sources fill it in. Silence does not protect you, it lets others write.
Make the same information identical everywhere.
Opening hours that differ between two bases are not a detail: they are a contradictory signal given to a machine that has to choose.
Treat OpenStreetMap as a public address.
It is an open base, correctable by you, and read by systems you will never choose.
Do not pay for what is free.
No provider can sell you a place in these corpora. They can sell you time and method, which is a different thing.