The Quiet Google File That Wants to Replace the Map of the Web

A new open standard born for company databases has slipped onto the public internet — and a handful of businesses are already betting it will decide how artificial intelligence understands them.

The most consequential thing Google shipped this summer did not come with a keynote, a product name you would recognise, or a price. It arrived as a short document on a developer blog and a folder of plain text files on GitHub. You could read the entire specification over a cup of coffee.

And yet, in the small, fast-moving world of people whose job is to make websites legible to machines, the Open Knowledge Format has set off the kind of argument that tends to precede a shift. The question being asked is deceptively simple, and its answer matters to anyone who has ever cared whether their business shows up when someone goes looking: as artificial intelligence replaces the search box, what will the machines actually read?

For twenty years, the answer was the sitemap — the humble file that hands a search engine a list of every page on a site. It was the closest thing the web had to an official map. The Open Knowledge Format, or OKF, was not built to replace it. But it does something the sitemap never could, and that difference is why a Fort Worth law firm, among others, decided not to wait for permission.

An early bet, placed in public

In late June, the Fort Worth-based firm Varghese Summersett published a structured knowledge graph for VersusTexas.com — an OKF bundle laying out, in machine-readable form, who its attorneys are, where its offices sit, and what kinds of cases it handles. It went up within weeks of Google releasing the format, an unusually fast move for a standard that, by its own author’s admission, is still version 0.1.

Benson Varghese, the firm’s managing partner, is candid that this is a calculated move by someone who has watched Google champion the next big thing before. “When Google says something, you listen — and then you make your own decision about whether to implement,” he said. “None of us has forgotten being pushed to put everything in AMP for the mobile web, and that went nowhere. But when you look at what consumers now expect and what these models can actually ingest, giving machines something they can read more easily seems like a sound step — or at the very least, one without any real drawback.”

The reasoning he describes is spreading quietly through marketing departments and engineering teams alike. When a prospective client no longer types a query into a search engine but asks an AI assistant — who handles federal cases in Fort Worth? — The assistant has to get its answer from somewhere. Left to its own devices, it scrapes whatever it can find and guesses at how the pieces fit. A firm that publishes a clear, linked map of itself is no longer guessing along with it.

That is the wager OKF invites every business to make. And to understand why it is more than a technical curiosity, you have to understand what the format actually changes.

What it is, in plain terms

Strip away the jargon, and OKF is almost aggressively simple. It is not an app or a service you pay for. It is a set of conventions for writing down what an organisation knows as a directory of ordinary text files — the same Markdown format that powers GitHub, Reddit, and the chat windows of the AI tools themselves.

Each file describes one concept: a person, an office, a product, a metric, a process. A few lines of structured tags sit at the top — a title, a description, a type — and a plain-language explanation follows below. Google sums up a bundle in three phrases: just Markdown, just files, just YAML. There is no software to install and no vendor to sign up with.

The feature that matters, the one that separates OKF from a folder of disconnected documents, is the link. When one concept points to another, those links turn the directory into a graph — a web of relationships a machine can walk. It learns not only that a thing exists, but how it connects to everything around it: that this attorney works in that office, that this practice area sits beneath that broader one.

This is the precise thing a sitemap cannot do. A sitemap is a list. It says a page exists. It says nothing about what the page means or how it relates to any other. That was the right tool for an era when a machine’s only job was to find documents and rank them. It is the wrong tool for an era when a machine is expected to understand.

Why now

The timing is not an accident. As the artificial-intelligence systems built on large language models have grown more capable, the thing holding them back has quietly changed. The bottleneck is no longer raw intelligence. It is context — whether the machine has the specific, current, correct information about the subject in front of it.

A model that does not know how a particular company defines a term, or which of its locations does what, will not stay silent. It will answer anyway, confidently and incorrectly. Every business that has watched an AI assistant describe it inaccurately has felt the edge of this problem. OKF is one answer to it: a portable way to hand a machine the truth about yourself, written once, owned by you, readable by any system that comes along.

The idea did not spring fully formed from Google. It formalises a pattern that engineers and researchers kept reinventing on their own — what the prominent AI researcher Andrej Karpathy called the “LLM wiki,” a knowledge base an AI reads and tends the way a programmer tends code. As Karpathy put it, the machines do not get bored and do not forget to update a cross-reference — exactly the bookkeeping that causes humans to abandon their own wikis. The same shape had been surfacing for a year under a dozen different names, each one bespoke, none built to cooperate. What Google added was an agreement: a common set of rules so that knowledge written by one team could be read by another team’s machine without translation.

The argument over what it means

Here the story turns, because not everyone believes OKF will become the map of the agentic web — and the people closest to it are the most careful.

Google built OKF for internal knowledge: the meaning of a database column, the steps in an emergency runbook, the institutional memory locked inside a few senior employees’ heads. Its published examples are datasets, not marketing pages. Nothing in the announcement mentions public websites or search at all. The businesses now publishing bundles for the open internet are using the format off-label, betting on a future that has not arrived.

And it may not. By Google’s own description, OKF v0.1 is “a starting point, not a finished standard.” The reference tools the company released are explicitly labelled proofs of concept. Most decisively: as of today, essentially no AI agent actually fetches and reads these public bundles. The audience that the early adopters are writing for does not yet exist.

Even Google does not speak with one voice. The same company whose data team published OKF has, on its search side, dismissed a related machine-readable proposal as “purely speculative” for ranking — while its Chrome team quietly added a check for it to a developer audit tool. The right hand and the left are placing different bets. That internal disagreement is itself a fair measure of how unsettled this all is — and it is exactly the history that makes practitioners like Varghese hedge. The AMP standard Google evangelised for the mobile web absorbed enormous effort across the industry before fading; no one who lived through it signs on to the next initiative without asking what happens if this one fades too.

Why the smart money writes one anyway

And still — the businesses moving early are not naïve. They are reading an asymmetry that is genuinely hard to argue with.

Writing an OKF bundle costs almost nothing. There is no infrastructure to buy, no contract to sign, no platform to be trapped inside. If the standard takes hold — as the connective tissue of an AI-first web, or simply as a cleaner way to feed a company’s own internal assistants — the firms that learned the language first will be fluent while their competitors are still reading the manual. And if it never takes hold, the downside is a tidy, version-controlled, human-readable account of exactly what a business is and how its parts fit together. Which, as more than one early adopter has noticed, turns out to be valuable on its own. The simple act of writing it forces a company to say plainly what it knows — and reveals, in the gaps, what it only thought it knew.

That is the quiet genius of the thing. It asks for very little and, win or lose, leaves you with something worth having.

The map and the meaning

So, is the Open Knowledge Format the new sitemap? Not exactly — and the difference is the whole story. The sitemap told the machines where your pages were. OKF is an attempt to tell them what your business means. One is a list of doors. The other is a description of the rooms, how they connect, and who is inside.

Whether the automated readers of the next decade come to lean on that description the way an earlier generation leaned on the sitemap is the open question now being settled, file by file, in public repositories like the one a Texas law firm posted in late June. The answer will not be announced. It will simply, one day, have become true — and the businesses that wrote their map early will have spent the wait already on it.

For now, the format sits where all genuinely new standards begin: too plain to look revolutionary, too useful to ignore, and waiting to find out whether the world will read it.