An Open Letter for Your Attention

I suddenly realized there is a game being played for control of how humanity accesses information. Very few, if any, even know the game is being played, and I don't think anyone fully comprehends the rules. I hope I'm wrong, but here is the game.

The three layers

To control what people know, you need three things.

You need a browser — the software people use to move through the internet. The browser watches everything. Every search, every click, every page, every moment of hesitation. It is the instrument that measures human attention.

You need a search index — a map of the entire web, updated constantly.

You need an AI — the layer that sits on top of the index and the browser signal and answers questions directly, without showing its work, without sending you anywhere, without letting you judge for yourself.

Own all three and you own the loop. The browser watches what people do. That behavior flows into the index. The index tells the AI what humans find useful. The AI gives better answers. Better answers attract more users. More users generate more signal. The loop closes. No one outside can audit it, replicate it, or offer a genuine alternative.

One company owns this loop for most of the world. Its name is Google.

Google's Chrome browser runs on 65% of the world's devices. Google Search handles 90% of queries outside China and Russia. Gemini sits on top of both. One company. One loop. Three billion people.

Baidu owns the same loop in China. Yandex owns it in Russia.

Everywhere on earth, the loop is closing. And the people whose attention built it own none of it.

What everybody gets wrong about Google's competitors

OpenAI. Anthropic. xAI. Everyone calls them Google's competitors in AI.

They are not Google's competitors. They are Google's customers.

OpenAI has a model. It is frantically building a search index because it has realized it doesn't have one. It has no browser. Anthropic has a model, rents its live search from an independent index it doesn't own, and is partly funded by Google itself. xAI has a model and a social media platform. None of them have the browser. None of them have the closed loop.

Without the browser you cannot watch the session. Without the session you cannot tell whether your AI actually helped someone or just sounded convincing. You can build a model that is articulate, confident, and wrong in ways you will never detect — because you have no instrument to measure whether it served a human being.

This is not a search monopoly. Regulators have spent years prosecuting the wrong case. A 90% search market share is the symptom. The disease is that Google owns the instrument of measurement itself — and has structured three products to ensure no one else ever can.

Here is a thing worth understanding about every "alternative" search engine that doesn't own a browser: it is not actually a search engine. It is a wrapper. DuckDuckGo is Bing with a privacy label. Perplexity is Brave's index with an AI on top. You.com is Bing again. This is not a criticism of those products. It is a structural fact. Without the browser layer, you cannot collect the session signal. Without the session signal, you cannot close the feedback loop. Without the closed loop, your index never improves from human attention — it only improves from whatever the company whose index you're renting decides to share with you. Every search product that doesn't own a browser is permanently, architecturally, a tenant. The rent can go up. The lease can be cancelled. There is no path to independence that doesn't go through building a browser or saving one that already exists.

The Microsoft question

Microsoft has Edge. Microsoft has Bing. Microsoft has Copilot. Microsoft, by any structural accounting, has all three layers of the loop.

Microsoft has also failed to matter.

The standard explanation is brand loyalty. Google is a habit too deep to break. But there is a more interesting possibility: Microsoft may have built three separate products instead of one loop. Three teams, three roadmaps, three data pipelines — and a gap somewhere between the browser session signal, the search quality signal, and the model training signal that means the loop never actually closed.

Google didn't build three products. Google built one loop and gave it three interfaces.

If that's the diagnosis, the fix is not a new product. It is a pipeline. The browser data flows to search quality. The search quality flows to the model. The model improvement flows back to search. If Microsoft has never fully connected those pipes — and the evidence suggests it hasn't — then one of the most consequential software projects in the world right now is an org chart change and a data architecture review at a company that already has everything it needs.

Microsoft: you may have already won and not know it. Figure out the loop. Connect the pipes. The world needs you to be a real competitor here, not because Microsoft deserves to win, but because the alternative to Google winning is worse than most people have thought through.

What everybody gets wrong about breaking up Google

The antitrust case against Google is about search market share. That is the wrong case.

You cannot break Google up. Not because it's legally impossible — though that's hard enough — but because breaking it up destroys the thing that makes it valuable without redistributing that value to anyone. If you separate Search from Chrome from Gemini, you don't create four competitors. You create four companies none of which has the loop. The signal that took 25 years to build — a billion people's attention, compressed into an index that knows what humans find useful — doesn't get redistributed. It evaporates. And Baidu, which has its own intact loop and a government that will never permit it to be broken up, fills the vacuum.

So here is the question nobody wants to ask out loud: is Google the good outcome?

Not because Google is good. Because the realistic alternatives are worse. A government controlling what humanity finds true is not obviously better than a corporation doing it. At least a corporation can be regulated, sued, and is theoretically subject to competitive pressure. China winning the AI loop by default because America broke its own is not a win for human freedom — it is a different kind of loss, faster and harder to reverse.

Google is not the enemy. Unaccountable control of the loop is the enemy. Google currently holds that control. So does Baidu. So, in its way, does Yandex. The problem is not who holds it. The problem is that no one should hold it alone, unregulated, without a people-owned alternative existing somewhere in the world.

The right frame for Google is not a monopoly to be broken. It is infrastructure to be regulated — the way power grids are regulated, the way railroads were regulated, the way telephone networks were forced to carry competitors' traffic. Not broken up. Not government-owned. Required to provide fair access. Prohibited from using its position to foreclose the possibility of alternatives. Obligated to be a utility, not a sovereign.

That is a political fight worth having. It is also a long one. It does not help anyone today.

Privacy was the wrong fight

The opposition to this consolidation organized around privacy for twenty years. Don't track us. Block the cookies. The frame was tracking versus freedom.

That frame was wrong. It cost us.

The internet requires collective behavioral signal to function at quality. That has always been true. The pre-Google internet wasn't private — it was just bad. You found things through link lists emailed by friends. The signal that made Google useful was never the problem. Knowing which pages humans collectively found valuable is not surveillance. A road doesn't know who drives on it.

The problem is not that the signal exists. The problem is who owns it and what they do with it.

Tracking that serves the people being tracked is infrastructure. Tracking that profiles individuals to sell their attention to advertisers is something else entirely. Google does both. The first kind makes search work. The second kind funds everything else. Conflating them handed Google twenty years of cover while the real question — who owns the commons — went unasked.

The choice was never between tracking and no tracking. It was between signal that belongs to the people who generate it, and signal that belongs to whoever built the box.

What is actually being lost

For twenty-five years, something extraordinary happened every time someone used a search engine. They voted.

Not in any official sense. But a billion people clicking on results, staying on pages, going back to search again when they didn't find what they needed — that behavior was a signal. Aggregate human judgment about what was useful, what was trustworthy, what actually answered the question. It was the most valuable dataset in human history, and it belonged, in practice, to whoever owned the search box.

That signal is going away. Not because people stopped having opinions. Because people are stopping before they click. The AI answers the question before you get to the results. You never go to the page. The signal never forms.

And here is what that means in practice, right now, today — not in some hypothetical future.

Search quality is already degrading. Not just Google. Every search product. The deeper problem is that the signal driving search quality improvement is thinning. Fewer humans making unguided judgments about what was actually useful. More AI-generated content flooding the index. The gap between what exists on the web and what is trustworthy and findable is widening, and the instrument that used to close that gap — human attention, expressed as clicks — is being routed around before it can register.

Google may not have fully understood what it is doing to itself. The moment you answer a question inside the search results page — the moment you stop directing people to the web — you begin starving the signal that told you what a good answer looks like. You are eating your own seed corn. Slowly, then faster. Each generation of the model gets trained on what the last model said, not on what humans actually found useful when they navigated the real web. Each generation inherits the errors of the previous one, slightly more confident, slightly further from reality. This has a name in AI research. It is called model collapse. Applied to search, it is something larger: the collapse of the mechanism that has kept human knowledge organized and accessible for a generation.

This is not only an AI problem. It is not only a search problem. It is the slow-motion collapse of the infrastructure that connects humanity to its own accumulated knowledge. The web does not disappear. It becomes a dark forest — vast, largely unnavigable, increasingly unmaintained, because the economic signal that told publishers what was worth writing and told search engines what was worth indexing has been silently rerouted into a closed loop that serves the loop's owner.

There are three search ecosystems that cover most of humanity. Google. Baidu. Yandex. Each of them is converting its search monopoly into an AI monopoly using exactly this mechanism — they already have the signal, the users, and the loop. They just have to build the answers on top.

No version of this story, in any country, ends with the people owning the layer that tells them what is true.

The last independent wall — and why it is also in danger

There is one independent web index left outside the three closed ecosystems and Microsoft's loop-that-doesn't-close. One company that actually crawls the open internet on its own, that has built the search backbone most major AI systems now depend on for real-time information. It also has a browser. Small — under 1% of global sessions — but real, with the architecture to collect the kind of anonymized, aggregate attention signal that could feed an AI that belongs to no corporation and no state.

Its name is Brave.

Brave did not set out to be the last wall. It got there by building something honest while everything around it collapsed. Microsoft killed its public search API. Google sued the scrapers. When the dust settled, Brave was the only fully independent index standing, and every major AI that needs to know what happened in the world today depends on it.

And that dependence is killing the thing that made Brave worth depending on.

Every AI system that queries Brave's index is not a human. It clicks nothing. It stays on no page. It returns no signal about whether the result was useful. It generates revenue for Brave — and it displaces exactly the human attention signal that keeps an index honest. The more successful Brave becomes as the search backbone for the AI industry, the faster the human signal in its index thins, and the faster Brave's index becomes indistinguishable from any other AI-trained corpus drifting slowly toward confident mediocrity.

Brave is not immune to the collapse it was built to prevent. Serving as the search backbone for the AI industry while losing the human signal that grounds that index is not survival. It is becoming Google on a slower timeline. The index stays alive. The thing that made it worth trusting dies quietly underneath it.

Brave is also a company with 175 employees, under 1% browser market share, and the same economic gravity that has pulled every other independent web property toward the business model it was built to resist. Not out of bad faith. Out of the ordinary math of survival.

Five things that must happen

Move to Brave. Download the browser. Use the search. Every session is a human attention signal that flows into an independent index instead of Google's. It costs nothing. It takes ten minutes. If enough people do it, Brave survives with its signal intact. If Brave's signal dies, it doesn't matter that the index still runs.

Brave must not let AI queries contaminate its human signal index. This is not a call to turn away revenue. It is a call to keep two data streams separate. Human browser sessions inform the index. AI API queries do not. The engineering decision is straightforward. The consequences of not making it are not. Every AI query that becomes part of Brave's ranking signal displaces a human judgment about what was useful. Over time, an index trained on AI curiosity rather than human need becomes indistinguishable from the AI-generated slop it was built to filter out. Brave's value — to its users, to the AI companies paying for API access, to anyone who cares about a web that humans can navigate — depends entirely on the index remaining grounded in human attention. That is the asset. Protect it. The API revenue keeps the lights on. The human signal is what the lights are for.

Microsoft: connect the loop. You have the browser. You have the index. You have the AI. You may simply have never connected them into a single feedback loop the way Google has. The signal from Edge sessions should be closing the gap between Bing and Google's search quality. If it isn't — find out why. The world needs a second working loop. You might be one architecture decision away from providing it.

Anthropic and OpenAI: build the full stack or save what exists. Without a browser and an independent search index, both companies are one business decision away from total dependence on the entity they are supposed to be an alternative to. The rent can go up. The lease can be cancelled. You know this. Move.

Find a better signal. We do not yet know if information quality can be measured without human attention. If it can — if there is a way to evaluate whether an answer actually served a human need without watching humans navigate — the browser dependency breaks entirely. This may be the most important open research problem in information science. It should be funded openly, answered openly, and owned by no one.

A suggestion

Form a cooperative around Brave and an AI — one of the existing frontier models, or a new one built for the purpose.

Give Brave the membership base and subscription revenue it needs to stay independent — to never have to choose between its principles and its survival. Build the full stack owned by the members who use it, governed by a single rule: this system exists to be useful to the people who fund it. Not to sell their attention. Not to answer to shareholders. To be useful.

The cooperative model is a century old. Mondragon, founded by a Basque priest after the Spanish Civil War, has 80,000 worker-owners. The Japanese consumer cooperative JCCU has 31 million members — a quarter of Japan's population. REI has been returning profits to its members since 1938. The model works. It has simply never been applied to the thing that matters most.

Everyone assumed OpenAI and Anthropic were Google's competitors. They were always building the model while Google owned the measurement. You cannot win a game you are not actually playing.

A cooperative built around Brave would be the first entity since Google — outside of authoritarian state control — to play the whole game. Not to beat Google. To exist alongside it, outside its walls, owned by the people it serves. So that the answer to "who controls what humanity knows" is not only Google, or Baidu, or a government. So that somewhere in the world, the answer is: the people.

Regulate Google as the infrastructure it is. Push Microsoft to close its loop. Fund the research that might make the loop unnecessary. And build the one thing that none of the above can give you — an alternative that answers to no one but its members.

If you can build this — build it.

If you can fund it — fund it.

If you can do neither — move to Brave. Tell someone why.

The window is open. It will not stay open.

This letter belongs to no one. Copy it. Fork it. Translate it. Post it wherever the right people will see it. The only thing that matters is that it lands.