What Is an AI Attention Exchange

An AI attention exchange is a marketplace that measures the attention people give while AI tools are working, checks that the attention was real and human, and sells it to advertisers, with a share of the revenue paid to the person who gave it. BackSpin is an open AI attention exchange built for developers: it runs during AI wait states in tools like Claude Code, Codex, Cursor and VS Code, and prices attention in quality-adjusted seconds instead of impressions or clicks.

That is the one-paragraph answer. The rest of this post explains why the category exists and how the pieces fit, using BackSpin's own numbers where they are public.

Why AI wait time became a market

Using an AI coding tool means waiting. You send a prompt and the model thinks, reads files, edits code or runs a command. Sometimes that is two seconds, sometimes it is a minute. During that time you are at your desk, looking at the screen, and mostly not doing anything else.

That is unusual attention. It is focused, it happens inside a work tool, and it repeats many times a day. Traditional ad formats do not fit it well, for three reasons:

• The user often cannot click. Clicking away mid-generation breaks your flow, so click-through rate says little about whether a message landed.
• Impressions are too cheap a unit. An impression counts a render. It does not care whether anyone was there.
• Bots are tempting. Anything that pays for wait time invites people to fake waits. Without verification the inventory is worthless.

An attention exchange is the response: measure the attention itself, verify it, and price that.

The building blocks

Attention windows

When a supported AI tool starts working, BackSpin opens an attention window and closes it when the tool is done. Each window carries content-free signals only: timing, whether the host app was focused, keyboard and mouse activity, and whether the discovery card was actually on screen. It never includes code, prompts or AI replies.

Quality-adjusted attention seconds

Not every second is worth the same. A window where the card was fully visible in a focused editor is worth more than one where the editor sat in the background. BackSpin scores each window with an Attention Quality Score and discounts the raw seconds by it:

weighted_attention = raw_seconds x attention_score / 100

The score combines activity, focus, on-screen visibility, session quality, workflow relevance, a per-account TrustRank built from verified history, and a fraud-risk penalty. Seconds that fail the checks count for nothing.

CPAS pricing

Advertisers buy those weighted seconds. The price unit is CPAS, the cost per 1,000 quality-adjusted attention seconds. A campaign sets a CPAS bid and a budget, and the budget is billed down only as verified attention is delivered. When it is spent, the campaign pauses on its own.

Share of voice, not highest-bid-wins

In many ad auctions the top bid takes the slot. BackSpin allocates share of voice instead. On the public board a campaign's weight is its bid weight times the advertiser's reputation times a fairness factor, and its share is that weight over the total. Reputation runs from 0.5 to 1.5, so a relevant, well-rated advertiser is not simply outspent. New campaigns get a 1.25x fairness boost for their first 7 days so the board does not freeze around early buyers.

The clearing price

The board also quotes a live clearing price. It moves between a floor based on a discounted average of active bids and a ceiling set by the top bid, depending on how scarce attention supply is. It never quotes below $0.50 per 1,000 attention seconds. You can see the current number on the open market.

The open ledger

Every settled bid, accrued reward and hourly price tick is written to a public ledger with no personal data. The scoring, ranking, revenue-split and pricing formulas are published in an open repository. The fraud thresholds stay private, because publishing them would tell farmers exactly what to fake.

How the money flows

Here is one wait, end to end:

1. A developer asks Claude Code to refactor a module. BackSpin's status line hook sees the tool is busy and opens a window.
2. The server picks one discovery card based on share of voice and relevance, say a database API, and the card shows in the status line, editor panel or browser.
3. The tool finishes. The window closes and is scored on the server.
4. If the window qualifies, the advertiser's budget is charged for the weighted seconds at the market price, and the developer is credited a share. By default that share is 50%.
5. The settled numbers land on the open ledger.

The developer earns reward credits at a fixed rate of 1,000 credits = $5.00 and can withdraw in USDT, Litecoin or Bitcoin after an admin review. The FAQ has the full rules.

The math on one wait

To make the units concrete, here is the arithmetic with made-up inputs. These are not real market numbers, just a way to see how the pieces multiply.

Say a wait lasts 20 seconds and scores 80 out of 100. The weighted attention is 20 x 80 / 100 = 16 seconds. If the campaign is paying a CPAS of $2.00, those 16 seconds cost the advertiser 16 / 1,000 x $2.00 = $0.032. At the default 50% split the developer is credited $0.016, which is 3.2 credits at $0.005 per credit.

Two things follow from that. A single wait is worth very little on its own, so the value comes from many real waits over time. And the quality score matters as much as the price: the same 20 seconds with the editor in the background might score 30 and be worth less than half as much.

Why "open" matters

Most ad exchanges are black boxes. Advertisers cannot see why they won or lost, publishers cannot see what their inventory sold for, and nobody outside can check the math. That works when trust is assumed. It does not work for a new category where both sides are asking whether the attention is even real.

Publishing the formulas and the ledger does two things. Advertisers can check that they are billed for verified attention, and developers can check that their share matches the published split. It also makes it hard for the exchange itself to quietly change the rules.

What an attention exchange is not

It helps to be clear about the edges:

• Not a CPM network. The unit is verified attention time, not a render.
• Not a click marketplace. Clicks are welcome, but the product is built for moments when people cannot click.
• Not surveillance. No code, prompts, replies or personal profiles are collected. The signals are about timing and presence.
• Not a get-rich scheme. Earnings depend on real advertiser demand and on how much of your time qualifies. There are daily caps, and suspicious accounts stop earning.

Where this is going

The first market is developers, because they wait on AI tools more than anyone and their attention is valuable to the companies that sell them tools, APIs and cloud credits. The same idea applies to any AI product with real wait time. For a buyer's view of the format, read advertising to developers inside the IDE. For the developer side, see how to monetize AI coding wait time.