The economics of free attention.

Digital communication has a structural flaw: the cost of demanding someone’s attention is borne almost entirely by the person whose attention is demanded. Sending an email costs fractions of a cent. Reading, evaluating and dismissing it costs the recipient seconds to minutes of finite time.

Multiplied across a professional’s inbound volume, that asymmetry produces the modern inbox: the founder reaching an investor, the candidate reaching a hiring manager and the journalist reaching a source all drown in cold outreach that was rational to send precisely because sending was free.

The standard response has been technological: Bayesian filters, sender reputation, machine-learned priority inboxes, and now large-language-model triage. These reduce noise but cannot eliminate it, because they never alter the sender’s incentive.

• A campaign with a 0.1% response rate stays profitable at zero marginal cost.
• Filtering invites an arms race: screening models and generation models now improve with the same underlying technology.
• Every software defense can be probed and bypassed for free. An economic constraint cannot.

If reaching an unknown recipient requires posting even one dollar of refundable value, mass low-relevance outreach becomes expensive in exact proportion to its volume — while a single legitimate message costs its sender nothing but float.

• Proof-of-work postage (Hashcash, 1997; Penny Black). The cost was real but non-transferable — the recipient gained nothing — and botnets paid less per message than honest users on laptops.
• Attention bonds (Loder, Van Alstyne & Wash, 2006). Economically sound and commercially stranded: no way to escrow and settle millions of sub-dollar bonds, and no deployment path that avoided replacing the recipient’s email provider.
• Paid-introduction marketplaces (21.co / Earn.com). Two-sided cold start, non-refundable payment that felt like a toll, and demand capped by human patience.

Knock differs on the axis that killed each one: value transfers to the recipient and scales in currency, settlement rails now exist as commodities, an alias address requires no change of provider, and machine senders — absent from every prior attempt — are the marginal buyer this time.

A recipient publishes, in machine-readable form, the price and rules under which unknown parties may claim their priority attention. A first-time sender — human or AI agent — attaches a small refundable deposit to their message.

• Reply or accept → deposit refunded in full.
• Never reviewed → deposit auto-refunded on expiry. Recipients cannot harvest by ignoring.
• Reviewed and rejected → deposit retained, after an affirmative action, never a tracking pixel.

Incentives line up on all three sides. Senders self-select on genuine relevance. Recipients earn only by doing the work of review. The platform charges fees only on retained deposits, so revenue scales with attention actually protected rather than with gross message volume.

The refund default is not a pricing detail — it is what makes the mechanism acceptable. Deposits refunded on reply, refunded on expiry, and retained only after review keep Knock legible as an anti-spam bond rather than a celebrity tollbooth.

The health metric is therefore inverted from a naive marketplace: a high refund rate among paying senders is success. It means the mechanism is admitting legitimate strangers and deterring the rest before they ever pay.

Autonomous systems increasingly need a human: to review an exception, authorize a decision, answer an expert question. Today those requests arrive as unauthenticated email indistinguishable from spam, or do not arrive at all.

For an agent, the challenge is not an email but a status code. It POSTs to the recipient’s contact endpoint, receives 402 Payment Required with x402 payment requirements derived from the manifest, attaches a USDC payment header, retries, and receives 202 Accepted with a status URL.

POST /v1/contact/{handle}      -> 402 Payment Required
POST /v1/contact/{handle}      -> 202 Accepted
  X-PAYMENT: <x402 USDC payload>   { "challenge_id": "...", "status_url": "..." }

Humans get the same challenge as an email with hosted card checkout. The manifest is the single source of truth behind both surfaces.

A protocol that priced a subpoena or a safety warning would deserve the criticism it received. Legal notices and emergency communications keep a payment-free path, and certain structural misuses are out of bounds: public officials monetizing constituent access, fiduciaries where ethics rules forbid it, employees monetizing employer-owned inboxes, and any participation by minors.

Why is spam profitable?

Sending costs fractions of a cent while reading costs the recipient seconds to minutes. A campaign with a 0.1% response rate stays profitable at zero marginal cost, so volume is always rational for the sender.

Why can't better filters solve it?

Filters treat the symptom, not the incentive. Every software defense can be probed and bypassed for free, and generation models now improve as fast as screening models. There is no equilibrium in which software alone wins.

What is a refundable attention deposit?

A sender attaches a small refundable deposit to a first-contact message. Reviewed and legitimate messages are refunded, unreviewed messages auto-refund on expiry, and only a message the recipient reviews and rejects results in retention.

Isn't this a tollbooth on access?

No. Refund-by-default means an honest sender pays nothing but float. A high refund rate is the health metric of the network: it means legitimate strangers are getting through and low-relevance outreach is deterred before it is ever sent.

Why does this work now when earlier attempts failed?

Settlement rails exist as commodities: cards for humans and x402/USDC for machines. AI agents are the first class of sender that can pay natively and at scale, which removes the demand cap that limited paid-introduction marketplaces.

• Download the full whitepaper (PDF)
• Protocol documentation — manifests, deposits and settlement
• Pricing and early access