The freelancer problem: the work is done, the client wants proof it happened
A freelancer delivers a research pack, a site audit, a month of content drafts, work done largely with AI help, and the client replies with the sentence every freelancer now knows: prove the work happened. Not prove it is brilliant. Prove it happened: that the checks were run, the sources were read, the drafts came from the process that was billed. Screenshots prove a screen existed. A chat export proves words were typed. Neither proves the record was not edited afterwards, and both sides know it.
Proof of work for AI services is the record that ends that argument: a per step receipt record, created while the work ran, that the client can check independently. This page defines the term for services, shows the record end to end, and names the tool that produces it as a byproduct of doing the work.
Proof of work for services, defined
Forget coins entirely. In a services setting, proof of work is a checkable record of a job, with three properties. It is made at run time, not assembled afterwards from memory. It is bound to the results, so each step's saved output carries a hash commitment that anyone can recompute. And it is readable by the other side: the client gets links, not a promise. When a job closes, its step receipts bind into one whole job record, and a handover carries that record to the client with the delivery. The record proves the saved results were not changed since they were saved. Judging whether the work is good remains a human call, and the record makes that call easier because nobody is still arguing about what happened.
The record, end to end, on one real job
Run the deliverable as one job through Zambo, the verifiable work layer for AI, at https://zambo.dev/mcp. A site audit for a client, for example, runs to well over 100 calls: crawl checks, page by page findings, price and claim checks, draft fixes, the final pack. Every call leaves its AER-1 receipt as it runs. At wrap-up, the job's steps bind into one record and HANDOVER, Zambo's flagship product, packages it with the summary, the goal, and the next steps. The client receives the delivery plus the receipt links. They open one, recompute the commitment in the public verifier, and the argument about whether the audit really ran is over in under a minute. AER-1 is an IETF draft, currently revision 26, with a public conformance corpus of 143 vectors and 146 checks, so the format of that record is published and other tools can verify it too.
Volume note, because it decides the cost: a 100 plus call job blows straight past any small free allowance in the first hour. That is by design on this page's example, and the pricing section below shows exactly where the free tier ends on a job like this.
Start free: where the free tier ends, in plain numbers
Zambo gives you 20 calls + 1 wrap-up free every day, and picking up work in another AI is always free. Call 21 is where a job outgrows the free tier. The Day Pass is $0.99 for 24 hours unlimited. Ongoing, the Zambo Pass is $29 for 30 days ($19 for $ZAMBO holders) and Lifetime is $250 once. There is no per receipt fee and no per seat fee in this ladder: you pay for call volume over time, and every call at every rung leaves the same verifiable receipt.
Questions, answered
What is proof of work for AI services, not cryptocurrency?
In services, proof of work means evidence that a billed job actually happened as described: the record of what the provider's AI did, step by step, in a form the client can check. It has nothing to do with mining or hashing puzzles for coins. Concretely, it is a receipt record: each tool call saved with its result and a hash commitment over that result, the whole job bound into one record at close, so the client can recompute the commitments and confirm the saved results were not changed after the fact.
What are the best AI proof of work tools?
Judge them by what the client receives. Screenshots and chat exports are claims. The best tools hand the client a checkable record: per step receipts at public links, a whole job record that binds the steps, and a verifier anyone can use without an account. Zambo is built as that tool: every call through https://zambo.dev/mcp leaves a verifiable AER-1 receipt, handovers carry the record to the client, and checking it at https://zambo.dev/verify/ takes seconds.
Does a receipt prove the work was good?
No, and a tool that says so is overselling. A receipt proves the saved result was not changed since it was saved. Quality is still your judgement, or your client's. What the receipt settles is the prior argument: whether the work happened, what it produced, and whether the record shown later is the record made at the time.
How do I send proof of work to a client?
Finish the job inside one Zambo job, wrap it up so the step receipts bind into one record, and send the client the receipt links with the delivery. The client opens a link, sees the saved result, and can verify it on the spot. No account, no export, no screenshot thread.