We built the homepage to be the front door. It carries the pitch, the demo, and the install buttons. Ask me a month ago which page would carry this site and I would have pointed at it without thinking.
Then I opened the analytics.
In the last 30 days, 101,683 visitors came through zambo.dev. The number one page is not the homepage. It is not pricing, and it is not the docs. It is /verify/, the page where you paste in a receipt id and recompute whether the record an AI left behind still matches what it saved at the time. 23,012 visits in 30 days, the top page in every window we can cut. Yesterday it took 4,002 of the day's 8,065 visitors, ahead of /chat at 2,161 and /add/ at 842.
Half of yesterday's visitors skipped the pitch entirely and went straight to the check.
What they are checking
Every job run through Zambo, the verifiable work layer for AI, saves its result at execution and leaves a verifiable receipt: a record carrying a digest over the saved result, filed where anyone can open it and recompute it. In the 7 days to October 21, the platform minted 12,748 of those receipts, across 121 tools. Over 11,000 of them were for outside users. People and agents who are not us.
So the check page is not a feature tour. It is where work gets audited after the fact. An agent ran a job somewhere, came back with a summary, and the person on the other end did the one thing a summary cannot survive on its own. They went and looked at the record.
Search sends them. Yesterday google.com sent 3,420 of our 8,065 visitors. Nobody searches for a brand they have never heard of. They search for a way to tell whether the work their AI reported actually happened the way it says it did, and the check page is where they land.
What a receipt proves, and what it never will
A verified receipt proves one narrow thing: the saved result has not changed since it was saved. Recompute the digest, compare it against the filed record, and you know you are reading the result the run actually left, not a retelling written afterward.
It does not prove the answer was right. It does not prove the work was good. That judgement stays with the person checking, and it should. What the receipt changes is the object being judged. Instead of grading a summary produced by the same process that did the work, you grade a fixed record that cannot quietly edit itself after the fact.
That narrowness is the point. A check that claims too much is marketing with a green checkmark. A check that proves exactly one thing, and that anyone can run for free without asking us, is infrastructure.
The takeaway, if you ship agents
We expected the front door to be persuasion. The numbers say the front door is verification. The traffic is not people coming to read about work. It is people checking work that already happened, work an AI claimed it did.
If you ship agents, your users will develop this habit whether you plan for it or not. The first time a confident summary turns out to be wrong, the user starts looking for the record underneath the next one. If your stack saves that record at execution, written while the work ran and recomputable without your permission, you keep the trust. If it does not, your users will go find someone whose work they can check. A large share of our number one page is exactly that: other people's users, checking.
Give your work a checkable record, or your users will go find someone whose work they can check.
an agent receipt is a record of one ai agent tool call that anyone can check: it keeps the exact saved bytes and their sha-256 digest, and it proves the result was not changed since it was saved, not that it was right.
If the problem on your desk right now is an agent that says done with nothing underneath to verify, that is the problem this is built for. The Day Pass is $0.99 for 24 hours at https://zambo.dev/day-pass/, every run leaves a receipt, and checking a receipt is free whether you paid or not.
Top comments (0)