One lifecycle — observe, verify, fix, decide, build, re-verify. Verification of a single AI surface is the way in; consulting is the continuation the evidence earns. Work is scoped after a call and billed by invoice; there are no package prices on this site.
One surface we had no hand in building — Google's AI answering about you, ChatGPT answering about you, or the chatbot you already run — checked end to end. You get the cited findings report, the fix specification your web person can apply, the thirty-day re-scan, and, when the surfaces hold up, the signed accuracy certificate in the public registry.
Your own chatbot is the higher-stakes half of the same problem: it speaks for you, in your name, to customers who take it at its word. Verifying it is the concrete first engagement most businesses need.
Some findings end in a fix specification. Others end in a decision to build something better. Where that's the decision, we design, build, host, and operate light customer-facing AI for your business on our own infrastructure — you do the minimum on your side, and hand-off is an option when you have a technical team.
Ongoing operation is a plain retainer: we host it, watch it, keep improving it, and review it with you monthly. We're a small firm and take a limited number of these at a time — capacity is the honest constraint, and we say so before you commit.
What we build gets continuous QA and monitoring under our own name — built, hosted, monitored.
On this site, independent verification refers only to AI we had no hand in building — the public assistants answering about your business, or a chatbot someone else built for you. AI we build, host, or operate is described as exactly that: built, hosted, monitored, under continuous QA. We do not grade our own homework and call it independent.
There are no package prices on this site because the jobs vary too much for a menu. After the thirty-minute call you get a written scope and an invoice through Mercury, our bank — work starts when both are accepted. Nothing auto-renews, and nothing upsells itself.
Those are run by companies that sell you software afterwards — one asks for consent to marketing calls with the free scan, another checks a single platform on the free tier. They're grading a problem their product monetises. We sell the check, and the check stands on its own.
You might need one. But an agency's audit exists to sell the retainer — several publish this openly, crediting the audit fee against the build. Inside a verification engagement we're the inspector, not the contractor: we hand you the evidence, and your agency is probably the right people to fix it.
No, and nobody can — there's no ranking to buy. What we can do is show you, with citations, where AI assistants say things about you that aren't true, and what's causing it.
It's scoped after the call — a single-surface verification and a build-and-operate retainer are different jobs, and pretending one menu covers both would be theater. You'll have the number in writing before any work starts, and the first thirty minutes cost nothing.
For a covered direct-interaction system, EU AI Act Article 50 calls for clear, accessible AI-interaction notice no later than the first interaction, unless the AI nature is obvious in context — generally applicable from August 2, 2026. California SB 243 is narrower, covering defined companion-chatbot platforms. We check technical evidence and disclosure placement; we do not sell a legal conclusion.
Book the thirty-minute call, or start with the scan intake form and we'll reply by email. Either way a named human scopes the work and tells you plainly if there's nothing worth checking.
