Copilot
Copilot open full view →
Ask Assay anything.
Straight answers about Assay, grounded in our own Truth Graph. Not a generic model's best guess.
Straight answers about Assay, grounded in our own Truth Graph. Not a generic model's best guess.
Your prices, features, and positioning move. Your published claims don't. Your own AI tools inherit the drift and send with confidence. Buyers' AI engines repeat it. The audit crawls what's public and shows you exactly where you contradict yourself.
2026 changed the blast radius
A contradiction used to sit quietly on page 9. Now it's training data for every model that reads you, and it ships at machine speed, in flawless prose, with total confidence.
They're trained and prompted on what you've published. If two pages disagree on price, they'll pick one and send it with confidence, and you won't know which until a customer does.
They compose the answer a buyer gets from whatever's public and current. The version they repeat is the one you left lying around, not the one you'd approve today.
The four kinds
No one ships a contradiction on purpose. It accumulates one honest edit at a time, until your own material argues with itself. The grader sorts it into four kinds.
Two live surfaces state different prices, terms, or numbers. Both are published. Both are being read.
A feature you sunset, a metric that expired, a tier you retired, still live on a page. The AI quotes the corpse.
Three versions of who you are, published in three quarters. Buyers meet all three and can't tell which is current.
An assertion with nothing behind it. Fine until a security review, a journalist, or an AI asks where it came from.
How it runs
No call, no demo, no promise. You give it a URL and it does the work in front of you, then hands you the receipts.
Work email and your company URL. That's the whole ask. No card, no call booked.
Your public surfaces, the same pages an AI engine reads. Read-only. Nothing internal, nothing behind a login.
Claims are extracted, cross-checked against each other, and scored. Contradictions, dead claims, forks, and unsourced assertions surface with their page and sentence.
The full report, yours to keep, every finding cited. Not a teaser. Not gated. The team picks up from there if you want it.
Not another audit
This one measures whether what's already out there still makes sense. It's a subtraction tool, not a content quota. Here's what it will never do.
The number comes from findings you can open and verify. No vanity metric, no black box you have to trust.
This measures what you say, not how you're secured. It crawls public pages. It doesn't touch your systems.
The whole thing is yours, with every citation, on the first pass. No "unlock the full findings" wall.
Read-only, always. The audit reads your public surfaces and reports. It never writes, posts, or changes a thing.
One number
The fastest way to see the drift is to point it at your own site. Free, read-only, yours to keep.