Content that cannot publish a claim you have not approved.
The Verified-Claims Content Gate sits in front of your publishing, and checks every piece against your approved-claims list and your brand voice before it can ship. When a post makes a claim that is not on your approved list, the gate does not warn you and let it through, it holds the piece, shows you the exact line, and hands you a suggested fix. Fail closed, on purpose.
Run the gate over a batch of your own content See a piece get held at the gate
THE PROBLEM
AI content engines write fast, and they will confidently assert a benchmark number that does not exist, a capability you never shipped, or a health or financial claim your lawyers would never sign. Most of the time the copy reads fine, which is exactly the danger, because nobody catches the one line that is false until it is already live and someone with a subpoena is reading it. The usual controls do not help here, a second model that scores how on-brand the copy feels is just vibes with a number attached, and a human reviewing every post at volume is the bottleneck you bought the AI to remove. As of August 2, 2026, the EU AI Act turns "prove what your AI did" into a board-level obligation, and for content that means being able to show that nothing shipped that was not on a list you approved.
HOW IT WORKS
The gate does not write your content and it does not decide what is true, you do that when you build the approved-claims list. What it does is sit between your draft and your publish button, and for every piece it runs three checks against your own DATA: it scans for forbidden terms and off-brand language across 400-plus enforcement rules, it cross-checks every factual claim in the copy against your approved-claims allowlist, and it scores the writing against your voice grammar. If any claim in the piece is not on the approved list, the piece is held immediately as an emergency stop, well past a warning, and downstream publishing halts. Where the copy is clean it passes, where it is not you get the offending span, the rule it broke, and a suggested rewrite. Your content, your list, your voice rules, all loaded as data you control.
WHAT YOU GET (the gate report)
For every piece you run through it, one clear verdict: - pass, or held, with a per-piece voice score, - for anything held, the specific violations (forbidden term, off-voice grammar, or, the one that matters most, an unapproved factual claim), each with the exact offending span, - a suggested fix for each violation, so the hold is actionable, not just a red light, - an emergency-stop flag that fires the moment any unapproved claim is found, so a false claim cannot slip through on a piece that otherwise looks fine, - a clean audit line you can show a regulator, nothing published that was not on your list.
WHO IT'S FOR
Regulated and claims-sensitive brands publishing at volume, health, finance, supplements, legal, and any company running an AI content engine where a made-up claim carries real liability. If a false or unsubstantiated line in your marketing can end up in front of an auditor, the FTC, or a plaintiff's lawyer, this is for you. If an unsubstantiated claim in your content costs nothing, it is not.
PRICING (the ladder)
We price the ladder, not a single number, and every number below is a hypothesis we validate with you, not a commitment. - Audit, a fixed-price one-time pass over a batch of your existing or queued content. You see exactly which pieces carry unapproved claims and which lines break voice, with fixes, before you commit to anything. - Value-metered, per piece gated pre-publish, once the audit proves the catch rate. Your content engine keeps generating, the gate checks each piece before it ships. - Platform, license the gate into your CMS as a required pre-publish check, with your own approved-claims list and voice rules loaded as data.
IP is licensed, never assigned. The gate stays ours, the list and the proof are yours. Any step that takes your money is gated and confirmed before it runs, nothing charges silently.
THE PROOF (dogfood)
We run this gate on ourselves before we sell it. AYA's own content engine routes every post it generates through this same approved-claims allowlist and the same emergency stop before anything can publish, and it exists for a blunt reason, an internal audit of our own draft marketing found that 15 of 27 benchmark scores we were about to publish were fabricated. So we built the gate that would not let that ship again, and then we pointed it at our own posts first. The control we are selling you is the one we already point at ourselves.
HONEST NOTE
We would rather under-promise. The gate is built, the validator pattern composes real atoms that resolve on disk, and the fail-closed emergency stop is written into it. Two things are honestly not finished yet, and both are data edits rather than code, so we will name them. First, the approved-claims allowlist is present on disk but is not yet registered in the engine's overlay registry, which means at this moment the allowlist does not bind at runtime until we register it. Second, the gate today produces suggested fixes but does not yet auto-run the correction step, the two correction pieces exist but are not yet wired into the flow. Neither is a rewrite, both are a few lines of JSON, and we would rather tell you exactly where the seams are than imply a finished loop we have not closed. What the gate has NOT done yet is run at volume against a live customer content pipeline, that is the first real test, and we will say so plainly.
Run the gate over a batch of your own content
*This page is a specification. The capability it describes is not built yet, and nothing here is a claim that it runs today.*