AI agents and checks · stage by stage

What our software does, and what it can never do.

Software reads your claims, rules them and drafts the rewording. A licensed pharmacist signs every file. Two of the stages are AI agents that work from your own file: the Watch agent on monitoring and Your claims agent on the dashboard. This page lists each stage, how often it runs and where it stops.

The test

We call a stage an agent only if it does all five.

  1. 01

    It works toward a goal the client sets in plain English, in more than one step.

  2. 02

    It works from that client’s own file and shows what it relied on.

  3. 03

    It produces something the client can use: a ruling, a rewording or a draft.

  4. 04

    It acts only when the client approves, or it hands the work to the pharmacist.

  5. 05

    It keeps a dated record the client can read.

By that test two of the six are agents today: Watch agent and Your claims agent. The others are listed as what they are.

The stages

Each stage, with its limit.

Weekly readCheck
What it doesReads each monitored Amazon listing and page and compares its text with the last reading on file.
How oftenOnce a week for each monitoring plan. A listing or page that could not be read is tried again the next day.
What it works fromThe listings and links on your file.
What it can never doIt never rules a line, and it never changes a listing, a page or an ad.
Who signsNobody signs a reading. It is a record of what the page said that day.
Where it runsAmazon listings are read from our own reading machine, on a schedule.
Why it is not called an agentIt does one fixed job on a timer. The client gives it no instruction beyond the list of listings and pages.Its output is a reading: changed, not changed or not read. It is not a ruling or a rewording.
RulingAutomation
What it doesRules each line of the text that was read against the published standards, quoting the line and citing the rule.
How oftenA full scan every 28 days, and a ruling of the changed text when the weekly read finds a change.
What it works fromThe text of your listings and pages as we last read it, and the published standards.
What it can never doIt never clears a line for good: a ruling is a draft until the pharmacist signs the file.
Who signsA licensed pharmacist reviews the rulings and signs the file.
LimitSupplement copy only. Telehealth pages are not ruled this way: the telehealth rules are a draft and not in force.
Where it runsThe scan and the ruling of changed text run on our own machine.
Why it is not called an agentIt rules whatever was read. It takes no instruction from the client and it is one step.
RewordingAutomation
What it doesWrites up to three rewordings of a flagged line and rules each one again. A rewording that draws a finding is dropped.
How oftenWhen you ask your claims agent for it, or check a draft, and once for each new finding the weekly check turns up.
What it works fromThe flagged line and the rule it broke.
What it can never doIt never puts a rewording on a listing, a page or an ad. You copy it there yourself.
Who signsA licensed pharmacist signs a rewording when you ask for sign-off, or as part of the signed file.
Why it is not called an agentIt has no goal of its own and is not a job of its own: the claims agent or the weekly watch hands it one line.
Watch agentAgent
What it doesFollows your monitored listings and pages from week to week: when the weekly read finds changed text, it has that text ruled again, sends a new Critical or High finding to the pharmacist, and drafts a rewording for each new finding for you to send to the pharmacist or dismiss.
How oftenA check every week, a full scan every 28 days, and an email within 48 hours of a new Critical finding.
What it works fromYour file: the last reading of each listing and page, and the rulings that stand on it.
What it can never doIt never edits or publishes anything, and it never signs.
Who signsA licensed pharmacist reviews each new Critical or High finding and signs the monthly summary.
Where it runsIt depends on the weekly read, which runs on our own reading machine. While that machine is off a check waits, and it runs when the machine is back.
Public recordAutomation
What it doesPublishes one fingerprint of every listing ruling we hold, so that no ruling can be backdated.
How oftenWith each site release, at most once a day. A fingerprint has been published for every day from 28 September 2026 to 9 October 2026.
What it works fromThe scan databases on our own machine.
What it can never doIt never shows a ruling to anyone. The published fingerprint reveals nothing about any brand.
Who signsNobody signs the fingerprint. The Internet Archive dates the page it is published on.
Where it runsIt runs on our own machine, as a step of each site release.
Why it is not called an agentNo client sets its goal. It fingerprints everything we hold.It works from our scan databases, not from one client’s file, and it does not say which of a client’s rulings a day’s fingerprint covers.
Your claims agentAgent
What it doesAnswers your questions from your own file, rules a line you want to publish and proposes rewordings, showing the finding or the rule each answer relied on.
How oftenWhen you ask. Most answers arrive within a few minutes.
What it works fromYour file: the stored findings and the listing text they were found in, and the published standards.
What it can never doIt never approves, signs or certifies an answer, and it never edits a listing. An answer is an AI draft until the pharmacist signs it.
Who signsWhen you ask for sign-off, a licensed pharmacist reviews the answer and can sign it.
For every stage

The limits that hold for all of them.

  1. 01

    No agent and no automated step approves, signs or certifies anything. A licensed pharmacist signs every file.

  2. 02

    None of them edits or publishes a listing, a page, an ad or a store. They draft, and you make the change.

  3. 03

    A ruling says a platform or a regulator can flag a line. It never says a line will or will not be flagged.

  4. 04

    None of this happens the moment you edit a page. The cadence is a check every week, a full scan every 28 days, and an email within 48 hours of a new Critical finding.

  5. 05

    Amazon listings are read from our own reading machine, on a schedule.

  6. 06

    The free telehealth site check is fixed software, not an AI model.

The rules each ruling cites are on the standards page. What a review does not cover is on the limits page.

Software drafts, you decide, and a licensed pharmacist signs.

© 2026 Claims Verified Privacy Terms Unsubscribe Every file signed by a licensed pharmacist Published from Toronto · 109 current standards · every figure recomputed from the scan databases before each deploy