Loading…
KNIGHTBYRD TECH LLC · VERIS
Buying a claim-evidence tool: nine questions to ask
Ask these of any vendor in this category. We have answered them for ourselves below — including the ones we would rather you did not ask.
01Do you record individual claims, or pages?
Ours: individual claims, verbatim, seven kinds.
02Can you tell me whether a cited source actually measures what the claim asserts?
Ours: yes — a source-class table drives it.
03Do you detect breadth claimed against breadth tested?
Ours: yes, and unstated scope stays unknown.
04Is every line dated to when the page was read?
Ours: yes, and the gate refuses to render without it.
05What happens when my product changes but the page does not?
Ours: that is the drift monitor, and it is the inverse of change detection.
06Do you test the commercial disclosure and the AI disclosure separately?
Ours: yes, in both directions.
07Will you tell me what your tool found on your OWN website?
Ours: 2 findings, published.
08How long until my first result?
Ours: under a minute, free, no account.
09What does your tool NOT do?
Ours: it does not determine whether a claim is adequate, and it is not legal advice.
How to read any answer, including ours
- Ask what the answer was measured on, and when. The two questions that separate evidence present from evidence sufficient.
- Treat an undated capability claim the way you would treat an undated accuracy figure. It is the same failure in a different suit.
What the category does, and what it does not
By product category, never by company. A dash means a capability is not described in published vendor materials as of 2026-08-19 — not that it is impossible. That distinction is the product, so we apply it here too.
| Capability | Change-detection platforms | Enterprise compliance suites | SMB marketing-review tools | Ad-creative archives | VERIS |
|---|---|---|---|---|---|
Reduces a page to its individual claims A substantiation request is never about a page. It is about one sentence, and someone has to find it first. | — | partial | partial | — | yes |
Says whether the cited source can carry the claim Evidence being present is not evidence being sufficient. A review ranking supports that people liked you; it does not measure performance. | — | partial | partial | — | yes |
Detects breadth claimed vs breadth tested This is the failure that ended a real enforcement matter: the claim was accurate for what was tested and unsupported for what was promised. | — | — | — | — | yes |
Dated provenance on every line Substantiation is judged as of the date the claim was made. Undated evidence is a finding of its own. | partial | partial | — | partial | yes |
Watches for claim decay against product versions Change monitoring fires when a page moves. The expensive failure is the opposite — the page has not moved while the product shipped four versions past the evidence. | — | — | — | — | yes |
Tests both disclosure requirements separately One label does not satisfy two rules. A commercial label does not cover AI involvement, and an AI label does not cover the commercial relationship. | — | partial | partial | — | yes |
Points the same inventory at a competitor You can find out what a rival published to support the claim that just beat you in a deal — as a record, not an accusation. | partial | — | — | partial | yes |
Time to a first result Speed is the whole difference between finding out now and finding out after someone asks. | yes | yes | yes | yes | yes |
Publishes its own results against itself Anyone can grade someone else. The question a buyer should ask is what the tool says about the company selling it. | — | — | — | — | yes |
Your claim exposure, worked out with your own numbers
We supply the arithmetic. You supply the counts.
Work out your own exposure
- A. How many pages publish measurable claims about your product?
- B. Of those claims, how many cite a source beside them?
- C. Of the cited ones, how many cite something that measures what the claim asserts?
- D. When was each supporting measurement produced, and has the product shipped since?
The published statutory figure
- The FTC's 2026 statutory maximum is $53,088 per violation, and roughly 670 companies were formally notified of potential civil penalties.
- Multiplying that by a claim count gives a THEORETICAL MAXIMUM only. It is not a prediction of enforcement, a penalty estimate, or a legal conclusion — and we will not present it as one.
We ran it on ourselves before running it on anyone else.
knightbyrd.com returned 2 claims of our own that would face the same question, and we published both. Separately, 16 companies were read on 2026-08-19: 368 claims inventoried, 133 would face a substantiation question, and 2 came back with nothing high.
Comparisons are by product CATEGORY, never by company. We do not name or rank competitors. "Not documented" means a capability is not described in published vendor materials as of 2026-08-19. It is not a claim that a competitor cannot do it. Counts come from the visible text of the pages submitted. Example output inside product demos can read as a claim, so a count is a starting point, not a verdict. No VERIS unit has been sold at any price. Nothing here reports a customer outcome, because there is not one yet. VERIS is an evidence inventory: it records what was claimed and what evidence was published alongside it. It does not determine whether a claim is adequate, and it is not legal advice or a legal conclusion. KnightByrd Tech LLC · veris@knightbyrd.com