Presell
How Presell Checks Claims
Presell generates from evidence rather than generating first and looking for support afterwards, and then grades what it wrote. This guide covers what counts as evidence, what the four risk classes mean, what blocks an export, and the difference between checking a claim and clearing it.
Quick Answer
Every factual sentence is graded green, yellow, orange or red against the evidence behind it. Red blocks the export. Health, financial and comparative claims are escalated whatever the evidence says.
- Evidence needs a source link and a date before it may back a claim on the page.
- Evidence without both can still steer what the page argues, and can never appear as a fact.
- Copy that asserts nothing gets no verdict at all, rather than a green one.
- We say your claims are source linked. We never say your page is compliant.
On this page
What counts as evidence
Evidence is a row with an excerpt, the source it came from and the date it was captured. Sources are your own site, whatever your review widget exposes, review sites and public discussion about competitors, competitor moves we track, and what the answer engines say about your category today.
A row that carries both a source link and a date is citable and may back a stated fact. A row missing either is not, and the distinction is enforced rather than advisory. An uncitable row can still tell the generator which objection dominates your category, which shapes the angle and the section order, and it can never become the sentence that asserts it.
Reddit evidence carries the community it came from and never a username. Somebody who posted about their sleep did not agree to appear in your advertising.
The four risk classes
Green means the claim cites evidence and the evidence supports it. Yellow means it needs a human look: something supports it partly, or the check could not complete, and abstention is never treated as approval. Orange means the sentence is health, financial or comparative, which are held to a higher standard than the rest of the page. Red means nothing supports it.
Escalation only ever goes one way. A model concluding that a claim is supported can never promote evidence that does not exist or may not be published, because if it could, the one sentence this product is allowed to say would stop being true.
- Green: source linked, and the source agrees.
- Yellow: needs confirming, with the reason shown.
- Orange: regulated category, flagged with the standard it is held to.
- Red: unsupported, and it blocks the export.
Why red blocks and orange does not
A red claim cannot be exported until you resolve it. Link the source you already have, rewrite the sentence as your own position rather than as a finding, or cut it. Those are all legitimate, and the third is the right answer more often than people expect.
Orange does not gate, deliberately. The regulated categories are most of what the customers in this product's core markets sell, and a gate on orange would make the tool unusable for them while teaching everyone else to ignore a badge. Orange is flagged with its source and its reason and left to you.
Copy is not a claim. A paragraph that asserts nothing and cites nothing gets no verdict at all, because a wall of badges over ordinary prose teaches you that a badge means nothing, and that costs more on the day one badge matters.
Checking is not clearing
We say that every factual claim on the page is source linked, and we show you the source. That is the whole of the promise. We do not tell you the page is legally compliant, because we do not know your jurisdiction, your category's rules, your substantiation file, or what a regulator has already said to you.
You still confirm the claims, the testimonials and the endorsements before anything goes live. This is compliance assistance rather than compliance, and the difference is the point.
FAQ
Can I check a page I wrote myself?
Yes. The claim check runs on a stored page or on a draft document handed over through the connector, so your own model can write a page and have it graded against your brand's evidence before anything is saved.
Why is a claim red when I know it is true?
Because red is a statement about sourcing, not about truth. From the page alone there is no way to tell a true claim from a confident one, which is also how an ad reviewer reads it. Link the source and it stops being red.
Does the same grading run on the free scanner?
The same rules and the same classifier, with one difference: a stranger's page has no evidence set, so the scanner treats a link to another domain as the citation and says so. Inside Presell the check is against your own vetted evidence.
Is the verdict the same every time?
Yes. The classifier decodes deterministically and carries a version, so the same claim against the same evidence returns the same verdict. When the grading behaviour changes, the version changes with it, so old verdicts are never quietly compared against new ones.
Related help pages
Presell
Getting Started With Presell
Go from a product URL to a page on your own domain: analyse a brand, generate a page in one of five formats, check the claims, then export it or publish it.
Presell
Measuring a Presell Page
Why the tag comes in two halves, what the collector records, how an in page test works, and why no winner is declared for you before the sample supports it.