A marketing audit of your website where every factual claim carries a numbered source you can open yourself.
An AI-written audit can be wrong in a way that is genuinely hard to see. It reads confidently, it sounds thorough, and nothing on the page separates the sentences grounded in something the tool actually looked at from the ones it produced because the sentence needed finishing. Acting on the wrong half costs real money.
Verdict is built to be checked. Every factual claim in the report carries a numbered marker that resolves to an entry in a Sources Verified appendix, listing the exact URL, when it was fetched, and a confidence rating on the source itself. Findings that a machine can test are re-tested against the site data before publication, and the ones that fail are downgraded or dropped rather than published. Page one is a reliability header that states what was verified and how fresh the data is, before you read a single recommendation.
The point is not that a language model wrote it. The point is that you can audit the audit. If a claim looks wrong to you, the source is right there and you can go and check it.
A written report you can read end to end, hand to your team, or put in front of a client.
Steps in the pipeline, not promises in a brochure.
Source markers on every claim. Claims are required to carry a numbered marker, and anything that cannot be tied to a fetched page is labelled as an inference instead of stated as fact. The markers are collected into an appendix with URLs, retrieval timestamps and a confidence rating per source.
A verification gate that actually drops things. High-confidence findings that a machine can test get re-tested against the crawled site: is that form really there, is the canonical tag really missing, is there really no structured data. Findings that fail the re-test are downgraded or removed from the body. This is the part that stops a confident sentence from surviving on confidence alone.
A critic pass, published not hidden. A separate reviewer reads the assembled report looking for unsourced claims, unlabelled hedging, invented figures, stale sources presented as fresh, scores that do not match the evidence, and sections that contradict each other. Its findings are printed in the report and counted on page one.
A cleanup pass on the writing. A render-time step strips em-dashes automatically and flags a named list of stock phrases that make text read as machine-written. The report should read like an analyst wrote it, because you are going to show it to someone.
Sources older than 48 hours are marked stale and are not used to support strategic conclusions. Where a second model disagrees with a strategic conclusion, the report is marked as a draft rather than delivered as settled.
No tracking pixel, no analytics login, no onboarding questionnaire.
A single website URL is the whole input. If the site sits behind a login, we can run it against credentials you supply.
Five specialist reviewers work the crawled site in parallel, each covering a different category, then the assembled report goes through verification, critic and cleanup passes.
You get the report with its sources appendix and reliability header attached, and a call to go through what it found and what is worth doing about it.
An audit that claims to know everything is the one you should not trust.
A one-off audit of a single site and an ongoing or white-label arrangement are not the same piece of work, so we scope it on a call rather than posting a number that is wrong for most people who read it. Fifteen minutes is usually enough to tell you whether a Verdict is worth running on your site at all.
Priced per engagement, once we know the target
If a report is not the right call for where you are, we will say so on the call rather than sell you one.
Bring the URL you are worried about. We will tell you what a Verdict would cover for your situation and what it would cost.