Answers to common questions about how Warrant's AI reviews your content, what happens when you give feedback on findings, and what the compliance score means. For step-by-step guides, see Scan an Asset for Compliance and Working with Scan Findings.
How does Warrant decide which policies apply to an asset?
A scan starts from what your admin has switched on: the regulations and company policies enabled on the AI Policies page, plus your enabled disclosures, trademarks, brand guidelines and Brand Portal libraries. The switches under Customize Results (or your scan preset) then decide which of those categories and regulators this scan includes.
Warrant then uses the file's details, such as its geography, products, brand and channel, to check the policies that apply to that file. For example, federal rules always apply, while a state rule is checked only when the file's geography includes that state. If the file has no geography, Warrant uses the places listed under Where you operate in your company profile. Disabled and retired rules are not checked, and for social posts, any policy or disclosure with Applies to Reach (Social) turned off is left out. Fill in each file's details so the right rules are checked.
To see exactly what was checked, click the Checked: line under the score on the Scan tab. It lists every regulation, disclosure and custom policy the scan included.
Learn more: Set Your Compliance Scan Policies and Conditional Rules for Disclosures and Library Records.
Does Studio search the open web, or only use our policies?
Studio reviews content only against your company's configuration: your enabled regulations and company policies, disclosures, trademarks, brand guidelines and Brand Portal libraries (and Content Quality checks, if they are on). Studio cites only those policies and doesn't mention regulations you haven't enabled. Studio does not run web searches.
If you paste a website link into Studio, Warrant captures that one page and reviews it as content. It doesn't follow links from that page.
You can narrow a Studio review in your message, for example "just check disclosures". If nothing matches what you asked for, Studio checks your full set of policies instead.
Learn more: Studio: Conversational AI Reviews.
What do thumbs up, thumbs down, Dismiss and Mark Relevant do? Who do they affect?
Action | What it does | Who it affects |
Dismiss | Closes that one finding, with an optional reason, and removes it from the file's score. It doesn't tell Warrant the policy was wrong. | That finding on that file, for everyone who views it. Other files are not affected. |
Undismiss | Reopens a dismissed finding. It counts toward the score again. | That finding only. |
Thumbs down (Not Relevant) | Records your vote that the finding didn't apply, with an optional note. It doesn't dismiss the finding. | Your whole company. Votes are counted per policy and shape future scans (see below). |
Thumbs up (Relevant) | Records your vote that the finding was correct. | Your whole company. Relevant votes count against not relevant votes. |
Mark Relevant | On a dismissed finding: reopens it and records a relevant vote. | The finding, plus a company-wide relevant vote for its policy. |
Votes from the File Viewer and from Studio are pooled for each policy across your company. They are used only for your company's scans and affect future scans, not findings that already exist:
When a policy is repeatedly marked not relevant, admins get a policy task suggesting they review it.
Repeated not relevant votes on a policy make Warrant apply it more strictly, and can lead to its new findings being dismissed automatically and labeled System Dismissed.
Relevant votes count against not relevant votes, so a policy your team finds useful keeps being flagged.
Each person has one vote per finding. In the File Viewer, clicking the other thumb replaces your vote. In Studio, clicking the same thumb again removes it.
Learn more: Working with Scan Findings and Policy Tasks and the Policy Audit Trail.
Someone dismissed a finding by mistake. How do we reverse it?
One finding: on the Scan tab, click Dismissed above the findings list, find the finding and click Undismiss (or Mark Relevant to reopen it and record that it was a real issue).
A policy that is now being dismissed automatically because of feedback: use Mark Relevant on its System Dismissed findings, or thumbs up its findings. As relevant votes build up, later scans stop dismissing that policy's findings. Anyone who voted thumbs down by mistake can also change their own vote.
Findings dismissed for low confidence: an admin can lower Review sensitivity or turn off Relevancy Confidence Auto-Dismissal. This applies to new scans; findings already marked System Dismissed stay that way until someone restores them.
If you need a policy's feedback history reset, contact [email protected].
Why did my score change when I rescanned without changing the file?
AI variability. The AI can return slightly different borderline findings on identical content from one run to the next. One borderline finding can move the score a little.
Your settings changed. Policies enabled or disabled, risk levels changed, a regulation customized, disclosures added or updated, or a different Review sensitivity all change what a new scan finds and counts.
Your team's feedback. A policy may now be applied more strictly or be dismissed automatically because of your team's votes (see above).
The scan or file settings were different. A different preset, a regulator or category switched off under Customize Results, or changes to the file's geography, products or brand change which policies apply.
Dismissing or undismissing findings also changes the score straight away, without a rescan. To see what changed between versions, see Compare Scan Results Across Versions.
How is the compliance score calculated, and what do the colors mean?
The score is out of 100. Each open finding lowers the score, and High-risk findings lower it most. Dismissed findings, including System Dismissed ones, don't count.
On the Scan tab and the Info tab:
90 to 100 (green): Excellent compliance
70 to 89 (amber): Needs attention
Below 70 (red): Review required
File badges inside Studio use slightly different bands: green from 80, yellow from 50 to 79, and red below 50.
The score is a guide to how much there is to fix. It is not a legal opinion, and a human reviewer makes the final decision.
Does the score block approval?
Not on its own. Admins can choose to use scans and scores in these places:
Minimum scan score to submit on an approval flow. Assets scoring below it can't be submitted to that flow. Leave it blank for no gate. Assets that have never been scanned aren't blocked by this setting.
Require completed scan before submit in Settings, under Company, Approvals. Turn it on so nothing can be submitted until a scan has finished on the version being submitted. Use it alongside the flow's Minimum scan score to submit so unscanned assets can't get through.
Approval trigger criteria in Settings, under Company, Approvals. Choose Auto-approve above a score and set a Minimum score to auto-approve. The score must clear the minimum you set for the asset to be approved automatically. Social posts can use their own setting or inherit this one.
Learn more: Create New Approval Flow and Get Content Approved.
What does review sensitivity mean?
Review sensitivity sets how confident Warrant must be before a finding is shown. The levels run from Flexible, which shows the fewest findings, through Mild (the default), Balanced and High, to Maximum, which shows the most. Higher sensitivity shows more lower-confidence findings. Findings below your chosen level are dismissed automatically as System Dismissed while Relevancy Confidence Auto-Dismissal is on, and don't count toward the score. Brand guideline findings can use their own Brand guidelines sensitivity. Only admins can change these settings.
Learn more: Global Sensitivity Control for Compliance Scans.
Why wasn't an issue flagged?
Work through these in order:
Was it dismissed? Click Dismissed above the findings list. The finding may be System Dismissed for low confidence or because of past not relevant votes.
Was the policy checked? Click the Checked: line under the score. If the policy isn't listed, it may not be enabled (built-in regulations stay off until an admin enables them), its category or regulator may have been switched off for this scan, or it may be scoped to a state, product, brand or channel that doesn't match the file. Check the file's details too.
Is the rule in your library? If no enabled policy covers the issue, an admin can add a regulation, create a custom policy or customize an existing one. For disclosures, check the disclosure exists in the Brand Portal and is enabled, not expired and not archived. For facts such as rates or offers, check there is a matching Brand Portal library record.
Could Warrant read that part of the content? Warrant can only review what it captured from the file or web page. If a website capture failed or part of the content didn't come through, upload the file directly and scan again.
Sensitivity. At lower sensitivity levels, borderline issues are filtered out. An admin can raise Review sensitivity to see more.
The AI can also simply miss things. Findings support your reviewers; they don't replace them.
Can I open these help articles from inside Warrant?
Yes. Many pages, including Studio, the File Viewer and the AI Policies settings, have a (?) help button at the top that opens the related help article in a new browser tab. To ask the support team a question, see Get Help from Warrant Support. For an overview of how Warrant uses AI, see How Warrant Uses AI.