AI check reads a policy against your own writing standards and comes back with specific recommendations you can apply in one click. Run it before a policy reaches your employees and you catch the compliance gap, the off-brand paragraph, and the sentence nobody will understand while they're all still easy to fix.

AI check is currently in beta.

A few things need to be true first:

Note: AI check runs on non-archived policies only. It doesn't run on documents or document sections, and your audience never sees any part of it.

Open the policy from your policy library, go to the AI Check tab in the policy details panel, and click Run AI Check...

A window opens asking Run an AI Check on this Policy? Untick any agents you want to skip, then click Run AI Check. It starts right away.

If you want to change what the agents look for before you run anything, click Edit AI Agents. That opens the Policy Skills page in a new tab, so your selection here is still waiting when you come back.

Select the policies you want in the policy library using the checkboxes on the left, then click AI Check... in the toolbar. The same window opens, and the title tells you how many policies you're about to check.

Each policy gets its own check, so results come back one at a time. Policies you only have view or comment access to are left out, along with any archived policies in your selection.

A banner appears at the top of the policy reading AI Check is running. Editing is disabled until the AI Check finishes. While a check is queued or running, the policy draft is read-only, so the text the AI Check's recommendations are based on can't change.

Checks usually take a few minutes, and a long policy checked against a large set of agents takes longer. If one doesn't finish, the panel tells you: The last AI Check didn't finish. Try running it again.

Note: Editing locks the moment a policy is queued, not when its check actually starts. If you check many policies at once, they'll be read-only for a while.

When the check finishes, the AI Check tab shows the result, the date it was last checked, which agents ran, and a result for each one.

A policy ends up in one of four states:

Note: These are separate from the phases you see in a policy review, which track what your colleagues decided. A policy can be Approved by your reviewers and still sit at Rework Needed here.

Each recommendation appears as a card in the AI Check panel, showing which agent raised it and why. Hover a card to see your options.

You can also reply on a card to leave a comment, and type @ to tag a colleague the same way you would on any policy comment. Useful when a recommendation needs someone else to weigh in.

The AI Check surface issues for you to judge. They don't approve a policy, and their recommendations aren't legal advice, so read each one before you apply it. Only your legal team can provide assurances that any recommendations are applicable or appropriate to your particular situation. Contact your legal counsel for advice on any particular legal matter.

For how Blissbook handles your content when AI features run, see AI governance at Blissbook.

Note: If the policy text changed after the check ran, some recommendations no longer match anything and get skipped. Apply All tells you when that happens, and re-running the check gets you a fresh set against the current text.

Open recommendations never stop you from releasing a policy. You can publish with them still outstanding and come back.

Run a check as often as you like. Once every recommendation is handled, the panel shows No Recommendations and you can check the policy again.

Note: Starting a new check discards any recommendations you haven't accepted or discarded yet, and Blissbook warns you first with Existing AI recommendations will be discarded. Accepted and discarded recommendations can't be brought back, so finish a set before you re-run.

Editing an agent doesn't re-check anything on its own. After you release a change to an agent, run the check again on the policies you want measured against the new version.

The AI Check column sits to the right of the policy name and appears on its own the first time you run a check. It shows the result of the most recent check, or Checking and Queued while one is in flight. Hover a cell to see when the policy was last checked and what's still open. The column is narrow, so No Recommendations is shortened to No Reco's there.

You can sort the column by severity, so whatever needs attention comes to the top. You can also filter it by No Recommendations, At Risk, Rework Needed, or Not Checked to work through one group at a time, and save that as a view if it's a list you'll come back to.

At the bottom of the library, the AI Check bar breaks your whole policy set down by result. It's the quickest way to see how much of your library has been checked at all, and clicking a segment filters the table to it.

When an agent gets something wrong, flag it. Hover the card and click the flag icon to open Report this recommendation.

Pick what's wrong with it, add detail under Anything else?, then click Send Report.

Also reject this recommendation is ticked by default, so flagging a bad recommendation also clears it from your policy. Untick it if you'd rather keep the recommendation and decide later. Each recommendation can be flagged once.

Flagging isn't just a complaint that disappears. At the bottom of your policy library, a flag icon shows how many flagged recommendations are waiting.

Click it to open the review, where every flag your team has raised is grouped by the agent that made it, because the agent is the thing you can actually change. Two flags under Brand Voice is your signal that Brand Voice needs an update.

Each group has an Update Skill button. Click it and Blissbook reads that agent's flags and drafts edits to the agent itself, so it stops making the same mistake. When it's done you'll see how many suggestions are waiting and a link straight to them: apply the ones you agree with, then release the agent so your next check uses it. See Set up and customize your AI check agents.

An update files away the flags it learned from, and you'll find them under the Archived tab. Restore puts one back if you'd rather it be picked up next time. A Skills left to review count appears next to the flag icon whenever an agent has suggestions waiting on it.

Note: The Legal Compliance check is based on underlying models and training data from Anthropic, with specific instructions and training data from Blissbook (and each customers' agent skill) layered on top to increase recommendation quality. We're confident it provides solid advice, but just like any AI agent (or human for that matter), it can make mistakes. It will improve over time; you can help by flagging any recommendations you think are off.