How do you catch AI hallucinations before shipping AI-generated output?

Genuinely curious how other builders here handle this, because I don't think there's a settled answer yet.

If your product generates text, summaries, descriptions, code comments, whatever using an AI model somewhere in the pipeline how much of that output do you actually check before a user sees it? Full verification pass?

• Spot-checks?

• Trust the model and fix things when users report them?

• Something automated?

What's making me ask: the failure mode that worries me most isn't obviously-broken output that gets caught fast, by you or by users. It's output that's fluent and confident and subtly wrong in a way that's easy to miss on a quick read, especially once you've seen a thousand outputs from the same model and start skimming instead of reading.

Curious whether people have landed on an actual process for this, or if it's mostly "ship it, watch for complaints" at this stage of the ecosystem. And if you have a process did it come from a specific incident, or did you build it in from the start?