Scam alert video for customers with an AI avatar: script rules

A short avatar clip works as a scam warning when customers keep getting fake texts or calls in your name. It is one fixed script, rendered once, and posted on your site and in your email footer. It cannot verify a caller, check a link or react to a worried customer, so every instruction in it has to be something the customer can do alone.

Script rules that fit 20 seconds

• Name the channel the scam uses (text message, phone call, email) in the first sentence.
• State one thing your company never does, for example never asks for a password by phone. Only claim what is true and approved by your security owner.
• Give one safe action: hang up and use the number printed on your account page.
• Do not read out a phone number or URL that could be mis-heard; put it in the page text next to the video.
• End on the page where customers can report a scam.

Why a synthetic presenter needs a label here

A warning about impersonation delivered by something that may look like a real employee is awkward if it is not labelled. Add a visible line saying the video is AI-generated, in the page and in the burned captions if you want it on the frame. The disclosure checklist lists where to put it. Check the final spoken words against the approved script too, since a wrong word in a security notice is a bigger problem than in an ad.

Captions and where the link goes

Inline captions on the avatar job are generated from the spoken script. For an on-screen line that is not spoken, such as a URL, use a standalone captions job with authored cues, each with text, start and end. A standalone caption job is a flat 0.20 dollars for videos up to 60 seconds per the docs.

Cost

The clip itself is billed by the second.

Keep it current

Scam wording changes. Put a review date in the video title and re-render when the channel changes. Because the avatar already exists, a new version is one more job at the same rate.

What an avatar clip cannot do for fraud prevention

It cannot tell a customer whether the message in front of them is real, and it should not claim to. It cannot be updated live when a new scam appears, so the script should describe the behaviour of scammers (urgency, requests for codes) rather than a specific phone number that will change. A written page with the current list of known fake numbers does that better.

Finally, make the clip easy to find. A scam alert that sits on a page nobody visits does nothing. Link the page from the footer of every transactional email you send and from your help centre's top-level search results.

A quick pre-publish check

Read the finished transcript against the approved script, confirm the AI label is on the page, confirm the reporting link works from a phone, and confirm that nothing in the clip asks the viewer to call a number. Ask your security owner to sign the final version. Record the date, script, job id and video id so that the clip can be withdrawn and replaced quickly if the facts change.