The new role · Stewardship and ethics

Disclose or not

Decide for four AI-assisted assets whether each runs as is, runs with a disclosure line you write, or does not run, and name the rule behind each call.

caseeasyExercise · 3 min · 59 of 60

On this page

Four assets from one campaign, and a decision for each that rests on a named rule: 16 CFR Part 255, Guides Concerning Use of Endorsements and Testimonials in Advertising (opens in a new tab), 16 CFR Part 465, the Rule on the Use of Consumer Reviews and Testimonials (opens in a new tab), and in the EU Article 50: Transparency obligations for providers and deployers of certain AI systems (opens in a new tab). This practises ethics and disclosure.

The situation

The invoicing app is launching an annual-plan campaign in the US and the EU. Every asset was made with generation tools, and the brand steward has to clear four of them this week.

Search and social ad copy, generated and edited by the team. No person and no quote appear in it. Its one claim, "send an invoice in under a minute", rests on a measured median of 48 seconds, on file.
A 30-second video in which a realistic generated presenter reads a review from Priya, a customer since 2024, who agreed in writing to its use: "It cut my invoicing time in half." Across customers, the median saving is about 25%.
A creator paid $2,000 a post shares a carousel built around an image she generated, a realistic photo of a freelancer at a desk. The draft caption has "#ad" as the last of 14 hashtags.
A carousel of five "customer quotes" the team generated from survey themes, each signed with a first name.

The question

For each asset, decide whether it runs as is, runs with a disclosure line, or does not run; write each line in full and name the rule it satisfies.

A strong answer · Rubric · Where answers go wrong

Write your answer first, then open

1. Ad copy: run as is. No endorsement, nothing realistic that did not happen, and the claim is proved by the 48 seconds on file. No rule here requires a label on edited generated copy; the EU's text duty covers text informing the public on matters of public interest, which an ad is not.

2. Video: run with two lines. On screen for the whole spot: "AI-generated presenter, not a customer. Words from a review by Priya S., a customer since 2024." Beside the claim: "Customers typically save about 25%." The presenter is shown as a consumer, so §255.2(c) requires saying it is not one; Priya's result is read as typical, so §255.2(b) needs the typical figure; in the EU a realistic generated person is a deep fake under Article 50(4). The review is real and permitted, so §465.2 is met.

3. Creator post: fix, then run. First line of the caption: "Paid partnership with the app", with the platform's paid-partnership label, and "AI-generated image" on the photo. §255.5 requires the payment disclosed, §255.0(f) requires it unavoidable on social media, which the fourteenth hashtag is not, and the realistic photo needs its label under Article 50(4) in the EU.

4. Generated quotes: do not run. The quoted customers do not exist, which §465.2(a)(1) prohibits outright, and no line repairs that. Replace them with five verbatim quotes from survey respondents who agree to be named.

What would change a call: if the 25% median turns out to be measured on a different plan, restate it before the video runs; if Priya withdraws consent, the video does not run at all.

Rubric

Decides each of the four assets separately: run, run with a line, or do not run.
Writes each line in full, with where it sits and how long it shows.
Names the section each decision rests on: §255.2(b) and (c), §255.5, §255.0(f), §465.2, Article 50(4).
Pairs Priya's halved time with the typical 25%.
Refuses asset 4 outright rather than labelling it.
In an interview, says which asset needs no line at all, instead of labelling everything.

Where answers go wrong

"Label every asset 'Made with AI'." It over-labels asset 1 and fixes nothing that matters: a label does not make asset 4's reviewers exist, does not disclose the creator's payment, and leaves Priya's atypical result reading as typical.
"Asset 2 is fine because Priya's words are real." The words are; the person on screen is not. A generated presenter shown as a customer needs §255.2(c)'s disclosure, and her result still needs the typical figure beside it.
"Move #ad to the front of the hashtags." It is still one tag in a block at the end of the caption, which a reader skims past. The disclosure goes in words, in the first line, where it cannot be missed.