Do we have to label AI-generated ads, and does labeling or AI voiceover hurt results?
Our answer
Add a short text label, shown from the start, whenever an ad shows a realistic AI-generated person. In the EU, label realistic AI-made objects, places and events too.
- Confidence
- Medium
- Strongest proof
- 4 controlled tests
- 7 Meta rules, 2 datasets, 1 practitioner, 3 opinions or claims
- Sources
- 15 sources
- 17 statements
- M
VLFH
New here? Take the 1-minute tour
Why this is the answer
3 reasons and our judgmentMeta's own AI label is not on the ad itself. It sits in the About this ad menu behind the three dots (Meta).
State laws already require a disclosure. New York requires a conspicuous one for AI-generated performers (Venable LLP), and California's SB 1050 adds AI narration from 2027 (Foley Hoag LLP).
Labels did not hurt in a test of 900 US viewers. They made no significant difference to brand choice or ad liking, though a full-length icon scored lowest (MediaScience, reported by Marketing Dive).
Our judgment
No source rules on Meta's menu label directly, so we judge it is not enough on its own. Reed Smith reads the EU rule as requiring a disclosure people can perceive at first exposure without specialist tools or an extra step, and Venable LLP advises meeting the FTC's clear and conspicuous standard in New York. We prefer text shown from the start over an icon because text labels scored highest on liking (Marketing Dive), and the AMCIS lab study found labels cost early attention, most of all before the ad. Your own A/B test on cost per purchase should override these results, because neither controlled test ran on Meta or measured purchases (MediaScience, Azerion and Differentology).
Our judgment, built from Meta, Reed Smith, Venable LLP, MediaScience, Marketing Dive, AMCIS 2026 Proceedings (AIS eLibrary) and Azerion and Differentology: our reading of these sources, not a finding from any one of them.
What the sources say
17 statements from 15 sourcesMeta says it put proposed changes to its commercial terms on hold after publishers raised concerns about how they apply to its web crawlers. It plans to consult publishers and regulators in early 2027 before any rollout.
“the proposed updates to our commercial terms of service have been put on hold”
An Auros.ai survey of 4,000 adults reports 59% of Australians do not trust AI-advertised products to match their description, and 44% are less likely to buy if they know AI was used. Stated attitudes.
“Close to half (44%) say they are now less likely to purchase a product if they know AI was used in its advertising.”
Meta told Reuters on October 28 that it will not change its terms without a full consultation with industry groups, which could run past November 20; until then its existing terms stay in place.
“Until then our existing terms remain in place.”
Resemble AI's scan of 37,768 Meta ad files flagged 8.9% as AI-generated or manipulated, and only 2 disclosed AI in the copy. It used its own detector and published no error rates.
A Marketing Science study of 4,633 paired Taboola ads found no average click-through gap for AI-generated images, and AI images that don't look like AI beat human-made ones. Clicks only, not sales.
“ads with AI-generated images outperform human-made ads when the AI images do not look like AI.”
Meta's Commercial Terms from October 30, 2026 make businesses solely responsible for what its AI features produce, bar removing AI watermarks, and let Meta crawl your product pages for ads and business products.
“You are solely responsible for and assume any and all risks associated with such AI features and any resulting Content”
Foley Hoag says California SB 1050, effective January 1, 2027, requires ads with a prominent AI synthetic performer, including an AI voiceover alone, to disclose it clearly. Violations count as false advertising.
“this performance features a synthetic performer”
The IAB's voluntary 2026 framework asks for labels on AI-generated images and video and synthetic people that could mislead, not on routine retouching or ad copy, and cites a 31.5% click-through drop from labeling.
“regardless of subsequent human refinement, editing, or compositing”
Reed Smith says that from August 2, 2026 the EU AI Act requires a clear, visible disclosure on deep-fake images, audio or video at first exposure, and ads are unlikely to qualify for lighter rules.
“Advertising is unlikely ever to qualify.”
Venable lawyers say New York has required a conspicuous disclosure on ads using AI-generated human performers since June 9, 2026, with no set wording. Audio-only ads and translation-only AI use are exempt.
“not recognizable as any actual person”
A UK test-and-control survey of 3,000 people on audio ads found 3% brand uplift for neutral AI and human voices alike, and 9% for regional-accent AI voices. It measures brand lift for audio, not sales or Meta video.
Common Thread Collective says standard edits like cropping or brightness do not trigger Meta's AI label, and outside detection relies on C2PA metadata. No data, and Meta has published none on performance.
“Meta has not published data suggesting the AI info label directly reduces ad performance.”
Meta automatically labels ads made or significantly edited with its own generative AI tools, and third-party AI content when it detects C2PA metadata; the label sits in the About this ad menu.
“automatically label ad content that has been created or edited using some of Meta's generative AI tools”
In a 900-person study reported by Marketing Dive, positive brand attitude was 44% with a full-length AI icon, 51% with no label, and 49-51% with text labels. 60% want disclosure when AI simulates humans.
“An AI icon elicited the worst brand attitude score, with just 44% of participants saying they had a positive attitude toward the brand.”
An AMCIS 2026 lab study found AI-origin labels on ads pull attention from persuasive content, and wording like AI generated carries steeper authenticity costs. Only the abstract was read.
A Mediascience experiment with 900 US respondents found no significant difference from an unlabelled control on brand choice, ad memory or brand attitude when AI-made video ads carried a disclosure label.
“if the creative is good, disclosure does not hurt it.”
The FTC's review rule, in force since October 2024, bans writing, creating or selling reviews or testimonials that misrepresent that the reviewer exists, used the product, or what their experience was.
“for a business to write, create, or sell a consumer review, consumer testimonial, or celebrity testimonial that materially misrepresents”
In the full answer
For readers who sign in- What to do this week3 steps
- The numbers behind it3 figures from cited sources
- The strongest disagreementThe best case against the answer, from a cited source
- How sure we areWhat its confidence rests on, and what would change the answer