AI Casting
Talent consent for an AI avatar in an ad: the ECHO model
Ogilvy and CAA launched ECHO, where talent approves every ad. Here is what it teaches brands about consent for an AI avatar, and how to build it into casting.
18 November 20264 min

If a testimonial's face appears in 40 versions of an ad, someone has to have said yes to all 40. Talent consent for an AI avatar in an ad is not a signature collected at the start of a project. It is a process that runs until delivery.
On 9 November 2026, Ogilvy and CAA unveiled ECHO, and the case shows how that process can be built. Our position is simple: consent should be treated as a stage of casting, with rules, checks and named owners, not as an appendix to the contract.
Consent is designed before the shoot
ECHO starts from a real shoot. As Ogilvy's announcement describes it, the platform turns a single celebrity or ambassador production into a content system, inside a governance framework meant to protect talent rights and IP. The models work from the talent's high-fidelity scans stored in the CAA Vault.
What matters for a buyer is the order of operations. First a legitimate source: an authorised shoot, with face and voice captured for a declared use. Then the variations. Reverse that sequence, generating first and asking permission later, and you may end up with material you cannot publish.
One general yes does not cover every version
The most interesting part of ECHO is repeated approval. As Marketing Dive reports, generated assets are automatically checked against machine-readable talent usage terms and brand safety guidelines, and nothing can be released without final sign-off through an approval portal.
These are two distinct layers, and it helps to keep them apart:
- Automated checks: stop what falls outside the contract (excluded product categories, territories, duration, unwanted contexts).
- Human approval: the talent or their agent sees the result and decides.
The first filters, the second decides. A single layer leaves a gap: either people eyeball what a rule would have caught instantly, or a consent nobody really gave gets automated.
From 8 to 50 assets: more versions, more approvals
Here is a number. As The Hollywood Reporter reports, Ogilvy and CAA say a traditional shoot of eight to ten assets can be expanded to roughly 30 to 50, for example to follow TikTok trends. Every consumer-facing asset carries a visible label: "[Talent Name] Approved | Recreated with AI".
The arithmetic for a brand is immediate. If each version needs approval, 50 versions mean 50 sign-offs to schedule. You need agreed response times, one contact per side and a clear review format. Without that logistics, the bottleneck moves from production to consent.
There is a communication upside too: a label naming the talent signals that the AI use is authorised. In the research Ogilvy presented, over 76% of respondents said explicit talent approval strengthens an ad's appeal. The study was commissioned by the agency itself, so we read it as an indication, not a guarantee.
How we handle consent in AI casting
In our AI Casting service, consent goes into the brief, not after it. With the brand and the talent we define in writing which uses, formats, territories and durations the likeness is licensed for, and which contexts are excluded. We then produce the versions and submit them for approval before delivery, keeping a record of each step. The same rule applies to real people such as founders or employees: informed consent and use limited to the project. For more on digital identities in communication, this article covers what brands can take from it (in Italian).
What to ask a supplier before you sign
ECHO has only just been announced. As ContentGrip notes, the service is commercially available through WPP and pilots are under way, but named client results have not been disclosed. So it is a good model of method, not yet proof of results. Whoever your partner is, we would ask:
- Where does the face data come from: an authorised shoot of the talent, or third-party material?
- What does the licence say: uses, duration, territories, excluded categories, and what happens if the talent wants to withdraw.
- Who approves, and how: every version or only the master? On what timeline?
- Which checks are automated: and who is accountable if something slips through?
- How is AI use disclosed: label, placement, languages.
- Who owns the rights to the output: a separate question from consent, covered in our piece on copyright (in Italian).
If you are planning an ad with a real face recreated using AI, start from this list and bring it to the first meeting. If you want to set it up with us, tell us about the project on the AI Casting page.


