AI for Marketing and Advertising Agencies

It depends on the jurisdiction and on how much human authorship was involved, and it is almost certainly not addressed in your current contract. Clients generally assume they own what they paid for, and the intellectual property position on generated material does not map cleanly onto that assumption. This belongs in the engagement terms before delivery rather than in a dispute afterwards. The related question worth settling at the same time is whether you may fine-tune on client brand assets and performance data, which is often what makes the output good and which most agreements did not contemplate.

It applied from 2 August 2026 and was not deferred by the Omnibus that pushed high-risk obligations to December 2027. For agencies it reaches disclosure where people interact with AI systems and the marking of certain generated content, which touches client-facing chat, generated imagery and synthetic media. The practical requirement is knowing which of your outputs were generated and being able to say so, which makes it an asset provenance and production tracking problem before it is a legal one. Agencies that cannot answer that question are the ones with a compliance problem.

Client reporting automation, and agencies keep skipping it. Every month account teams pull the same numbers from the same platforms, rebuild the same decks and write similar commentary, largely unbilled. Automating the assembly and the first-draft commentary — with the account team supplying interpretation — returns hours nobody was charging for. There is no rights exposure, no claims risk and no client approval gate, which is exactly why it loses budget to creative applications that carry all three.

Within an encoded brand system, yes, and the review model matters as much as the generation. Encode tone, vocabulary, prohibited claims and mandatory disclosures as constraints the generation cannot leave, and put a hard gate on regulated claims — financial, health, environmental and comparative — because a generated claim nobody checked becomes a compliance incident with your client's name on it. The failure mode people underestimate is that generation scales and review does not, so a thousand variants is trivial to produce and impossible to check unless the review workflow was designed alongside.

Where you bill for production hours, yes, and agencies have felt this earlier than the rest of professional services. The responses that work are moving production to an output or retainer basis, using the capacity for volume that was previously uneconomic, or repositioning on the strategy and selection judgement that generation does not supply. The response that does not work is holding the hourly production line and quietly delivering it faster, because clients who use the same tools know roughly what it now costs.