AI has already transformed creative production. An image that once required a photographer, studio, talent, location and several days of post-production can sometimes be generated in minutes. Designers can create campaign environments without organising a shoot, social teams can turn a single asset into dozens of variations, video teams can generate footage that would previously have required significant production budgets, and copywriters can move from a blank page to a workable first draft almost instantly.
For the last few years, the commercial argument has therefore been relatively straightforward: AI can make creative production faster and cheaper. As of August 2026, that calculation has become more complicated because the transparency provisions of Article 50 of the EU AI Act began applying on 2 August 2026, while California’s AI Transparency Act has introduced its own requirements around identifying and tracing AI-generated media. Social platforms are also developing their own systems for detecting and labelling content created or significantly edited using AI.
For creative agencies, internal marketing departments and social teams, this does not mean that every use of AI suddenly requires a large “AI-generated” warning. It does, however, mean that businesses increasingly need to understand how their assets were produced, what information may travel with those assets, what a platform might disclose automatically and, in some circumstances, what the audience needs to be told explicitly.
The question for marketers is therefore changing from simply asking whether AI can make something faster to asking whether the business is comfortable with customers knowing how that content was made. That is not simply a compliance question; it is a brand, marketing, production and financial decision.
Legal disclosure, platform labelling and provenance are not the same thing
One of the reasons the current debate around AI labelling has become confusing is that several different things are being described as though they are one and the same. For marketing teams, it is much more useful to separate machine-readable provenance, legal audience disclosure and platform labelling.
Under Article 50 of the EU AI Act, providers of generative AI systems have obligations to make synthetic outputs detectable through machine-readable marking, subject to exceptions including systems that are simply performing standard assistive editing. There are then separate obligations for businesses using certain types of synthetic content, particularly deepfakes that resemble real people, objects, places, entities or events and could falsely appear authentic or truthful. Where that audience-facing disclosure obligation applies, simply having invisible metadata within a file is not enough, because the audience needs an understandable and perceivable disclosure.
Social platforms add another layer because they are developing their own rules and technical systems for identifying AI-generated content. Their approach does not necessarily map perfectly onto the legal definition of a deepfake, which means a piece of content may not require a specific legal disclosure from the brand but could still acquire an AI indicator when it reaches a social platform.
This distinction is important because marketers should not think of AI transparency as a single label that appears only when a brand has created a realistic fake photograph. The ecosystem is becoming considerably broader than that.
AI labels are not only about photorealistic fake scenes
A lot of attention around the new rules has understandably focused on realistic synthetic content, including fake people, fake locations, cloned voices and video depicting events that never happened. Those are obvious cases where transparency matters, but they are not the only types of content that may now carry signals showing that AI was involved.
The EU’s audience-facing deepfake disclosure requirement focuses on content that could falsely appear authentic, but the separate requirement for providers to make synthetic outputs machine-readable is broader. This means that a highly stylised illustration, generated graphic, synthetic background or other clearly creative piece of work may still contain information indicating that generative AI was used, even when nobody could reasonably mistake it for an untouched photograph.
Social platforms can use those signals to apply their own labels. TikTok, for example, says it may automatically apply an “AI-generated” label to content it identifies as completely generated or significantly edited using AI, including through Content Credentials based on the C2PA provenance standard. Meta also uses industry-standard signals to help identify content generated or edited with AI and has expanded the way AI information can appear around organic and advertising content.
Are brands comfortable with customers knowing they use AI?
A business may be perfectly comfortable with its agency or internal creative team using AI behind the scenes, particularly when it has already benefited from faster production and lower costs. The more difficult question is whether that same business is comfortable when customers can see that AI was involved.
An “AI-generated” or “AI info” indicator could have very different implications depending on the brand and the context. A technology company may be entirely comfortable talking openly about its use of generative tools, while a luxury brand built around craftsmanship may take a different view. A food company may care about whether customers believe its product photography represents a real product, while a property company may need to consider whether buyers understand which images represent completed spaces and which are synthetic visualisations.
There is unlikely to be one answer that works for every business. Some brands will be comfortable being transparent about AI and may even choose to make that part of their story, while others may decide that certain categories of content should continue to be photographed, filmed, illustrated or produced conventionally.
Those are not simply legal questions. They are brand, marketing and financial decisions because the choice of production method has a direct impact on cost, turnaround time, resource and potentially how the finished campaign is perceived.
The budget versus brand trade-off
The commercial tension becomes clearer when you consider the alternative production routes available to a creative team. A campaign image might take ten minutes to generate using AI, and if the client is comfortable with synthetic production and any associated transparency, that efficiency remains extremely valuable.
However, if the client decides that it does not want its campaign appearing with an AI indicator, the creative team may need to take a completely different route. Instead of generating a model, it may need to cast one. Instead of generating an environment, it may need to hire a location. Instead of generating a product visual, it may need photography or CGI. Instead of synthesising a voice, it may need to book voice talent and secure usage rights. Instead of generating video, it may need a crew, equipment, talent, lighting, editing and post-production.
The cost difference between those production routes can be considerable, which means the conversation should no longer stop at asking whether a client is comfortable with its agency using AI. The more commercially useful question is whether the client is comfortable with its customers knowing AI was used and, if not, what additional time and money it is prepared to invest in an alternative production route.
This may be one of the first moments when businesses begin to see the true cost of AI more clearly. AI can deliver enormous production savings, but those savings need to be considered alongside the cost of review, provenance, approvals and the possibility that a brand may ultimately decide not to use synthetic production at all.
Work that recently took minutes may start taking longer again
There is another change that creative and marketing teams need to discuss now. Even where AI remains acceptable to the brand, the 30-second generation is not necessarily a 30-second production job anymore because somebody may need to establish what was generated, identify the tool used, preserve relevant provenance, consider whether a disclosure is required, check rights or consent issues, obtain approval and retain enough information for the decision to be understood later.
For one social post, this may add very little time, but across hundreds of assets, multiple campaigns, markets, freelancers, agencies and channels, it can become a meaningful production overhead.
Creative teams therefore need to speak to their marketing teams as soon as possible. If marketing has become accustomed to the speed and efficiency that AI-enabled workflows have created, it needs to understand that the transparency and provenance environment may introduce additional production steps. If the organisation then decides that it does not want certain types of AI-generated content at all because of how that use may be communicated to customers, the impact can be greater still, with work that recently took minutes potentially returning to a production schedule measured in hours or days.
Those changes need to be reflected in deadlines, resource planning and budgets rather than being silently absorbed by creative departments or agencies.
Agencies need to agree AI policies with their clients before production starts
There is a contractual dimension to this as well, particularly because the organisations using AI systems professionally may themselves carry obligations rather than being able to assume that all responsibility sits with the technology provider.
Agencies should therefore review their terms, scopes of work and production agreements with appropriate legal advice so that they are clear about how generative AI may be used, who approves its use, what provenance information is retained and supplied, who is responsible for relevant disclosure decisions, how rights and consent are handled, and what happens to timings and costs if a client changes its position on AI after production has begun.
This does not need to become a frightening page of disclaimers attached to every brief. The purpose is simply to remove ambiguity around the production method being used and the commercial implications if that method changes.
Internal marketing teams need the same conversation with their creative teams
The issue is not limited to agency-client relationships because internal marketing departments face exactly the same challenge. A brand cannot realistically have one AI position sitting inside a corporate policy while the designers, social teams, video producers, freelancers and external agencies producing hundreds of pieces of content are all operating according to different assumptions.
Marketing teams therefore need to align with their creative and production teams around a practical policy that explains what can be generated, what AI can assist with, what types of work need additional review, how the organisation feels about platform AI labels, what information should be recorded and who makes the final decision when the answer is unclear.
They also need to agree the production consequences of that policy. If marketing does not want certain types of synthetic content because it is uncomfortable with customers seeing an AI label, the creative team needs to know this before work begins, and the marketing team needs to understand that the alternative may require more time and a larger budget.
The purpose is not to make every use of an AI-powered Photoshop feature a compliance exercise. It is to make sure that the people planning the work and the people making the work are operating from the same assumptions.
Creative teams need a simple provenance system
For most organisations, the sensible response will be a lightweight internal system rather than an enormous archive of every prompt and rejected generation. The purpose of provenance recording should be to ensure that the knowledge of how an important asset was created does not disappear when the person who made it hands it over to somebody else.
A simple internal classification could include:
- Human-created: The asset was conventionally produced, even if routine assistive AI tools were used for minor editing.
- AI-assisted: Generative AI materially helped with the production process, but the underlying human-created content remains central.
- Synthetic or materially AI-generated: AI created or substantially altered important parts of what the audience ultimately sees or hears.
- Review required: The nature of the asset means disclosure, rights, provenance or brand suitability needs to be checked before publication.
These are not legal classifications. They are simply useful production labels that allow somebody opening the asset to understand quickly how it was made and whether anything else needs to happen before it is published.
Figma is a sensible place to start recording provenance
For many design teams, Figma is an obvious place to introduce this because it is already where the asset is being produced, reviewed and approved.
A small status component could sit beside the final frame rather than being included within the artwork itself. An AI-assisted asset might state that generative fill was used to extend an existing studio background while the original model and product photography were retained. A synthetic asset might state that the model and environment were generated and that the final disclosure decision needs review before publication.
The status is an internal production note rather than a customer-facing label, so it should not accidentally appear inside the exported creative. Its purpose is to allow the designer, producer, account manager, marketing manager or client to understand what happened during production.
That information can then move into the approval system, campaign record or digital asset management system alongside the finished master, creating a simple production chain from brief to production, provenance review, approval and publication.
What provenance records should creative teams actually retain?
Creative teams do not necessarily need to retain every generation and every prompt forever, and the EU rules do not prescribe a universal creative-production archive that every marketing department must maintain for a fixed number of years.
A more proportionate approach is to retain enough information around significant AI-assisted or synthetic assets for the organisation to understand how they were created and why a particular publishing decision was made. Depending on the asset, that might include:
- the final asset and campaign it belongs to;
- its internal AI classification;
- the AI system or tool used where this is materially relevant;
- what part of the asset was generated or significantly altered;
- the level of human creative input or review;
- any relevant rights, consent or third-party source considerations;
- whether audience disclosure was considered or required;
- whether machine-readable provenance or Content Credentials are present;
- who approved the final asset;
- and which version ultimately went live.
The objective is not to maintain a forensic record of every failed generation. A much more useful test is whether the business could answer, six months later, how an important asset was made, what AI contributed, who reviewed it and why it was considered appropriate to publish.
What should agencies provide to clients?
An agency does not necessarily need to hand over every prompt, rejected generation, Figma annotation and internal discussion whenever it delivers a campaign. There should be a distinction between the agency’s deeper production record and the information the client needs in order to understand, publish and reuse the finished work responsibly.
For significant AI-assisted or synthetic assets, a simple client-facing record could explain the AI status of the asset, what generative AI contributed, whether a disclosure requirement was considered, whether relevant provenance has been retained and whether there are any restrictions or actions the client needs to understand when republishing the work.
The agency can retain deeper production evidence internally according to its own governance procedures and contractual arrangements. The principle is that clients should receive enough provenance information to understand and responsibly use the creative without turning every asset handover into a full technical audit.
Do provenance records need to be uploaded to social platforms?
The internal production records described above are not generally something a marketing team uploads to Instagram, TikTok or another social channel. Figma annotations, prompt histories, internal AI statuses and approval records exist to help the business understand and govern its own production process.
Audience disclosure is a separate requirement. Where a particular type of synthetic content requires an audience-facing disclosure, that disclosure needs to be presented appropriately when the content is published. Machine-readable provenance is another layer again because some information can already travel inside or alongside the asset itself and can potentially be detected automatically by the platform receiving the file.
California’s AI transparency requirements illustrate the direction of travel because covered generative AI providers are required to include latent disclosures in certain AI-generated image, video and audio where technically feasible, while qualifying large online platforms will have additional obligations from January 2027 around detecting recognised provenance standards and making relevant information available to users.
The future workflow is therefore unlikely to involve a social media manager manually uploading a separate provenance spreadsheet alongside every post. Increasingly, some provenance can travel with the file, while the business retains a separate internal record and adds any audience-facing disclosure that is required or appropriate.
What about AI content that brands have already published?
Businesses do not need to panic and begin manually relabelling years of social content simply because Article 50 has come into effect. The European Commission has stated that content generated before 2 August 2026 does not need to be labelled retroactively, although it encourages greater transparency around older content where practical.
This means that a brand which published an AI-generated Instagram asset in 2025 is not suddenly required under Article 50 to reopen every historic post and add a disclosure.
However, the legal requirement and a social platform’s own detection system are separate things. Platforms may already have technical signals allowing them to identify some AI-generated or significantly edited assets, so it is possible for platform labelling to operate differently from the legal requirement. That does not mean every historic AI post will suddenly be identified and labelled, because the platforms do not claim that every old asset can be reliably detected.
The more important issue for marketing teams is what happens when an old asset is reused. If a synthetic campaign image created several years ago is brought back into a new campaign, it makes sense to put that asset through the organisation’s current AI review process before publishing it again.
This is another reason why provenance records become more valuable over time. A creative team opening an old campaign folder in several years should ideally be able to tell whether the person, location or environment was photographed or generated without having to track down whoever happened to make the original file.
AI-written marketing copy is a slightly different issue
caption, blog post or social update that began in an AI writing tool to carry a disclosure.
The relevant Article 50 requirement concerns AI-generated or manipulated text published for the purpose of informing the public about matters of public interest where there has not been meaningful human review or editorial control. Genuine human editing and responsibility therefore remain important.
For marketing teams, the practical principle should already feel familiar. AI can be useful for research, ideation, structure, alternatives and creating a first draft, but somebody who understands the subject should still review, challenge, fact-check and ultimately take responsibility for what the brand publishes.
Human involvement is therefore not becoming less important as AI becomes more capable. In some areas, it is becoming more important.
Copyright adds another reason to understand how the work was made
Transparency is not the only reason to understand where human work ends and AI generation begins. Copyright and intellectual property also matter, particularly when a business is creating major campaign assets that it expects to own, license, defend and reuse for years.
The US Copyright Office has concluded that using AI as an assistive tool does not prevent copyright protection, while purely AI-generated material without sufficient human authorship does not receive copyright protection simply because somebody supplied prompts. Copyright law varies by jurisdiction and these questions need to be assessed carefully, but the broader commercial point is straightforward.
There is a significant difference between generating a disposable social graphic and generating the hero creative for an international brand campaign.
AI is therefore becoming an intellectual-property decision as well as a production-efficiency decision.
Perhaps this is when we discover the true cost of AI
For the last few years, discussions about the economics of generative creative have largely focused on how quickly an asset can be produced and how much traditional production cost can be removed.
The transparency era makes that calculation more sophisticated because businesses also need to consider the time involved in reviewing assets, documenting provenance, obtaining approvals, assessing potential disclosures and deciding whether customers seeing an AI indicator affects the value of the finished work.
In many situations, AI will still be dramatically faster and cheaper than conventional production, even after those additional steps are accounted for. The point is not that AI has suddenly become inefficient.
The more interesting point is that the business can now make the decision with a clearer understanding of the trade-off.
If a brand is comfortable with synthetic production and transparent about how it uses AI, it may continue to enjoy significant production efficiencies. If it is not comfortable with its customers knowing AI was involved, then the alternative may require traditional production methods, longer timelines and materially larger budgets.
That is why the new question for marketing teams is not simply whether AI is worth using. It is whether the time and money saved by AI remain worth it once transparency, provenance and brand perception are included in the calculation.
This is the moment to agree the process
The businesses that handle this well are unlikely to be those that ban AI entirely, and they are equally unlikely to be those that use it everywhere without considering the implications. The strongest approach is to decide the rules before production starts.
Creative agencies should talk to clients about what forms of AI production they accept, how they feel about customer-facing transparency, what provenance information they want and how a decision not to use generative production could affect timing and budget. They should also make sure their scopes and contractual terms reflect the way creative work is actually being produced.
Internal marketing teams need to have the same conversation with their designers, content teams, social teams, video teams and production partners because workflows that have become dramatically faster through the use of AI may now require more review and record-keeping, while a decision to avoid synthetic production altogether may slow those workflows down considerably.
Creative teams should introduce a proportionate provenance process so that significant assets can be understood without turning every piece of work into a compliance exercise. Marketing teams, in turn, need to recognise that these additional steps take time and that the financial advantages of AI depend partly on the organisation’s appetite for transparency.
Most importantly, businesses need to understand that legal disclosure, machine-readable provenance and platform labelling are not the same thing. A brand may not itself be legally required to place an AI badge on a particular stylised graphic, but the asset may still contain information identifying its origin and a social platform may choose to surface that information to users.
That brings the issue back to three questions that creative and marketing teams should now be asking before they start production: Are we comfortable with how this was made? Are we comfortable with our customers knowing how it was made? And, if we are not, what are the implications for the brand, the timeline and the budget?
AI is not going away, and neither is the extraordinary creative potential it offers. What is changing is the assumption that its role in production will remain invisible.
The new transparency environment may therefore represent something more significant than another compliance requirement. It could mark the beginning of an era in which the provenance of creative work becomes part of the creative and commercial decision itself.
This article provides general information about emerging AI transparency and copyright requirements and is not legal advice. The application of these rules depends on the content, organisation, use case and jurisdictions involved, and businesses should seek appropriate legal advice where required.
