Every brand already has an AI creative policy. The question is whether you wrote it or the market wrote it for you.
I have watched this cycle before. In the mid-nineties the question was photoshopping, and the brands that worked out an answer early (Dove among them, Aerie and Target a decade later) spent the next ten years using the answer as part of their positioning. The ones who did not were periodically caught, and the apology was always more expensive than the policy would have been. The pattern repeated with influencer disclosures. It is repeating again with generative AI in creative work, and this time the clock runs faster: an AI-made asset can be posted, screenshotted, and litigated in a single news cycle.
When I ask a marketing lead whether their team has written down what AI is allowed to do inside their creative output, the answer is usually no. Whether the agency has is usually no as well. That silence is the policy by default: the market picks the position, and the brand reads about it in the trade press.
Alessandro Manfredi, then Chief Marketing Officer at Dove, published his version of the line on 9 April 2024. His framing is still the clearest working example of a brand treating the AI question as a positioning decision rather than a tooling decision, and the eighteen months since have handed us three other brands who did the opposite, with three different bills to pay.
“Not Algorithms,” Said the Brand That Was Ready for the Question
In the press release Dove issued on 9 April 2024, Manfredi put it plainly: “At Dove, we seek a future in which women get to decide and declare what real beauty looks like, not algorithms.” He added that the pledge was “just one step.” Marketing Dive’s Peter Adams confirmed the policy the same day, with Brazil-based agency Soko on the lead campaign.
Read that quote twice. The operational problem Manfredi is naming is not whether AI tools are good or bad. It is who gets to decide what the brand looks like when the tool is in the room. His answer, in public, before anyone caught Dove doing anything, was that the brand gets to decide and the tool does not. That is a positioning statement first, a creative constraint second, and a tool choice a distant third. Most brands have inverted that order: they have told their creative teams which tools to use (usually by silence, letting the agencies pick) without telling them the brand’s position. The position is being written anyway, one ad at a time.
The Scale of the Question You Have Not Decided
Here is what we know on the record. Digiday’s Kimeko McCoy reported on 10 November 2025 that at least five creators across different follower sizes had confirmed their 2025 brand deals carried no AI, intellectual property, or copyright clauses. Danielle Wiley, CEO of the influencer shop Sway Group, told the reporter it was “strange to me how little my clients are talking about it.” Joseph Lawlor, an advertising and IP lawyer at Haynes and Boone, said creators and brands were still relying on existing name, image, and likeness language that was never written with generative AI in mind.
On 15 June 2026, The Fashion Law reported the predictable consequence. Influencer Molly Tranchin, who posts as FashionVeggie, had filed Tranchin v. EBY, Inc. (No. 3:26-cv-05559) in the Northern District of California on 9 June 2026. Her complaint alleges that EBY, the intimates brand backed by Sofia Vergara, posted an AI-altered video showing her partially nude in one of its sheer bras, without her knowledge, through a whitelisting arrangement that gave her followers reason to think she had posted it. The claims span California’s non-consensual deepfake statute, defamation, copyright, breach of contract, right of publicity, and more. These are allegations and have not been tested in court. The point for a brand buyer is that the contract Tranchin signed with EBY was almost certainly the standard shape of a 2025 influencer deal.
Between those anchors sit two more public positions worth examining. Vogue ran an AI-generated model in a Guess ad in its August 2025 US print issue; Marketing Interactive’s 29 July 2025 coverage captured both the backlash and Vogue’s internal clarification that the ad “was approved by the advertising department, not the editorial department.” In a different direction, Mattel announced a formal OpenAI partnership on 12 June 2025, with Chief Franchise Officer Josh Silverman confirming that Mattel does not license its intellectual property to OpenAI and keeps full control over anything produced. Four brands, four approaches, and the cost shows up in different ledgers. Dove paid in strategy work. Mattel paid in contract architecture. Guess paid in a week of trade-press scrutiny. EBY is paying in court.
Lever One: Publish the Position Before Someone Else Writes It
Manfredi’s instinct is correct on the mechanics as well as the ethics: a brand position stated in public is a cheap purchase. It costs you optionality (Dove gave up the right to use generative AI to depict women in advertising) and buys you a defence against the next twelve months of accidents. The Dove pledge has been treated as the industry benchmark for eighteen months, and I have not seen a competitor successfully argue they match it without putting their own statement next to it.
Mattel’s position is also a published one, in a quieter register. Silverman’s line to TechCrunch that “Mattel is not licensing its intellectual property to OpenAI” is a positioning sentence as much as a legal one. It tells a parent, a regulator, and a competitor where the asset boundary sits. If you are briefing an agency to work with a frontier model and you cannot write a sentence that short about where your IP goes, you have not finished the policy.
Lever Two: The Contract Clause Your Creator Deal Does Not Yet Carry
Wiley told Digiday that Sway Group had started writing explicit anti-deepfake language into its creator contracts about two years ago, ahead of what Lawlor says most deals still carry. Alessandro Bogliari, co-founder and CEO of Influencer Marketing Factory, said his agency would start considering clauses that require influencers to appear in their own content and disclose any AI-generated material. These are not hypothetical. They are the missing paragraphs the EBY complaint relies on. A clear “no AI alteration without written consent” clause in Tranchin’s contract would have made the suit a different and much shorter document.
For a brand buyer: ask your head of marketing to pull a signed influencer agreement from the last ninety days and search it for “artificial intelligence,” “generative,” and “synthetic.” If none appear, your position on AI alteration of a creator’s likeness is whatever the standard rights language implies, which is to say undefined. Steven Stein of Greenberg Glusker’s entertainment litigation team told Digiday the exposure runs in both directions, so the clause protects the brand too.
Lever Three: Who Approves the Ad Needs an Answer in Writing
The sharpest line in the Guess story is not from Guess. It is from Vogue, which told reporters that the AI ad “was approved by the advertising department, not the editorial department.” That is a correct, careful, legal statement, and it is also an admission that no single person inside the publication had read the ad against a brand-safety policy that explicitly covered AI. In a publisher, that gap gets the publisher criticised. In a brand, the same gap gets the brand sued.
The fix is boring and I would still argue for it. Any paid creative that uses generative AI anywhere in its production chain, including stock imagery with AI components, agency-side composition, or voice synthesis, needs a named approver before it goes live. That approver is not the agency account lead, and it is not the junior person running the ad server. It is the same person who signs off on brand voice. If you cannot name them in a sentence, write the sentence.
Lever Four: The Governance Seat Nobody Has Filled
In the Digiday follow-up to the Dove pledge, Jeff Rosenblum, co-CEO at Questus, called the commitment “an absolutely ethical maneuver to make,” then added the honest second half: most brands will need AI, so the real question is how to use it. Marty Senn, Chief Creative Officer at Carmichael Lynch, framed Dove’s move as the trigger for an internal conversation about whether AI delivers on each brand’s promise. Those are two different jobs: the first is a position paper, the second is weekly operational governance. Most brands I meet have neither owner.
My recommendation is to put the second job somewhere senior enough that “we do not do that” sticks when a creative team asks for an exception. In practice that is usually the CMO or a brand governance committee reporting to the CMO. I would not route it through legal alone, because legal protects against the lawsuit after it exists and will not design the position before. Marketing’s job is the design.
The Lever That Is Not Working Yet: Enforcement After the Fact
Honesty requires naming the lever nobody has made work. If a bad AI ad goes out, pulling it back is slow, and the damage during the delay is exactly what you were trying to prevent. The EBY case tests California’s non-consensual deepfake statute with precedents still unsettled. Takedown via platform policy is uneven: TikTok’s synthetic media guidelines, Meta’s manipulated media rules, and X’s authenticity policy each read differently, and none of them resolve a dispute between a brand and a creator in hours. The deterrent is still the policy you wrote before anyone had reason to come after you. The remedy afterwards is partial.
I would rather tell you we have cracked the enforcement side. We have not. Neither has the industry.
What This Looks Like Inside a Real Content Stack
The four levers above all pass through content infrastructure at some point, and this is where our side of the work lives. On a WordPress multisite that carries a brand’s main site plus a handful of regional storefronts, there are three concrete places the policy has to be legible to the system, not just to the humans.
The first is a provenance field on the media library. A post-meta pair on every attachment answering two questions (“Was any part of this asset generated by an AI tool? Yes / No / Partial” and “If yes, which tool and under which brief”) takes a few hours via a custom block and a REST field, and from that moment on every ad asset carries its own answer to an audit question. We write to _twc_ai_provenance, exposed in the block editor sidebar and surfaced in the Media screen column view. Half a day on a mid-market store, longer at enterprise scale because the field also feeds DAM exports.
The second is approval routing. WordPress’s own user roles plus a lightweight editorial workflow plugin (we use PublishPress Flow on client installs, PublishPress Permissions where the role mix needs tighter control) can enforce that any post or page carrying an attachment flagged as AI-generated or AI-partial requires a specific named role before it can move from Pending to Publish. That named role is the person who signs off on brand voice. The build cost here is the policy itself, not the plugin: deciding whose job this is, and whose job it is not, takes longer than configuring the gate.
The third is contract templates. Any site that publishes creator content (a brand hub, a press room, a creator landing section) is already generating PDFs or forms somewhere, and the contract clauses from Sway Group’s playbook belong in those templates, versioned, with the agency’s own approval field alongside. We treat them as content under revision history, not as a one-off document the legal team keeps in a shared drive. On a WooCommerce estate running an affiliate or ambassador programme, that is a Gravity Forms route with a signed PDF attachment and a webhook to the CRM. One to two days to ship.
None of this is glamorous and none of it earns a case study alone. All three together make the written policy enforceable the next time an agency ships an asset, which is the point.
If you buy a platform or an agency on the basis of someone’s deck, this is the plumbing that tells you whether you are buying a slide or a system. Ask whoever is pitching you how they would log AI provenance on an asset from a stock library. A blank look means the policy will be enforced by goodwill, which does not scale and does not survive a lawsuit.
The brands who did this work in the photoshopping era (Dove most prominently, Aerie and Target in the same company) spent the next decade not having to explain themselves. The ones who did not are still periodically explaining themselves. No reason to expect this round to be different, except that the clock is tighter. If you want a conversation about where the levers land on your own estate, that is the kind of work we take on.
Last modified: November 3, 2026
