Legal wants an AI-label on everything, marketing on nothing
Both silos are right about their own risk, and together they get it right
86% of people say content should clearly state when AI made it. 32% would trust a brand less once they know. Your readers want to know, and many hold it against you.
Your legal team wants a label on everything AI touched. Your marketing team quietly stopped asking. Both are protecting something real, each from their own side of the table.
Stop letting either silo decide your AI labels alone. Legal says what must carry a label by law. For everything else, marketing brings one question to the table: would our reader feel misled if they knew AI was involved? Agree on that once, per type of content, and sign it off together. From then on, whoever approves a piece checks it against that list. Legal only sees the doubts.
It flips when much of your copy is public interest by default, as in finance or health. Then legal keeps a say per piece.
Most AI-generated copy doesn’t need a label
Deepfakes need no discussion: AI-made images, audio and video that could pass for real always get a label.
Copy is where it gets unclear. The rule is Article 50 of the EU AI Act. Text only needs a label when you publish it to inform the public on a matter of public interest: politics, public health, justice, consumer safety, or an economic, scientific or cultural debate.
Even then, you don’t need a label if a person with real editorial responsibility reviewed it. A quick read doesn’t count.
Most of my own copy is AI-generated. This piece too. But I steer the AI, do the reasoning, review and correct every line, and decide what goes out. That’s real editorial responsibility, so it needs no label. It’s in my editorial policy, so anyone can check. And because of my process, I hold the copyright. Copy written by AI alone isn’t protected.
→ For images, audio and video, the law is clear. For copy, the law leaves most of the decision to you.
Legal and marketing both have a point
What breaks trust most is feeling misled: 63% of people name it, more than any other reason. That’s where legal and marketing need each other.
Legal’s caution makes sense. A missing label can cost up to € 15 million or 3% of global turnover. And when a piece goes wrong, legal gets asked why it signed off. From where legal sits, labeling everything is the only safe answer.
Marketing’s resistance makes sense too. Across 13 experiments, people who disclosed their AI use were trusted less than people who didn’t. But getting caught was worse. The trust penalty was bigger when AI use was exposed than when people disclosed it themselves.
→ Label everything, and you pay the trust penalty on every piece. Label nothing, and you risk the bigger trust penalty for getting caught.
Decide it once, together
3 easy steps so legal keeps control without reading every post, and your reader’s interest comes first.
Step 1: Must label. What the law requires, as set out above. Legal owns this part. Marketing doesn’t negotiate it.
Step 2: The reader trust test. Everything else gets one question: would our reader feel misled if they knew AI was involved? That depends on what the reader trusts the piece for.
- A person: a CEO letter, a personal opinion, “I tested this.” The law often doesn’t require a label here. But the reader trusts the voice, the experience, the “I”.
If AI wrote it and the person only signed it, that trust is misplaced. So label it, or have the person write the brief from their own experience, review the copy closely and approve the final version.
- Expertise: advice, analysis. The reader trusts the expert’s judgment. If an expert reviewed the AI-written copy, had the final say and puts their name on it, the judgment is human. Even on a public-interest topic, that human judgment earns the exemption. No label needed.
- Information: product pages, campaigns, FAQs. This copy sells or explains your product, so the label rule usually doesn’t apply. Nobody feels misled by a correct product page written with AI. No label.
Step 3: Sign it off once. Legal and marketing map your content types against steps 1 and 2, in one session. Both sign off. Where you still can’t agree on a content type, you decide, together with your General Counsel. Put the result in your editorial policy, so anyone can check it.
From then on, whoever approves a piece checks it against that list. Only a new content type, or real doubt, brings both back together.
Start here
Take your last 5 AI-assisted pieces. Ask your General Counsel and your marketing lead to mark each one: label or no label. Then compare.
Where they differ, one silo made the call on its own, without putting your reader first.
If legal and marketing still disagree on AI, I’ll get them in one room. Mail me.
9 AI dilemmas
This is the seventh of 9 AI dilemmas for the C-suite. I'm publishing them over the coming weeks. Each piece names the decision, what it costs to leave it open, and what I'd do.
The one you recognize immediately might be the one you've left open too long.
- Faster with AI means more errors
- Secure AI means weaker AI
- Your board wants AI to make marketing cheaper. Your brand can't afford it
- Want the AI results you read about? Look at your workflow, not your tool
- An AI license costs less than 1% of a salary. Too many boards still say ‘no’
- You can delegate AI. You can’t delegate understanding it
- Legal wants to AI-label everything, marketing nothing (this page)
Sources
- EU AI Act, Article 50
- Meltwater and YouGov, Trust in the Age of Generative AI (April 2026), YouGov survey of 9,869 adults in 7 markets
- Schilke, O. & Reimann, M. (2025). The transparency dilemma: How AI disclosure erodes trust. Organizational Behavior and Human Decision Processes
- European Commission, Guidelines on transparency obligations under Article 50 (20 July 2026)
- ActuIA, AI Act obligations after Omnibus 2026/1744
- You probably don’t need the AI-generated label
- Who signed off on that? Most brands can’t say
Join 2,750+ professionals
AI changes how the work gets done. I test it in my own work and write about what actually works.
Free. Unsubscribe anytime.