A practical framework for AI disclosure in marketing

As an adjunct professor at Georgetown University, I (and my students) live under an AI disclosure policy. If you use generative AI — whether it’s to write, design, brainstorm or something else — and submit that work for a grade, you’d better disclose it. Fair enough. We talk about it in class, we work with it responsibly and we treat it like any other assistive tool.

Outside the classroom, the rules are murkier. Lately, I’ve been reviewing emerging laws around AI disclosure. It got me thinking: disclosure isn’t inherently the problem. But the way it’s being discussed and, more importantly, applied, often is.

Dig deeper: What privacy and email laws reveal about today’s compliance risk

To date, there’s no broad U.S. federal law that requires AI disclosure in marketing. But several states have introduced mandates in specific contexts: political advertising, employment screening, healthcare decision-making and chatbot interactions. Some are already in effect. Most social media platforms have stepped in, too, requiring or strongly encouraging creators to label AI-generated content.

When AI disclosure becomes noise

There’s a growing push, from platforms, regulators and even consumers, for marketers to disclose AI use more broadly. The concern? That AI-generated content could mislead, manipulate or undermine trust.

I’m on board with the spirit of that. Truly. I’ve got no issue disclosing when AI lends a hand. But the current vibe, where some people are calling for brands to slap a label on everything AI, is a bit like the backlash we saw over em dashes (too dramatic, too frequent). Not every use of AI needs a disclosure.

Yes, I follow the disclosure rules and I support transparency. But here’s my argument: we need to move beyond the binary of “always disclose” or “don’t disclose at all.” Instead, we need a continuum that’s based on context, consequence and audience impact.

Why a continuum model works better

If we want AI disclosures to mean something, to actually build trust, not just tick a compliance box, we need to apply a little more judgment and strategy. That means moving away from blanket disclosure rules and thinking instead about context, consequence and audience impact.

Context: Where and how is the AI being used?

AI tools are everywhere, from spell checkers to subject line testers to fully generative writing engines. But not all use cases are created equal. Context matters.

• Was the AI used behind the scenes (e.g., grammar fix, content outline)?
• Did it generate the content itself (copy, image, etc.) in a way that directly reaches the audience?
• Is this internal use (like segmentation or data modeling) or external, consumer-facing content?

Disclosure should be shaped by the role AI played, not just by its presence.

Consequence: Could this mislead or distort perception?

This is where the materiality test comes in. If the AI’s involvement changes how someone interprets the content, then disclosure matters more.

• Would the audience feel misled if they knew the image wasn’t a real person?
• Would they assume a human expert wrote this advice, when it was mostly machine-generated?
• Would nondisclosure cross a line legally, ethically or reputationally?

If the AI’s contribution affects trust, credibility or interpretation, that’s not a gray area, that’s a red flag.

Audience impact: Who’s on the receiving end, and what do they expect?

Different audiences bring different assumptions. What raises eyebrows in one context might feel totally normal in another.

• In an academic journal? Full citation required.
• In a marketing email? Readers expect curated content, but not necessarily full disclosure on whether the headline came from ChatGPT or a team brainstorm.
• On a political ad? Disclosure should be immediate, unmissable and enforceable.

Audience expectation shapes how disclosure lands and how necessary it is. When transparency adds clarity, great. When it’s just noise? Not so much.

Dig deeper: In an age of AI excess, trust becomes the real differentiator

Here’s how the disclosure continuum applies across common marketing scenarios.

Internal productivity or planning tasks

Use of AI to segment an email list based on engagement data

Sample prompt: “I’ll upload a spreadsheet with recency, frequency and monetary (RFM) data for each person on our email list. Please segment into groups based on this data.” 

• Context: Internal use, behind-the-scenes.
• Consequence: None to the end user. They’re unlikely to know or care that AI was involved.
• Audience impact: Zero. This segmentation could be done manually. AI just speeds it up.

Continuum model: No AI disclosure needed. I see this as akin to using any other analytics tool for segmentation. It increases your team’s productivity, but it’s invisible to the recipient. 

Caveat: AI-driven segmentation, a type of automated processing, likely triggers a disclosure obligation under GDPR and similar data protection regulations, since it involves personally identifiable information (PII). 

Use of AI to draft an internal creative brief for a consumer-facing campaign

Sample prompt: “I’ll upload information about this campaign. Please use it to develop a creative brief.” 

• Context: Internal document, not customer-facing.
• Consequence: Minimal, human team reviews/edits final brief.
• Audience impact: Zero. This brief could be created manually. AI just increases productivity.

Continuum model: No AI disclosure needed. In this case, I see AI as a smart template. You give it a format and data, and it plugs it in. 

Dig deeper: AI productivity gains, like vendors’ AI surcharges, are hard to find

Written content creation and transformation

Use of AI to brainstorm headlines or subject lines

Sample prompt: “I’ll upload the body copy; please provide 10 subject line options for this email message.”

• Context: Creative assistance. Human prompts, AI responds, human chooses or edits.
• Consequence: Minimal. The AI’s influence is limited to generating options. A human makes the final decision.
• Audience impact: Low. Users care more about whether the copy resonates than how it was written.

Continuum model: No AI disclosure needed. For me, this is like kicking copy ideas around with a colleague. Although headlines and subject lines are important elements of marketing copy, they are a small part of what goes into a campaign. 

Use of AI to organize a human brain dump into a draft

Sample prompt: “Here are my notes, please turn them into a rough draft.”

• Context: Human-generated input, AI-assisted structuring and phrasing.
• Consequence: Moderate. Depends on whether AI is simply formatting ideas or adding substantial new content.
• Audience impact: Variable. If the final product reflects your original thinking, disclosure isn’t expected. But if AI is adding material beyond your input, readers may assume more authorship than you actually contributed. 

Continuum model: AI disclosure may or may not be required. If AI is acting like a ghostwriter, shaping your thoughts into a clearer, more organized form, then disclosure would not be required under my model.  But if AI is inserting ideas, claims or other information you didn’t originate, you’re crossing into co-authorship territory and disclosure would make sense under my model. 

Use of AI to fully generate written content

Sample prompt: “Please write a 600-word post on marketing automation trends.” (And content is published with minimal edits under a person’s byline.)

• Context: Generative. AI creates the content from start to finish.
• Consequence: High. The output is largely or entirely machine-authored, not human-created.
• Audience impact: Significant. Readers assume the content reflects the author’s own expertise, voice or judgment.

Continuum model: AI disclosure is required (or better yet, don’t do this at all). This is where the academic in me kicks in.

If you’re passing off content that is not based on your own ideas and input as your own, that’s essentially plagiarism. It doesn’t matter whether the content was created by AI or another human. I like Georgetown’s AI disclosure policy, which requires that you disclose how AI was used, not just that it was used, for situations like this. 

Yes, disclose that you used AI and include the prompt language that you used. Or better yet, do a brain dump of your own ideas on the topic (see the example above) or summarize third-party content on this topic and provide attribution to the source (see the example below). 

Passing off fully AI-generated content as original work is why “AI slop” was created.

Use of AI to summarize or paraphrase third-party content

Sample prompt: “Please summarize the ideas in this MarTech article for our newsletter.”

• Context: Source material originates elsewhere, and AI is used purely for efficiency.
• Consequence: None, the AI isn’t generating original thought, just speeding up summarization.
• Audience impact: Zero. This summary could be created manually. AI just increases productivity.

Continuum model: AI disclosure is not necessary. In this case, AI is a productivity tool. Readers don’t care whether you summarized the article yourself or whether an intern or AI did the work. 

Caveat: Failing to attribute the original source when summarizing third-party content, whether manually or with AI, raises intellectual property and ethical concerns. This isn’t an AI disclosure issue. It’s about proper citation. Attribution is still required to avoid misrepresentation or plagiarism.

Dig deeper: Why AI content strategies need to focus on tasks not transactions

Visual content generation 

Use of AI to create a background image

Sample prompt: “Please create a background image we can use on our website.”

• Context: Supporting visual. AI replaces stock image or simple design task.
• Consequence: None, the visual doesn’t affect the message or meaning.
• Audience impact: None, there’s no expectation of human authorship.

Continuum model: AI disclosure is not necessary. In this case, AI is acting as a faster, less expensive option than stock images or a bespoke design. It’s a workflow win. 

Use of AI to create a visual metaphor or campaign concept image

Sample prompt: “Please create an illustration of ‘work burnout’ with flames to support this blog post.”

• Context: Image is conceptual or symbolic, not literal, but it plays a central role in message delivery.
• Consequence: Low to moderate, it depends on how literally the audience interprets the visual.
• Audience impact: Low, as long as the image is clearly an illustration or a metaphor.

Continuum model: AI disclosure is unlikely to be necessary. As long as viewers won’t assume the image is an actual photo, it functions more as an illustration than documentation and doesn’t need to be disclosed. 

Use of AI to generate images of people who appear to be real

Sample prompt: “Please generate a picture of a customer for this testimonial.”

• Context: Visual is presented as a real person (or implies realism).
• Consequence: High. Risks misleading the audience into thinking this is a real human or customer.
• Audience impact: High. Audiences may interpret this as an authentic representation, which affects trust.

Continuum model: AI disclosure is required (or better yet, don’t do this at all). Years ago, I worked for a brand that gathered testimonials and then had its designers match them with stock images, without AI. It was a bad idea then, and it’s just as bad an idea now, whether or not you use AI. This is an ethical issue, not an AI issue. 

Caveat: If the AI-generated image is a realistic likeness of a celebrity or public figure, then you’re in deep fake territory. This has the potential to bring with it lawsuits around rights, misrepresentation and potential defamation — whether or not AI is used. 

Dig deeper: How to protect customer trust when using AI

Use AI responsibly, disclose when it matters

I’m not anti-disclosure. I’m pro-useful disclosure. There are moments when AI use needs to be transparent, like when it fabricates a person, distorts a truth or presents machine-generated content as expert human insight. In those cases, the ethical (and sometimes legal) line is clear.

But blanket disclosure? Labeling every background image or brainstormed subject line as “AI-assisted”? That’s not transparency, that’s noise. It dilutes the moments where disclosure actually protects trust.

As marketers, we’ve been through this before. Remember the early days of sponsored content? Influencer ads? Cookie banners? When everything got labeled and eventually, nothing got read?

AI is just the latest tool in the stack. Like spellcheck, Photoshop, Grammarly and Google Translate. Its presence doesn’t always change what the audience sees or how they interpret it. When that’s the case, a disclaimer isn’t just unnecessary, it’s distracting.

Let’s stop treating AI like a secret or a scandal. Let’s treat it like what it is: a powerful creative partner. One that deserves disclosure when it changes the meaning, the message or the trust. And one that can stay behind the curtain when it doesn’t.

That’s not hiding anything. That’s respecting the audience and their attention.

The post A practical framework for AI disclosure in marketing appeared first on MarTech.