Should You Tell Customers When You Use AI in Your Marketing?

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Should You Tell Customers When You Use AI in Your Marketing?
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Should You Tell Customers When You Use AI in Your Marketing?

Tell them when AI changes what they'd believe about you: realistic AI images or video of people, products or work that don't exist as shown, AI avatars, and chatbots that could pass for a person. Never publish AI-written testimonials at all. Copy that AI helped draft and you checked needs no label. Some platforms, and EU rules, make certain labels compulsory.

A simple test covers most cases: if a customer found out afterwards, would they feel misled? Nobody feels misled that an email subject line was drafted with AI. Plenty would feel misled if the "happy client" in an ad never existed, or if the tattoo in a studio's portfolio was never tattooed. That's why the answer matters most for businesses whose product is their craft, and least for businesses using AI to write routine copy. For the wider question beyond marketing, the general guide to telling customers you use AI covers service delivery, advice and operations.

Follow me on Instagram@sagnikteaches

Three questions that decide whether to label

  1. Could it be mistaken for something real? A photorealistic person, place, product or piece of work, or a voice that sounds human. Stylised illustrations, obvious cartoons and text don't raise this.
  2. Does it show what the customer will get? Portfolio images, results photos, product shots and "our venue" pictures are evidence. Evidence that was generated is a problem even if it's pretty.
  3. Is the customer interacting with it? A chat, DM reply or voice call. People are entitled to know whether a person or a machine is answering, and in some places the law requires you to tell them.

Two or three yeses: label it, or better, replace it with the real thing. One yes: judge by context, and lean towards a light label. None: no label is needed, though the content must still be accurate.

Connect on LinkedInSagnik Bhattacharya

Label or not: common marketing assets

AssetLabel?Why
Caption, email or ad copy AI drafted and you editedNoYou're responsible for it; nobody is misled about what it is
Website guide or FAQ drafted with AI and fact-checkedNoAccuracy matters, origin doesn't; Google judges quality, not authorship
Real product or work photo with an AI-generated backgroundUsually noThe thing on sale is real; label if the setting implies something untrue
Routine retouching of a real photoNoCustomers expect colour fixes and tidying
AI-generated realistic image of your product or workYes, or don't use itIt stands in for evidence of what customers get
AI-generated realistic person in an adYesViewers assume a real customer or model
AI avatar or cloned voice as a spokespersonYesIt speaks for your business as if it were a person
Chatbot or AI replies in DMsYes, at the start of the conversationPeople are talking to it directly
Concept visuals and mock-ups ("how your day could look")Yes, lightlyEasy to mistake for a past job
AI-written or AI-"polished" testimonials and reviewsNever publishAttributing words to real or invented customers is misleading, and unlawful in many places

The testimonial row isn't a labelling question at all. A label doesn't make an invented review acceptable. Where the legal line sits on AI and testimonials explains why and what you can do instead.

Subscribe on YouTube@codingliquids

What the platforms already require

Several of the places you advertise have their own rules, separate from any law. As of September 2026:

  • Meta (Facebook and Instagram ads). Meta applies an "AI info" label to ads fully created or materially edited with its own generative AI creative features, such as background or image generation; minor edits like resizing or colour correction don't trigger it. The label usually sits in the three-dot menu, but moves next to "Sponsored" if the image includes an AI-generated photorealistic person. Ads about social issues, elections or politics must disclose digitally created or altered media, including from third-party tools. Meta has said it's rolling this out gradually, so check the current help page.
  • TikTok. Requires creators to label realistic AI-generated content, using its AI-generated content setting or a clear caption, and may remove realistic AI content that isn't labelled.
  • YouTube. Requires creators to disclose realistic altered or synthetic content (something a viewer could mistake for a real person, place or event) through the AI disclosure setting in YouTube Studio when uploading. Clearly unrealistic or animated content, special effects and AI used for production help, such as scripts or captions, don't need disclosure.
  • Google Merchant Center (shopping listings). AI-generated product images must carry the IPTC DigitalSourceType "TrainedAlgorithmicMedia" metadata, and AI-generated titles and descriptions must be supplied through the structured title and description attributes marked as AI-generated.

The practical upshot: if you post realistic AI video or images on TikTok or YouTube, turn the label on yourself; it's quicker than dealing with a takedown. Labelling AI images and video on social media goes through each platform's settings step by step.

If you sell to customers in the EU

Article 50 of the EU AI Act adds legal transparency duties that have applied since 2 August 2026. Three matter for marketing:

  • Chatbots must make clear to people that they're interacting with AI, unless it's obvious.
  • Deepfakes, meaning realistic AI-generated or manipulated images, audio or video of people, objects, places or events, must be disclosed. The Commission's 2026 guidelines on Article 50 treat a realistic AI-generated image of a product in an advert as a deepfake that needs disclosure, while a real product shown against an AI-generated background generally doesn't, provided it isn't misleading. They also say that where content is both commercial and creative, the commercial character generally prevails, so the lighter "artistic work" treatment rarely applies to ads.
  • AI-generated text published to inform the public on matters of public interest must be labelled unless a person has reviewed it and someone holds editorial responsibility. Ordinary product and service copy is rarely about matters of public interest, and human review usually covers it anyway.

These duties apply according to where your audience is, not where you are based. How they apply to a particular campaign is a legal question, so if you run realistic AI imagery in ads that reach EU customers, ask a qualified adviser.

A tattoo studio sorts its last 20 marketing assets

An illustrative three-artist tattoo studio had been using AI for about six months and wasn't sure what it had published. The owner listed the last 20 marketing assets and ran the three questions over each one. The result:

DecisionCountExamples
No label needed11AI-drafted captions, the flash-day email, a gift voucher graphic with an AI-generated texture, the FAQ page
Add a label4AI concept sketches used in a "which should we draw next?" poll; an AI-generated studio interior used on a "coming soon" post for the new room
Replace with real work3A realistic AI mock-up of a fine-line sleeve on a model's arm; two AI "healed tattoo" images on the website
Remove2A "client review" graphic where the wording had been "tidied" by AI; an AI-generated portrait of a "happy client"

The replacements took an afternoon: the artists photographed healed pieces that clients came back to show, with permission. The labels took five minutes each. The two removals were the ones that mattered most, because they were the assets most likely to make a customer feel deceived. After that, the studio made the three questions part of posting: any new asset that gets a yes is labelled or replaced before it goes up.

Where hiding AI backfires, business by business

A photography studio's sample images

A family photography studio used AI to swap dull grey skies for golden-hour light in its sample gallery. It looked wonderful. Then a client, booked on the strength of those images, got photos with the grey sky that was actually there that evening, and asked why hers didn't look like the website. Minor retouching is expected; a changed sky in a sample image is a promise about results. The fix: label samples with replaced skies as "sky replaced", or offer the same edit to clients as a paid option.

A wedding planner's concept visuals

A wedding planner uses AI to create styling visuals for couples: "your barn venue with a dried-flower theme". Those are genuinely useful in a proposal. The problem arises when the same images appear on Instagram without context and followers assume they're past weddings. A short caption, "Concept visual for a couple's spring wedding, made with AI", keeps the usefulness and removes the ambiguity.

A yoga studio's website people

A yoga studio replaced its dated website photos with AI-generated images of calm, diverse members in a bright studio. New members arrived to a room that looked different and didn't recognise anyone from the pictures. Nothing illegal, but it set the wrong first impression. Real photos of real classes, taken with members' consent, would have done the job better.

A personal trainer's results images

Before-and-after images are evidence of results, so AI-generated or AI-altered versions are the clearest case for "don't use at all", label or not. Advertising rules in many places treat misleading results claims seriously, and a trainer's reputation rests on them.

Wording that doesn't sound like a confession

A label works best when it's short, specific and says what's real. Compare the generic disclaimer many businesses paste in, "This content may have been generated using artificial intelligence", with "Concept image made with AI. The final design is drawn by hand." The second tells the customer what they'll actually get. Templates for common cases:

  • Concept image: "Concept visual made with AI, to show the idea. Your final [design / styling / result] is made by us."
  • AI background on a real product: "Real product, AI-generated backdrop." (Only needed where the setting could mislead.)
  • Chatbot opener: "Hi, I'm the studio's AI assistant. I can answer questions about prices, bookings and aftercare. For anything else, I'll pass you to one of the artists." What to tell customers at the start of a chat has more versions.
  • AI avatar in a video: "This presenter is an AI avatar. The advice comes from our team."
  • Website note (optional): "We use AI tools to help draft some of our content. Everything is checked by our team, and all the work shown in our portfolio is our own." Writing an AI disclosure statement for your website covers the longer form.

When a customer asks: "Did you use AI for this?"

Answer plainly, and say what's human. Some honest replies for different situations:

  • Drafted copy: "Yes, I use AI to help with first drafts of posts and emails, then I rewrite them myself. The advice and the prices are mine."
  • Concept image: "That one's an AI concept to show the idea. Everything in the portfolio section is our real work."
  • Edited photo: "We retouched it, as with all our images, but the setting and the people are real."
  • Something you shouldn't have posted: "You're right, that image was AI-generated and we should have said so. We've taken it down and replaced it with real work."

The last one is worth rehearsing. A quick, straightforward correction usually ends the matter; a defensive one turns a small mistake into the story.

When someone else makes the content for you

Disclosure doesn't stop at what you make yourself. If a freelancer, agency or content creator produces marketing for your business, customers will hold you responsible for it, whatever tools they used. A designer who hands over a "lifestyle photo" for your website may not mention that it was generated, and an influencer's sponsored video might use an AI voice. You find out when a customer does.

The fix is a few lines in the brief or agreement. Something like: "Tell us which assets use AI-generated or AI-altered imagery, video or voice. Don't use AI to create images of people, results or our work without our written approval. Supply original files with metadata intact." For a wedding planner commissioning styled-shoot content, or a photography studio hiring a retoucher, that one paragraph prevents the most awkward version of this problem: explaining to a client that you didn't know. It also means you can label accurately, because you know what you're publishing.

Keep the proof: metadata and a simple log

Many AI image tools embed provenance data in the file, often as Content Credentials, and platforms such as Meta can read these signals to apply labels automatically. That's useful evidence of what was generated, but it's fragile: re-saving, compressing or passing an image through some editing apps and website tools can strip it. Google Merchant Center explicitly asks you not to remove the AI metadata from product images. So keep original files, and check that your website or shop doesn't strip metadata from images it's required to carry.

Alongside that, a simple log helps more than any policy: a spreadsheet with the date, the asset, where it was posted, what AI did (drafted text, generated image, edited background) and whether it was labelled. It takes seconds per post, answers "did we use AI for that?" in seconds when someone asks, and makes the next audit a ten-minute job instead of an afternoon. And if the question is really whether AI-written pages hurt your search rankings, that's a separate issue; whether Google penalises AI-written content covers it.

Disclosure questions that come up

Does labelling AI content reduce reach?

The platforms say not in itself. YouTube says disclosing altered or synthetic content won't limit a video's audience or its eligibility to earn money, and TikTok says turning on its AI-generated label doesn't affect distribution if the content follows its guidelines. What can hurt reach is content being removed or labelled by the platform after you failed to disclose.

Is AI retouching the same as AI-generated content?

Not usually. Removing a blemish, fixing colour or tidying a background are edits most customers expect and don't need a label. It becomes a disclosure question when the edit changes what the image shows as real: a different setting, a changed body, an added person, or a result the service can't deliver. Ask whether the edit changes what a customer would expect to get.

Do I need to mention AI in my privacy notice?

If AI tools process customers' personal data, such as a chatbot reading messages or an assistant analysing a customer list, your privacy notice should cover that processing like any other supplier. That's a data-protection question rather than a marketing-disclosure one, and the right wording depends on what you do, so check it with your data-protection adviser.

Further reads

Sources: Meta help pages (AI info labels on ads); TikTok and YouTube help pages on AI-generated content disclosure; Google Merchant Center Help (AI-generated content); EU AI Act Article 50 and the 2026 guidelines on its transparency duties, as summarised by law firms.

Want your AI marketing checked for disclosure?

On a 1:1 call we'll go through the ways you use AI in your marketing, sort what needs a label from what doesn't, and agree wording that fits your brand and the platforms you use.

Book a 1:1 call with me