Quick answer: what are the ethics rules for AI in advertising?
AI advertising ethics rests on three duties: disclose AI-made people, voices and scenes that look real, check that automated delivery is not unfairly skewed, and keep every claim and testimonial true. In 2026 these are no longer optional. The EU, India and New York have disclosure rules, and Meta and Google now label AI ads.
AI advertising ethics comes down to three duties
Ethics in AI advertising is not a philosophy debate for a brand. It is three practical duties: disclose, stay fair and stay true.
- Disclose: tell people when AI made or changed something that looks real, such as a person, a voice or an event.
- Stay fair: check that the algorithm is not hiding your ads from groups of people for no good reason.
- Stay true: keep claims, reviews and testimonials honest, whoever or whatever wrote them.
Until 2025 these were mostly good manners. In 2026 they became rules, with dates and enforcement. That shift is the reason this topic now belongs in a monthly review, next to tax and billing.
This article covers the rules and a routine to follow them. For the wider picture of how AI changes roles and budgets, see our guide to what AI changes for brands, agencies and platforms.
Disclosure rules in 2026: what the law and the platforms require
Regulators in the EU, India, New York and the US now have written rules that touch AI in ads, and so do both big ad platforms. They differ in detail, but they aim at the same thing: realistic content that could fool a viewer.
| Rule | Where | What it asks for | Status |
|---|---|---|---|
| EU AI Act, Article 50 | European Union | Whoever uses AI to create or change image, audio or video that is a deepfake must disclose that it is artificial | Applies from 2 August 2026 |
| Synthetic performer law | New York | Whoever produces or creates an ad must identify AI-generated synthetic performers in it | Signed December 2025, in effect since June 2026 |
| IT Rules amendment | India | Platforms must ensure clear labelling and traceable metadata for AI-generated content | Amended rules, reported by the government in July 2026 |
| Consumer Reviews and Testimonials Rule | United States | No fake or false reviews and testimonials, with civil penalties for knowing violations | In effect since 21 October 2024 |
The European Commission's guidelines on Article 50 also list exceptions, such as standard editing. Cropping a photo or fixing its colour is not the target. A generated person is.
What Meta does
Meta's help page on AI info labels says it automatically labels ads when you use its generative tools, such as Background Generation, Image Generation or Add Animation, to create or significantly edit an image or video. It also labels ads made with outside tools when it detects standard metadata such as C2PA.
In regions with their own AI rules, Meta lets advertisers self-disclose, and the label then shows next to the Sponsored label. Meta currently lists the European region, California, New York, India and Taiwan.
What Google does
Google's July 2026 policy update allows text or visual AI labels inside image and video ads. Google says these labels will not count as a breach of its rules on text overlays and watermarks. Google is also rolling out an AI label setting across Google Ads, Display & Video 360, Campaign Manager 360, Merchant Center and Ads Editor.
Google states that using its AI label setting does not guarantee compliance with any specific regulation, and tells advertisers to seek legal guidance. The duty stays with you, not with the platform.
When to label an AI ad: a 4-question test
You do not need to label every ad that AI touched. You need a fixed order of questions, asked for every creative before it goes live.
- Does it show a realistic person, voice or event that AI created or materially changed? If yes, label it in every market. This is the one case all the rules agree on.
- Is anyone in the ad presented as a customer, expert or user? If yes, that person and their words must be real. AI may not invent them.
- Will the ad run in the EU, India, New York, California or Taiwan? If yes, turn on the platform's AI disclosure for any generated or heavily edited asset, even a product scene.
- Is it an election or political ad? If yes, platform disclosure of synthetic content is mandatory. Google's update repeats this for all election advertisers.
If all four answers are no, as with a resized image or a cleaned-up background, you have no extra duty. The platform may still add its own label, and that is fine.
When you are unsure whether a viewer could take an AI scene for a real one, label it. A label costs little. An ad that fools people can cost the account, and the trust of the customers who saw it.
Keep one more record: a simple asset register. For each creative, note which tool made it, who approved it and whether a real person's face or voice was used with written consent. Our article on AI and advertising creativity covers how to keep AI work recognisably yours.
Bias: how to check whether AI delivery is skewed
Delivery bias means the algorithm shows your ads far more to some groups than others, in a way your real customer base does not justify. You find it by comparing where the money goes with who actually buys.
The platforms have had to deal with this themselves. Meta built its Variance Reduction System after an agreement with the US Department of Justice. It launched in 2023 for US housing ads and was later extended to US employment and credit ads. For those categories, Meta had already removed some targeting options in 2019.
Most brands do not sell housing or credit, but the same question still matters for commercial reasons. An algorithm trained on your past buyers keeps finding people like them. It can quietly stop testing everyone else.
Run a delivery skew check
Use two numbers for each age, gender and region group: its share of ad spend from the platform, and its share of buyers from your store data. The method for pulling the first is in our guide to placement, age and region breakdowns.
Worked example: a skincare brand spending ₹50 lakh a month
Step 1: Put spend share next to buyer share
Example: last month the brand spent ₹50,00,000. Men aged 25-44 received ₹6,00,000 of it, which is 12% of spend. Store data shows the same group placed 30% of all orders.
Skew index
Share of spend ÷ share of buyers = 12% ÷ 30% = 0.40Step 2: Read the index
An index of 1.0 means spend matches the customer base. At 0.40, this group gets less than half the spend its buying would suggest. Women aged 18-24 show the opposite: 28% of spend and 15% of buyers, an index of 1.87.
Step 3: Find the cause before changing targeting
In this example, every ad showed only women. The algorithm did what the creative told it to. The fix is two new ads that show men using the product, not a manual targeting change.
Our rule of thumb: look closer at any group with at least 10% of your buyers and an index below 0.5 or above 2.0. Check creative, landing page and product range first. Change targeting last.
Honesty: AI avatars, testimonials and claims
AI presenters are allowed. Fake customers are not. That one line settles most questions about avatars and AI-written reviews.
The US Federal Trade Commission's questions and answers on its reviews rule say the rule has no blanket ban on AI-generated avatars in marketing, and that it was drafted so as not to prohibit virtual influencers. The same rule bans fake or false reviews and testimonials, and courts can impose civil penalties for knowing violations.
In practice:
- An AI avatar can explain how a product works.
- An AI avatar cannot say "I used this for 6 weeks" as if it were a customer.
- A real customer's written review can be read out by an AI voice, if the words are unchanged and the ad is labelled where the rules require it.
- AI-written claims need the same proof as any other claim. "Clinically tested" needs a test, whoever typed it.
Generated copy brings one extra risk: it can state things about your product that nobody checked. Keep a human approval step for every claim. We list other common failures in AI advertising mistakes that waste budget.
What most guides get wrong about AI advertising ethics
Much of what ranks for this topic was written before the 2026 rules arrived, or treats ethics as a list of values. Here are the claims to stop repeating.
- "There are no laws on AI in ads yet." Outdated. Article 50 of the EU AI Act has applied since 2 August 2026, and New York's law has been in effect since June 2026.
- "You must disclose every use of AI." No rule we read says that. The laws target deepfakes and synthetic performers, and the EU guidelines list standard editing as an exception.
- "The platform's label makes you compliant." Google says the opposite about its own setting. Labels help, but the legal duty is yours.
- "Bias is the platform's problem." The platform controls delivery, but you control the creative, the product feed and the conversion data it learns from.
- "AI labels ruin performance." We found no reliable public data either way. Treat it as something to test in your own account, not a reason to hide AI use.
For the benefits, risks and regulation in one place, see our overview of AI advertising benefits, risks and regulation in 2026.
A monthly ethics routine for brands spending ₹10 lakh to ₹1 crore
At this level of spend, a team launches dozens of creatives a month across several markets. Ethics only holds if it is a routine with an owner, not a value on a slide.
- Name one owner. Usually the marketing head. Agencies and freelancers follow the same checklist.
- Run the 4-question test on every new creative before launch, and record the answer in the asset register.
- Get written consent before you use a real person's face or voice with AI, including staff and creators.
- Check the skew index once a month for your main age, gender and region groups.
- Keep a proof file for every claim and every testimonial in live ads.
- Review market rules each quarter, because the list of regions with AI disclosure laws is growing.
Sellers in India and the Gulf should apply the same care to cash on delivery (COD) offers and festive claims around Diwali and Ramadan. A generated "70% off" banner still needs a real 70% price cut behind it.
The checks themselves are light. ScaleWin, for example, already breaks results down by age, gender and region every day, so the spend side of the skew index is ready when you need it. The decisions stay with a person.
Read next in this series
Key takeaways
- Label any ad where AI created or changed a realistic person, voice or event. That is the case every law and platform rule targets.
- EU AI Act Article 50 has applied since 2 August 2026, and New York's synthetic performer law has been in effect since June 2026.
- A platform label is not legal cover. Google says its AI label setting does not guarantee compliance with any regulation.
- Check delivery skew monthly with a simple index: share of ad spend divided by share of real buyers, for each age, gender and region group.
- AI avatars are allowed in marketing, but an avatar may not pose as a real customer. Fake testimonials can bring civil penalties in the US.
Ai and advertising: frequently asked questions
Do I have to label every ad made with AI?
Is it legal to use AI-generated people in ads?
Can I use an AI avatar for a customer testimonial?
What does EU AI Act Article 50 mean for advertisers?
Who is responsible if an AI tool makes a biased or misleading ad?
How do I check my ads for delivery bias?
Sources
- European Commission: Guidelines on transparency obligations under Article 50 of the AI Act
- EU AI Act Service Desk: Article 50
- Governor of New York: Law requiring disclosure of AI-generated synthetic performers in advertisements is in effect
- Press Information Bureau, Government of India: IT Rules amendment on synthetically generated information
- US Federal Trade Commission: Consumer Reviews and Testimonials Rule, questions and answers
- Meta Business Help Centre: About AI info on ads created or edited with generative AI tools
- Google Advertising Policies Help: Updates to AI labeling requirements (July 2026)
- Meta: An update on our ads fairness efforts (Variance Reduction System)
Let ScaleWin read your ad data every day
ScaleWin checks your Meta and Google Ads daily, judges every ad on profit and tells you exactly what to change.
Start free, no card


