AI dalam Iklan Kini Perlu Konteks yang Lebih Jelas: Audit 15 Menit untuk UMKM

An AI label is not cosmetic. On August 18, 2026, the IAB released its V2 framework for AI disclosure in advertising, only two weeks after Article 50 transparency obligations under the EU AI Act became applicable. For an Indonesian small business, this is not a reason to copy European rules blindly. It is a practical reason to make sure customers are not confused when an ad uses an AI-made image, voice, avatar, or chatbot.

Many businesses use AI to speed up concepts, remove a photo background, create copy variations, and answer initial questions. That can be useful. But material that looks like a record of reality or a human conversation raises a simple question: does the audience know what it is looking at? The answer should arrive before a customer has to ask.

🧾 Why this is a live issue this week

On August 18, 2026, the IAB released AI Transparency and Disclosure Framework V2. Two weeks earlier, Article 50 transparency obligations under the EU AI Act became applicable on August 2. That sequence has returned AI labelling to the operating agenda for advertisers, agencies, and publishers. A small business does not need to wait for an identical local rule before taking the practical lesson. When a brand uses an image, voice, avatar, or conversation that a person could mistake for human-made, the audience needs honest context. A marketing team that records where an asset came from can answer a customer without guessing.

For a small business, start with an inventory of active campaigns: product photos, short videos, voice-overs, service avatars, chatbots, and promotional material made with generative tools. A business owner can assign one person to review material before it goes live, then retain the brief, asset source, talent permission, and final version in the campaign folder. That process does not make an ad stiff. It gives staff shared language when a customer asks what they are seeing.

The boundary matters. A label does not replace truthful product claims, image-use permission, or data protection. A store still has to substantiate price, availability, benefits, and promotional terms. AI transparency helps a reader understand how a piece was made; it does not license promises that cannot be supported.

📌 What has actually changed

The European Commission says Article 50 covers machine-readable marking by providers and disclosure duties for deepfakes and certain AI-generated text publications without human review or editorial responsibility. Its code of practice is voluntary, while the transparency obligations are legal requirements. A small business does not need to wait for an identical local rule before taking the practical lesson. When a brand uses an image, voice, avatar, or conversation that a person could mistake for human-made, the audience needs honest context. A marketing team that records where an asset came from can answer a customer without guessing.

Do not reduce this to a rule that every use of AI needs a sticker. First map who created the material, who publishes it, who sees it, and whether it could mislead people about identity or authenticity. A business owner can assign one person to review material before it goes live, then retain the brief, asset source, talent permission, and final version in the campaign folder. That process does not make an ad stiff. It gives staff shared language when a customer asks what they are seeing.

The boundary matters. A label does not replace truthful product claims, image-use permission, or data protection. A store still has to substantiate price, availability, benefits, and promotional terms. AI transparency helps a reader understand how a piece was made; it does not license promises that cannot be supported.

🎨 A practical line for creative work

The IAB recommends targeted disclosure when AI materially affects authenticity, identity, or representation. Its high-risk examples include prompt-generated images and video, some synthetic voices, avatars, and digital twins that could cause confusion. A small business does not need to wait for an identical local rule before taking the practical lesson. When a brand uses an image, voice, avatar, or conversation that a person could mistake for human-made, the audience needs honest context. A marketing team that records where an asset came from can answer a customer without guessing.

When a visual looks like a real customer, model, place, or product, place an easy-to-read explanation beside it. A team can use plain wording such as 'AI-generated illustrative visual' when that wording is accurate. A business owner can assign one person to review material before it goes live, then retain the brief, asset source, talent permission, and final version in the campaign folder. That process does not make an ad stiff. It gives staff shared language when a customer asks what they are seeing.

The boundary matters. A label does not replace truthful product claims, image-use permission, or data protection. A store still has to substantiate price, availability, benefits, and promotional terms. AI transparency helps a reader understand how a piece was made; it does not license promises that cannot be supported.

💬 Chatbots and voices need context

European Commission guidance says providers of directly interactive AI systems must make people aware when they interact with AI. The IAB also names chatbots or assistants that could be mistaken for humans as an example that merits disclosure. A small business does not need to wait for an identical local rule before taking the practical lesson. When a brand uses an image, voice, avatar, or conversation that a person could mistake for human-made, the audience needs honest context. A marketing team that records where an asset came from can answer a customer without guessing.

Write an opening that identifies the digital assistant, offer a route to a human, and review automated answers about price, stock, refunds, and high-risk advice. The people behind the business remain responsible for the experience a customer receives. A business owner can assign one person to review material before it goes live, then retain the brief, asset source, talent permission, and final version in the campaign folder. That process does not make an ad stiff. It gives staff shared language when a customer asks what they are seeing.

The boundary matters. A label does not replace truthful product claims, image-use permission, or data protection. A store still has to substantiate price, availability, benefits, and promotional terms. AI transparency helps a reader understand how a piece was made; it does not license promises that cannot be supported.

📊 Trust is not cosmetic

IAB research with Sonata Insights, published in January 2026, found that more than half of respondents wanted brands to disclose fully AI-generated advertising or AI imagery and video. The IAB's August 18 release also said 73% of Gen Z and Millennials said clear disclosure would increase or not affect their likelihood to buy. A small business does not need to wait for an identical local rule before taking the practical lesson. When a brand uses an image, voice, avatar, or conversation that a person could mistake for human-made, the audience needs honest context. A marketing team that records where an asset came from can answer a customer without guessing.

Use that finding as a reason to test clarity, not as a sales statistic. Ask someone who did not build the campaign to read the label and explain what they think it means. A business owner can assign one person to review material before it goes live, then retain the brief, asset source, talent permission, and final version in the campaign folder. That process does not make an ad stiff. It gives staff shared language when a customer asks what they are seeing.

The boundary matters. A label does not replace truthful product claims, image-use permission, or data protection. A store still has to substantiate price, availability, benefits, and promotional terms. AI transparency helps a reader understand how a piece was made; it does not license promises that cannot be supported.

🔎 What to do this week

This week's developments are not an order to stop using generative tools. They are a reminder that creative technology works alongside context, permission, and editorial responsibility. The IAB itself distinguishes targeted disclosure from universal labelling. A small business does not need to wait for an identical local rule before taking the practical lesson. When a brand uses an image, voice, avatar, or conversation that a person could mistake for human-made, the audience needs honest context. A marketing team that records where an asset came from can answer a customer without guessing.

Choose one active campaign, run a fifteen-minute audit, and repair the one thing most likely to confuse a customer. A business website with a full offer, FAQ, policies, and human contact remains the strongest place to hold context that an ad cannot contain. A business owner can assign one person to review material before it goes live, then retain the brief, asset source, talent permission, and final version in the campaign folder. That process does not make an ad stiff. It gives staff shared language when a customer asks what they are seeing.

The boundary matters. A label does not replace truthful product claims, image-use permission, or data protection. A store still has to substantiate price, availability, benefits, and promotional terms. AI transparency helps a reader understand how a piece was made; it does not license promises that cannot be supported.

🧠 Separate work assistance from public representation

Staff can use AI to tidy a draft, find headline variations, or extend a plain background behind a product that they actually photographed. That back-office work does not automatically change what a customer experiences. The risk changes when the output becomes public material that claims, or appears, to record a real person, place, event, or conversation.

If a team invents a customer who appears to give a testimonial, a reader may believe that person exists and bought the product. A clear label does not cure a fake testimonial. The business needs real, accountable reviews. The useful question is not the name of the application used. It is what the material represents to the person who sees it. The same test applies to animation, voice-over, and avatars.

📱 Test material on the screen customers use

A label that looks good in a design file can fail when an ad appears on a phone. Text may sit behind a button, flash too quickly in a video, or disappear in a vertical crop. Open the material at real screen size before approval. Ask two staff members to view it without verbal context, then ask what they understand about the origin of the visual or voice.

The test does not need to become a large research project. Record the date, material version, and changes made. If the label is not visible, move it or change the format. If wording sounds like legal jargon, replace it with everyday language. The destination page matters too because a small ad cannot carry every piece of context. A website can explain the offer, how the service works, return policy, and a human contact route.

🤝 Who makes the final decision

AI can produce alternatives, but a person in the business needs to approve material that represents the brand. Set out who can give final approval. In a small business, that may be the owner or marketing manager. The important part is that the person is known and the decision does not disappear into a group chat.

That person should check whether the material states facts accurately, whether rights to supporting material are clear, whether customers can understand relevant AI use, and whether a customer can reach a human. A short record helps when material is reused. The habit also protects designers and chat administrators from claim decisions that should sit with the business owner.

A pre-publication check can stay short:

• Confirm that price, availability, and benefit claims have evidence.
• Confirm that people, voices, and supporting material have clear permission.
• Confirm that customers can understand relevant AI context and reach a human.

🧭 Keep the claim, the tool, and the channel separate

A useful approval conversation has three separate parts. First, what is the business claiming? A discount, product availability, customer result, or delivery promise needs evidence regardless of how the content was produced. Second, what did the tool create or alter? That record helps a team decide whether an audience needs context. Third, where will people see it? A website, ad, email, social post, and chat interface each give different room for explanation.

Keeping those parts separate prevents a common mistake. A team may spend time perfecting an AI disclosure while overlooking an unsupported claim. Or it may use a truthful offer but present an invented person in a way that creates confusion. A short review sequence can catch both problems: verify the commercial claim, verify the source and permission for creative elements, then decide whether the audience needs an AI explanation.

The record can remain simple. Save a screenshot of the published item, the final copy, the version of the visual, the person who approved it, and the link to the landing page. If a platform changes its crop or format, the team can inspect the actual published version rather than rely on an old source file. If a customer asks a question, staff can give a factual answer quickly.

🛠️ Start with one repeatable campaign template

The first implementation does not need a new software purchase. Add a small section to the campaign brief: AI used? Yes or no. What was generated or altered? Does it portray a real person, real event, real place, or staff member? What wording appears for the audience? Who checked it? Where is the final asset stored? The answers create a useful trail without turning a small team into a compliance department.

Use the template on a campaign that already has a clear owner. After publication, collect ordinary customer feedback and support questions. If people ask whether a person, image, or voice is real, the team has learned where context was insufficient. Adjust the next version. This is more productive than treating disclosure as a one-time badge.

For a business with a website, put stable explanations in places people can find later: a campaign landing page, FAQ, contact page, or policy page. An ad can signal the essential point. The website can carry the details, evidence, and human contact that let a customer make an informed decision.

🌐 Where the ad, website, and support team meet

A promotional item rarely stands alone. A person may see a video on social media, open a search ad, visit a landing page, and then send a question through chat. Each point can create a different impression about what is real, who is speaking, and what is being promised. That is why a team should inspect the customer journey from start to finish instead of reviewing only one creative file.

Start with the link used by an ad. Confirm that the destination page names the business, gives a checkable product description, includes relevant policy information, and provides working contact details. If a video uses a synthetic voice or an illustrative visual, the page does not need to repeat the label excessively. It must not contradict the context that the ad provided. Consistency helps a customer understand the offer without guessing.

Customer-service staff also need to see the material in circulation. Give them the final copy, likely questions, and answers that do not make new claims. When a customer asks whether an image was AI-made or whether they are speaking to a bot, an honest answer works better than a defensive response. When the bot cannot handle a question, route the conversation to an accountable person.

This cross-channel habit helps even a business that is not advertising in Europe. Platforms change, formats change, and customers carry expectations from one application to another. The website gives a business the most stable place to explain its offer and identity. An AI audit is therefore also a wider digital-communication audit.

🧩 Two clearer decisions in practice

Imagine a coffee shop creating a poster with an illustrative barista character. The poster does not say that the character is a real employee. The wording 'AI-generated illustration' can help when the image resembles a photograph. The case changes if the shop uses a voice that imitates the owner without permission. The issue is not only the production technique. It is identity and permission.

A chatbot provides another example. A bot that opens with 'I am this shop's virtual assistant' gives a customer context at the start. A bot using a human name, realistic profile photograph, and no route to staff can create confusion. Small design choices determine whether a customer feels guided or deceived after learning how the system works.

A team does not have to solve every possibility at once. Take the example closest to an active campaign, document the decision, and improve the next version. Small repeated practice gives a business owner more control than a long policy that nobody opens.

Sources: IAB, European Commission, and OpenAI. All sources accessed August 24, 2026.