If you are using AI to create campaign images and blog drafts, you must now include "how to disclose that they were created by AI" in your pre-publication checklist. Based on the AI transparency guidelines released by the European Commission, we have outlined which content marketers and businesses should review first and who to divide the work with.
EU AI law guidelines have been released; what has changed?
On October 22, 2026, the European Commission released guidelines to support the implementation of transparency obligations under Article 50 of the AI Law. The guidelines explain the obligations of AI system providers and distributors and provide guidance, along with examples, on the display and detection of AI-generated content, as well as the scope of disclosure for deepfakes and certain texts intended for public interest information. These transparency obligations are effective from August 2, 2026.
The chronological order is noteworthy. Given that the guidelines were released after the obligations began to apply, it would be inappropriate to interpret them merely as a "preview of rules to be prepared in the future." It is more practical to view them as a resource that reviews existing AI-based operations and specifies what verification procedures are required for which deliverables. However, one cannot conclude that the exact same wording must be attached to all AI-using content simply because the guidelines have been released. The nature of the deliverables, the circumstances of their disclosure, and the role played by the business operator must all be considered together.
For example, the likelihood of reader misunderstanding differs between using AI to refine sentences in a product description and creating and posting a video that appears as if a real person is speaking. If you place the two results on the same line based solely on the record of "AI utilization," it is difficult to determine whether or not they are disclosed. The starting point is not the name of the tool used, but what the public result is and who encounters it.
There is a point many people overlook here: the approach of trying to limit AI transparency to a single-sentence notice at the bottom of the site. If readers encounter videos on social media, images in advertisements, and text on individual pages after navigating search results, the notice must also be reviewed based on each respective point of exposure. This week, try categorizing the AI-powered content currently being published into 'Text, Images, Audio, and Video,' and start by noting where each is displayed. Deciding on the display method comes next.

Should I display it on all content even if I use only a little AI?
The most frequently asked question in practice is, "Where does the line lie between AI-generated content and what is not?" Texts rewritten directly by a person receiving title candidates from an AI, advertisements created by compositing AI-generated background images, and informational videos featuring synthesized voices differ in both their production processes and methods of disclosure. The fact that the guidelines provide guidance on the indication and detection of AI-generated content, deepfakes, and the scope of disclosure for certain public interest information purposes through specific examples can be interpreted as a signal that simply checking whether a tool was used is insufficient.
When making a judgment, align the final product with the audience's perception rather than the production method. For instance, illustrations explaining how to use a product and synthetic videos that can be accepted as actual customer testimonies create different kinds of misunderstandings. The technical term "deepfake" is easy to understand if you think of results created by synthesizing or manipulating the images, audio, or video of real people or events to make them appear authentic. While a high level of video quality is not problematic in itself, you must consider whether the viewer can accept it as an actual record.
Open the list of contents and write the following questions next to each item.
- In this output, what parts were generated by AI—sentences, images, audio, or video? At what stage was human editing performed?
- Is there a possibility that readers might mistake this for actual statements by real people, real scenes, or public interest information written directly by a person?
- Where are the results published—not only on our own website but also in advertisements, external posts, or affiliate channels?
- Is it possible to track the relationship between the original and the final version so that the parts generated by the AI can be checked later?
These questions are not checkboxes that automatically arrive at legal conclusions. They are closer to a classification table designed to ensure that no content requiring review is omitted. In particular, it is safer not to interpret "some text intended for public interest information" as equating it with the entirety of ordinary product descriptions. Conversely, if you are writing content that conveys facts regarding public issues or social matters, do not exclude it from review simply because it is a blog post. You must verify the purpose of the post and the writing and review process, and then compare it against the examples in the guidelines.
Rather than applying the same label to all AI usage, it is more accurate to review potential misunderstandings by type, starting with publicly available content. It is advisable to establish a procedure to separate content where you are unsure whether labeling is necessary from the publication schedule and hand it over to a reviewer. This is to prevent the cycle of "posting first because it's urgent, and deciding later" from repeating.
AI tool vendors and content uploading companies: who should take charge of what?
The fact that the guidelines explain the obligations of both AI system providers and distributors is particularly useful when dividing practical responsibilities. Here, the provider should be understood as the party offering the AI system, while the distributor is the party using the system in actual work to produce deliverables. This means that if a marketing team uses an external AI tool to produce an advertising video, they should not simply pass the responsibility for verifying the tool and the public content onto one person. Which party has which obligations must be determined based on specific roles and usage methods.
Separating the check questions for the two areas makes it easier to communicate with partners.
| Inspection area | Questions to ask in the workplace | Materials to keep |
|---|---|---|
| AI system provision stage | What functions or information are provided regarding the display and detection of results? | Tool Guide Materials and Supplier Responses |
| Content creation and publishing stage | Who modified the generated product, and where is the final result published? | Production history and list of posting locations |
| Pre-release review stage | Is there no possibility that this falls under deepfakes or certain types of text intended for public interest purposes? | Review comments and approval records |
This table is not a legal determination table for establishing obligations, but rather a business memo categorizing matters to verify with different personnel. It is easier to obtain a response by asking specifically whether there are features or guidance available to help identify the generated results, rather than asking the tool vendor in a single sentence like, "Does it comply with EU AI law?" Conversely, even if you receive confirmation that the tool has such features, who will ultimately verify the descriptions or landing pages of the videos our company has posted is a separate issue.
In organizations that use agencies, the contract manager, content manager, and publication manager are often different. If the brief simply states "AI allowed," you will have to verify exactly which parts were composited upon receiving the deliverables. Create a form during the ordering stage that asks about the scope of AI usage, the presence of elements that make the person look real, the final release channels, and the production records to be provided. The questions can be short. What is needed when issues arise later is not a long statement of principles, but a record that explains how the deliverables were created and where they were published.
How are search traffic and AI content labeling connected?
The search team might initially worry that disclosure phrases could negatively impact clicks. However, there is no basis to claim that AI content labeling raises or lowers search rankings solely based on the public release of these guidelines. Disclosure obligations and search performance are separate issues. When deciding whether to display the label, conduct a review in accordance with laws and guidelines first, and separate performance measurement to verify how that decision is actually conveyed in posts.
Readers arriving via search typically view individual documents first, rather than homepage announcements. For product comparison articles, the body text serves as the starting point; for videos, the playback screen and post description; and for images, the context of the post where the image is used. There is a difference between something being "written somewhere on the site" and "being aware of it by the viewer." While the necessary disclosure method should be determined based on the content type, it is better to directly open and examine the screen the reader actually encounters during internal review.
Disclosure and content quality must also be managed separately. Disclosing the use of AI does not automatically correct outdated product information or verify claims without sources. Conversely, the possibility of reader misunderstanding does not disappear simply because the text was written by a human. For articles intended for search, compare figures, citations, and product descriptions with source materials regardless of whether a draft was created, and record the dates and scope verified by the actual person in charge in the internal editing log. The purpose of disclosures shown to readers and verification records maintained within the organization are different.
The tasks that can be applied immediately are as follows.
- Identify pages that use AI images, text, and voice among the major pages with search traffic, and check if readers can find relevant explanations on the individual screens they enter.
- If you have reused the same AI results across blogs, advertisements, and social media, please compare the posting styles for each channel. Notices from one channel are not automatically transmitted to others.
- Please record the inquiries and instances of reader confusion separately before and after page revisions. It is difficult to judge whether the display is sufficient based solely on click counts.
The principle remains the same, even when considering search screens that use AI to provide answers. Rather than assuming how a search service will display a sentence, edit the original page to ensure there is no confusion regarding what is actual experience and what is generated or synthesized. For instance, you should first identify and correct sentences that describe composite images as real-life examples, or those that exaggerate verified product specifications without citing. This is also a matter of trust that cannot be resolved by a single label.
What review order should I create starting from the next post?
When creating a new review process, attaching a lengthy approval step to every post makes it easy to bypass in practice. Instead, try incorporating questions into your existing publishing flow. This involves documenting whether AI was used during the drafting stage, checking for potentially confusing phrasing during the design and editing stages, and verifying the screen the reader sees just before publication. For small companies, one person can handle multiple stages, but it is better to keep a record of what was checked at each step.
- Write the content you plan to publish line by line and indicate the parts where AI will be used. Do not lump it together as 'entire post'; instead, divide it into sections such as title draft, body part, background image, and audio.
- Classify whether further verification is required regarding deepfakes or text intended for certain public interest purposes. For ambiguous items, do not finalize them based on the 담당자's (person in charge's) speculation; instead, mark them as subjects for review.
- Distinguish between what to verify with the provider and what to verify with the publishing organization. The key is not to mix questions about tool features with questions about how to present them to readers.
- Check how the deliverables and instructions appear to be related on the final publication screen. If it is a scheduled publication, check not only the screen at the time of scheduling but also the screen after actual release.
- We link and archive the original source, edited version, review comments, and posting location used. This is to avoid repeating the same questions from the beginning when modifying content or reposting it on other channels.
For instance, let's say you use a short video produced by an external production company for both your blog and advertisements. The marketer asks the production company if there are any composited characters, voices, or scenes, while the editor checks to ensure the caption isn't written in a way that misleads viewers into believing it is actual footage. The person in charge of publishing can verify what is visible in the blog post and the ad creative, respectively. Although the final decision regarding the required disclosure scope must be made by comparing the actual nature of the content with guideline examples, following this sequence speeds up the process of gathering the necessary data for that judgment.
Do not overlook content that has already been published. Given that the relevant obligation takes effect on August 2, 2026, reviewing only content to be produced after the guidelines are released on October 22, 2026, may not be sufficient. Among currently publicly available assets, prioritize identifying and reviewing AI-synthesized videos, generated images that resemble real photos, and text intended for public interest information. However, this does not mean that the same measures must be applied uniformly to all past posts. It is a suggestion to establish review priorities based on the public status and the nature of the output.
Frequently Asked Questions
If it is a site operating outside the EU, do I not need to check?
It is better not to jump to conclusions. Rather than determining applicability based solely on the location of the site, you should verify who the service is intended for and how the AI results are disclosed. If you operate a business related to the EU, organize your role and circumstances and obtain a legal review.
Do I only need to put the AI product label inside the image file?
Do not assume that looking at a single file is sufficient. The display, detection, and disclosure of generated content may involve different issues to review, and images may be downloaded or reposted on other channels. Along with verifying the file, you must also check what readers can understand from the actual published screen.
What should I do if I cannot tell if a partner has used AI?
Start by requesting verification regarding the production process. You can ask about the use of AI and the scope of final editing for each piece of text, image, audio, and video delivered. For unverified deliverables, do not arbitrarily record that "AI was not used"; instead, mark them as unverified and review them before publication.
Do I need to check this when reposting content that has been marked once to another channel?
It is safer to double-check the repost screen. Explanations that appeared alongside the results in the original post may be omitted in short advertisements or social media posts. Even if you use the same file, please review the publication method if the format in which the reader encounters it has changed.
Which department would be best to manage the AI usage records?
It is practical for the team responsible for publishing decisions to determine the location of the records and the update procedures. Roles can be divided so that the production team records usage and editing history, while the publishing team records public channels and final screenshots. The key is whether the history of a single deliverable can be traced back later, rather than the name of a specific department.
This week, check the post list first, rather than the label text.
The first step in putting the transparency obligations under Article 50 of the EU AI Act into practice is to transform the large folder titled 'AI Use' into a list organized by output. This week, please note the parts of the videos, images, and text currently in use that involve AI and the channels through which they were disclosed. You should separately mark content that could be mistaken for real people or scenes, or content that may fall under the category of public interest information, and compare them against examples from the guidelines.
Next, separate the questions to verify with the provider from those our organization needs to decide, and designate a person responsible for reviewing the published screen. Managing the deliverables that readers encounter alongside production records that can be explained internally is how you transform AI content labeling from a one-time task into an operational process. For items that are difficult to judge, rather than forcing the classification, note down what is unclear and leave them for review.
AI Summary
On October 22, 2026, the European Commission released guidelines to support the implementation of transparency obligations under Article 50 of the EU AI Law, effective August 2, 2026. Rather than simply checking whether AI is used in bulk, businesses must classify the types and locations of published results and separate matters to be verified with providers from matters to be reviewed directly.