The June Transparency Code — What's Coming and What You Can Do Right Now

Last Tuesday I used Midjourney to create a header image for our newsletter. Wednesday, ChatGPT drafted three LinkedIn posts. Thursday, an AI voiceover for a product demo video. By Friday I was staring at the AI Act's transparency requirements wondering: do I need to label all of this?

Short answer: probably yes.

The EU's Code of Practice on marking AI-generated content drops its final version in June 2026. That's four months from now, two months before the August 2 deadline when transparency requirements become enforceable. If you're using AI to create images, video, audio, or text that reaches customers, this affects you.

Here's what we know, what's still uncertain, and what you can start doing today.

What the Transparency Code Actually Is

The AI Act requires companies to mark synthetic content — anything generated by AI — in a way that's clear to people and readable by machines. Think watermarks, metadata tags, labels that say "AI-generated."

The Code of Practice is the technical specification. It'll define:

How to mark different content types (image, video, audio, text)
What machine-readable formats to use
Where labels need to appear
What counts as "sufficiently clear"

The first draft came out in December 2025. It gave us a preview, but the final technical specs aren't locked yet.

Who This Affects

You're in scope if you're generating synthetic content that reaches the public. Common scenarios:

Marketing teams using AI for social media graphics, blog illustrations, ad copy variations, email subject lines.

Content teams using AI to draft blog posts, generate video scripts, create voiceovers, produce podcast intros.

Product teams building features that generate images, text, or audio for users (think AI avatars, automated reports, chatbot responses shown to customers).

Anyone running ads with AI-generated visuals or copy.

This isn't about whether your internal process uses AI. It's about what you publish. If a customer sees it and AI made it, you're probably obligated to disclose that.

What We Know From the December Draft

The draft Code gave us some clues:

Machine-readable markers will be required, not just visible labels. Expect metadata standards, possibly watermarking for images and video.

Content types are treated differently. Images and video will likely need embedded markers. Audio might use audio watermarks. Text is trickier — the draft suggested metadata tagging and visible disclosures.

"AI-generated" vs. "AI-assisted" matters. Fully synthetic content (AI created it from scratch) needs clear marking. Heavily edited human content with AI assistance is murkier. The line isn't perfectly drawn yet.

User-facing clarity is non-negotiable. Whatever the technical format, people need to understand what they're looking at without a decoder ring.

What's Still Uncertain

Here's what we're waiting on:

Final technical specifications. Which watermarking standards? What metadata schemas? The June version will answer this.

Thresholds for "AI-assisted" content. If you use Grammarly to fix typos, do you mark it? Probably not. If you use ChatGPT to write 80% of a blog post and you edit the rest, do you mark it? Probably yes. But the exact line isn't defined.

Enforcement details. How will regulators check? What happens if you miss marking one image in a campaign of 50? We don't know yet.

Retroactive requirements. Do you need to go back and mark everything you've published since January 2025? Unclear.

The uncertainty is real. But waiting until June to think about this is a bad bet.

What You Can Do Right Now (Regardless of the Final Specs)

The technical details will change. But three things are safe bets no matter what the final Code says:

1. Inventory Your AI-Generated Content

You can't mark what you don't know about. Start a running list:

Where are you using AI to generate content? (Social media, blog, ads, product features, emails, videos)
What tools are you using? (Midjourney, DALL-E, ChatGPT, Jasper, Descript, Eleven Labs)
Who on your team is generating content with AI?
What content is fully AI-generated vs. AI-assisted?

This inventory is useful regardless of the final rules. You'll need it in June anyway.

2. Tag It Clearly Today

You don't need to wait for the official watermarking standard to start disclosing. Simple visible labels work now:

"Image created with AI"
"This script was generated with AI assistance"
"AI-generated voiceover"

Add these to your content publishing workflow. When someone exports an AI image from Midjourney, the next step is adding the label before it goes live. Make it a habit now.

For machine-readable tagging, you can start simple: add metadata fields to your content management system. Even a basic "ai_generated: true" tag in your CMS will make the June transition easier.

3. Prepare Your Technical Infrastructure

If you're publishing images, you'll almost certainly need to embed watermarks or metadata. Start researching:

Does your image editing workflow support EXIF metadata?
Can your video platform handle C2PA watermarks?
Does your CMS have fields for AI-generation flags?

You don't need to implement the final standard today. But you need the capability to add metadata and markers when the spec drops. If your publishing pipeline can't handle that, fix it now.

The Practical Timeline

Now through June: Inventory your AI content. Start visible labeling. Prep your technical infrastructure.

June 2026: Final Code published. Review the technical specs. Map them to your inventory.

June to August: Implement machine-readable markers. Update your workflows. Mark existing content if required.

August 2, 2026: Transparency requirements enforceable.

You've got four months until you know the final rules and four more to implement. That sounds like time, but content backlogs are large and workflows take time to change. Start the easy parts now.

AI-Generated Content Audit Template

Use this to get started:

Content Inventory

[ ] List all channels where you publish content (website, social media, email, ads, product)
[ ] For each channel, list AI tools in use
[ ] Tag each content type: fully AI-generated, AI-assisted, or human-only
[ ] Identify who on your team generates AI content

Current Labeling

[ ] Do you currently label AI-generated content? Where?
[ ] Is the labeling visible to users?
[ ] Is there any machine-readable tagging?

Technical Readiness

[ ] Can your CMS add metadata fields for AI-generation flags?
[ ] Can your image tools embed EXIF data or watermarks?
[ ] Can your video platform support watermarking?
[ ] Do you have a process to mark content before publishing?

Workflow Changes Needed

[ ] Who needs training on labeling requirements?
[ ] Where in your publishing process should labeling happen?
[ ] What templates or tools need updating?

Backlog Review

[ ] How much AI-generated content have you published since January 2025?
[ ] Can you feasibly go back and mark it if required?
[ ] What's your plan if retroactive marking is mandatory?

What to Do Today vs. What to Wait For

Do today:

Run the audit above
Start visible labeling on new AI content
Check if your tools support metadata/watermarking
Make AI disclosure a standard step in your content workflow

Wait for June to decide:

Which specific watermarking standard to use
Exact metadata schema formats
Whether to retroactively mark old content
How to handle edge cases (AI-assisted vs. fully generated)

The uncertainty is uncomfortable. But basic transparency practices — knowing what you've published, labeling it clearly, building the technical capability to add markers — those are useful regardless of the final Code.

Start now. Adjust in June. Be ready in August.

And if you need help tracking all of this, we built a tool for exactly this problem. Scan your AI-generated content, tag it, and prepare for the transparency requirements — regardless of how the final Code shakes out.

This document supports readiness preparation. It does not constitute legal advice.