AI-Generated Images Now Have Provenance Labels—But the Text Your Brand Publishes Still Doesn't

C2PA embeds a cryptographically signed manifest directly into a file: creator, tool, timestamp, edits, AI involvement. Adobe, Microsoft, BBC, Intel, and Truepic launched it in February 2021, and Adobe Firefly, DALL-E 3, Sora, and Google Imagen all embed C2PA credentials now. Google has watermarked over 100 billion files with SynthID since 2023.

But the system has a real failure mode. C2PA metadata lives in the file container, so a screenshot or re-encode silently strips it. C2PA metadata can disappear silently when files pass through platforms that strip or ignore the container. Midjourney still embeds no credentials at all. Absence of a label proves nothing; it only means provenance did not survive distribution.

Why Text Provenance Is Technically and Structurally Harder Than Image Provenance

Images carry redundant information across millions of pixels, giving watermarks room to hide. Text has none of that, and paraphrasing alone defeats most current schemes. ChatGPT embeds no watermark on any plan. Google's SynthID-Text ships inside Gemini but detects only Google's own model output; OpenAI built a text watermark internally and never released it.

Even a working watermark answers one narrow question: was this likely written by Model X? It cannot confirm human authorship or catch a different AI's output entirely.

What the Regulatory Environment Now Requires of Brand Publishers

The EU AI Act's Article 50 became fully enforceable August 2, 2026, requiring machine-readable markings on synthetic text, with fines reaching EUR 15 million or 3% of worldwide turnover. The technical reality is that no single technique currently satisfies Article 50 alone. In the US, regulatory pressure around disclosure of AI involvement in advertising has been building at both the federal and state level. Some US states have moved toward requiring disclosure when AI plays a significant role in producing consumer-facing content. Regulators require disclosure; the infrastructure to automate it for text does not yet exist at scale.

How Consumer Trust Is Shifting as AI Content Becomes Visible, and Why Disclosure Is Not a Clean Solution

A Klaviyo/Datalily survey found only 7% of consumers say visible AI-generated marketing content increases their trust in a brand, while 31% say it actively reduces it. Gartner research found 68% of consumers already suspect content is AI-generated regardless of any label. That raises an important question: if suspicion is the baseline, does disclosure actually help? 2026 academic research found reviews labeled "AI-assisted" landed better than those labeled "AI-generated." The framing of disclosure shapes reception in ways regulations have not caught up to.

What the Image-Text Credibility Split Looks Like to a Consumer Encountering a Brand Ad

Google's "How this ad was made" panel, rolled out across Search, YouTube, and Discover in July 2026, provides automatic disclosure for AI-generated images. The copy running alongside those images carries no equivalent signal. Long-form branded content sits most exposed here: thought leadership, founder perspectives, expertise articles. These are formats where human voice has historically been the primary trust mechanism.

What Strategic Content Ownership Actually Looks Like While Technical Infrastructure Catches Up

Without a machine-readable provenance layer for text, a brand's editorial process becomes the provenance. Who owns the content, and how that ownership shows through consistency and depth, are the signals available right now. That means four things matter most:

• Internal records of where AI tools enter the workflow, required for EU AI Act compliance regardless of watermark availability
• Consumer-facing language that meets emerging disclosure standards' proximity-to-content expectations
• Named human authority on key content: bylined experts, accountable editorial review, subject-matter depth generic generation cannot replicate
• Publishing decisions that are documented, not just made, platforms such as Letterstory, an end-to-end content automation platform, keep that workflow record as a byproduct of how they operate.