How agencies and brands can label AI-assisted content honestly

To label AI content honestly, decide on a small set of labels, apply them to every deliverable, and keep evidence of how each piece was made. Clients increasingly write AI rules into contracts, platforms require disclosure for some AI content, and the EU AI Act's Article 50 transparency rules apply from August 2, 2026. AI detectors cannot settle a dispute, because they guess.

The Trust Identity Protocol (TIP) by The AI Lab turns the label into evidence. Your agency or brand signs each deliverable under its verified organization TIP-ID with one of four origin labels, Original Human, AI-Assisted, AI-Generated or Mixed. The client, and anyone who reads the content later, can check for free who produced it, when and how. Agencies that use AI well can say so openly, and agencies that sell human writing can prove it.

Get a free TIP-ID How TIP works

The problem

Where AI disclosure goes wrong

Where	What goes wrong
Client contracts	Clients ask for human-written work or limit AI use, and the agency has no way to show compliance beyond its word.
Detector disputes	A client runs a detector, gets a false positive and disputes the invoice.
Hybrid workflows	Most content is now made by people and AI together, and a yes or no label does not describe it.
Regulation	Disclosure rules for AI-generated content are arriving in the EU and elsewhere.

Every option, honestly

What each option does, and where it stops

These work together. Use more than one, and read the last column for each, including ours.

Option	What it does	Where it stops
Contract clauses and AI policies	Set what AI use is allowed for each client.	Rely on trust, with no evidence attached to each deliverable.
AI detectors	Estimate whether text is AI-written.	Produce false positives and cannot prove anything.
Invisible watermarks (such as SynthID)	Mark output from specific AI models.	Cover only some models and can be removed. They say nothing about human work.
C2PA Content Credentials	Record editing history inside supporting files.	Often stripped, and rarely present on text.
Trust Identity Protocol (TIP) by The AI Lab	Signs each deliverable under your verified organization TIP-ID with one of four origin labels and a fingerprint, checkable by anyone for free.	A signed declaration, not a detector. It holds you accountable for the label you choose, and a false label can be disputed.

Step by step

How to put it in place, in order

Adopt the four TIP origin labels, Original Human, AI-Assisted, AI-Generated and Mixed, and define when each applies in your workflow.
Write the labels into client contracts and briefs.
Get an organization TIP-ID and sign each deliverable on TIP with its label before delivery.
Send the record link with each deliverable, and let clients publish it with the content.
For EU audiences, use the signed label as part of your Article 50 disclosure.

Get a free TIP-ID How TIP works

About TIP

What the Trust Identity Protocol is

The Trust Identity Protocol (TIP) is an open standard from The AI Lab for proving who made a piece of content and whether it has changed. A person or organization is verified once by an accredited Verification Provider and gets a TIP-ID. Each time they sign something, their own device signs it with a post-quantum signature (ML-DSA-65, NIST FIPS 204) after a fresh fingerprint or Face ID, and the TIP Network records the signature, the time, an origin label and a fingerprint of the content. Anyone can check a record for free, with no account.

TIP does not detect deepfakes, does not stop copying and does not decide whether content is true. It proves who stands behind a piece of content and whether a copy matches the original. It is governed in the open by the AI Trust Council, and the specification is public on GitHub. Signed content is called a Verifiable Asset.

Questions

Questions people ask

How do I prove to a client that content was written by a person?

Keep drafts and version history, and sign the deliverable on the Trust Identity Protocol (TIP) by The AI Lab with the Original Human label under your verified agency identity. The client can check the signed record for free. A detector score cannot prove it either way.

Is it bad to use AI for client content?

No. Many clients welcome AI-assisted work when it is disclosed and the quality is there. The problem is hidden use. TIP's AI-Assisted and Mixed labels let you show how you work.

Do I need to label AI content under the EU AI Act?

Article 50 requires disclosure for certain AI-generated or manipulated content, including deepfakes and AI-generated text published to inform the public, from August 2, 2026, with some exceptions such as text that a person reviewed and takes editorial responsibility for. Check the rules with your counsel. A signed TIP label is one practical way to disclose. More at theailab.org/eu-ai-act.

What do the four TIP labels mean?

Original Human means a person made it. AI-Assisted means a person made it with help from AI. AI-Generated means AI made it, with a person responsible for publishing it. Mixed means it combines human and AI-made parts.

What is the Trust Identity Protocol?

The Trust Identity Protocol (TIP) is an open standard from The AI Lab for proving who made a piece of content and whether it has changed. A person or organization is verified once by an accredited Verification Provider and gets a TIP-ID. Each time they sign something, their own device signs it with a post-quantum signature, and the TIP Network records who signed it, when, how it was made (Original Human, AI-Assisted, AI-Generated or Mixed) and a fingerprint of the content. Anyone can check the record for free.

What does TIP cost an organization?

Checking any TIP record is free for everyone, with no account. Signing is free for individuals, nonprofits, schools and universities, government bodies, journalism organizations for editorial use and businesses under USD 100,000 in annual revenue. Larger businesses use a published commercial license, with every tier listed at theailab.org/tip-license.

SourcesEU AI Act, Article 50, Google DeepMind, SynthID

Other use cases