The Trusted Creator

The Trusted Creator standard v1.0

This is the standard every creator is audited against. It is public so that anyone can see exactly what a certificate means.

Document
TTC-STD-001
Version
1.0
Published
28 October 2026
Last updated
28 October 2026

Scope

The standard covers video creators on YouTube, Instagram and TikTok. A certificate names a person, the channels covered, a tier, an audit period and the version of this standard.

We certify evidence of human production. We do not claim to detect AI in a finished video. Detection tools may be used as a second opinion, but a detector score alone never passes or fails a creator.

A certificate covers one named person. If a channel has other regular presenters, each needs their own audit. Guests are checked for synthetic voice and face only, and the creator must show the guest agreed to appear.

Tiers

There are two tiers. Both mean a real person made the work. The tier tells viewers how much AI was involved. An audit is measured against the rules of the tier the creator asked for, and the auditor must confirm those rules were met before a certificate can be issued.

Human Made

No generative AI in the script, research, voice, face, footage, music or thumbnails. Helper tools such as captions and noise removal are declared.

  1. The script and ideas were written by the creator. No generative AI drafted, outlined, rewrote or polished them, and no AI text is used word for word or nearly so.
  2. No generative AI was used for research or ideas.
  3. Every voice is a real recording of the creator, or of a named guest or collaborator. No synthetic voice, voice clone or dub.
  4. Every face and body is a real recording. No generated avatar, digital double or face swap. A human made avatar driven live by its verified performer is allowed.
  5. All footage was filmed, recorded, drawn or built by the creator, except declared and licensed stock. Nothing generated is shown as real.
  6. Stock and third party footage is under 25 percent of each video's runtime and is never presented as the creator's own.
  7. Music and sound effects are made by humans or licensed. None are generated.
  8. Thumbnails and graphics are made or edited by a person from real images.
  9. Besides these, only helper tools are used, and each is declared: captions, noise reduction, loudness, stabilisation, colour matching, classical (not AI) upscaling and denoise.
  10. Drawn, painted, modelled and keyframed pictures are made by a person. No AI inbetweening, auto colour, generated backgrounds or generative fill.
  11. Platform AI labels, where a video needs one, match the declaration.
  12. Every tool and every contributor is declared, and the declaration is true.

Human Led

A real person presents and makes the work. AI may help with research, music and thumbnails if declared and labelled. Never AI avatars, voice clones, dubs or face swaps.

  1. The script was written by the creator. AI may help with research and ideas if declared. No AI text is used word for word or nearly so.
  2. Every voice is a real recording. No synthetic voice, voice clone or dub, labelled or not.
  3. Every face and body is a real recording. No generated avatar, digital double or face swap. A human made avatar driven live by its verified performer is allowed.
  4. All footage was filmed, recorded, drawn or built by the creator, except declared and licensed stock. No generated video, and nothing generated is shown as real.
  5. Stock and third party footage is under 40 percent of each video's runtime and is never presented as the creator's own.
  6. Generated music and sound effects are allowed only if declared and labelled, and they do not imitate a real artist.
  7. AI may help with thumbnails and graphics if declared, and they depict nothing false.
  8. Helper tools are declared: captions, noise reduction, loudness, stabilisation, colour matching, and upscaling that does not invent faces.
  9. Drawn, painted, modelled and keyframed pictures are made by a person. No AI inbetweening, auto colour, generated backgrounds or generative fill.
  10. Platform AI labels, where a video needs one, match the declaration.
  11. Every tool and every contributor is declared, and the declaration is true.

Never allowed at either tier: AI avatars and digital doubles, including consented and licensed ones, voice clones of anyone, including the creator's own voice, synthetic dubs, face swaps, generated video presented as real, and any undisclosed synthetic media.

Helper tools that are allowed when declared
Noise reduction, loudness and EQ, captions and transcripts, subtitle translation, auto framing, scene cut detection, stabilisation, upscaling that does not invent faces, colour matching, background blur. These do not change the tier.
Stock and third party footage
Allowed if declared and licensed. It is measured for each sampled video, as a share of that video's runtime: up to 25 percent for Human Made and 40 percent for Human Led. Everything else must trace to an original file. It may never be presented as the creator's own capture.
Footage with no metadata
Phones and apps often strip metadata. The auditor looks for the original file on the device first, then asks the creator to film prompted clips on the same phone, then relies on the editing project and its media cache. A video that rests only on the last of these passes with a note and needs corroboration.
Live streams
The creator supplies the local recording and settings, not only the platform copy. The auditor watches one live stream in real time and checks that no virtual camera, face swap or voice changer is in the chain.

Evidence

Evidence is ranked by how hard it is to fake. Auditors look for at least two different kinds of evidence for each check, and documents alone never pass one.

LevelKindExamples
E1Signed captureC2PA Content Credentials from a camera or phone, validated.
E2Original filesCamera and phone originals, multitrack audio, intact metadata.
E3Project filesPremiere, DaVinci, Final Cut or CapCut projects with save history.
E4Watched liveA live call and screen share the auditor controls.
E5DocumentsScript history, receipts, signed declarations.

Definitions

Generative AI
Any model that produces or materially changes text, images, video, voice or music, beyond fixed signal processing. A tool that only measures or filters, such as loudness, is not generative AI.
Helper tool
A tool that does not produce the creative substance of the work: captions and transcripts, noise reduction, loudness and EQ, stabilisation, colour matching, auto framing, scene cut detection, subtitle translation, and upscaling that does not invent faces. Helper tools are declared and do not change the tier.
Research
Finding and reading sources, and generating ideas or outlines. At Human Made no generative AI may be used for it. At Human Led it may help, if declared, but never writes the script.
Declared
Written in the creator's signed declaration, naming the tool, its version and what it was used for, before the auditor decides.
Labelled
Stated where the audience can see it: in the description or on screen, and in the platform's altered or synthetic content setting where the platform has one.
Original file
The first generation file written by the camera, phone or recording app. A download, screen recording or re-upload is not an original file.
Stock or third party footage
Footage the creator did not film, used under a licence. It is declared with its source, counted as a share of each video's runtime, and never shown as the creator's own.
Avatar or digital double
A generated or captured likeness of a person that stands in for them, including a licensed or consented one of the creator themselves.
Voice clone and dub
A synthetic copy of a voice, or a synthetic translation of a voice. Neither tier allows them, whether the voice is the creator's own, a guest's or anyone else's, and whether or not it is labelled.
Format
What viewers mainly see and hear: on camera, voice and screen, drawn or animated, or an avatar performer. A channel can have more than one. The ten checks are the same for every format, and four of them (C1, C3, C5, C7) are worded for the format.
Source material
The first generation creative files for what viewers see or hear: camera originals, raw screen captures, layered or scene working files, an avatar model with a tracking session, and multitrack voice or music sessions.
Animation
Moving pictures whose frames are made by a person by drawing, painting, modelling, posing, keyframing, compositing or photographing physical objects. Hand drawn and modelled frames count as original. Frames made or interpolated by a model from a prompt or a learned style do not.
Analyse, enhance, generate
The test for any tool. Analyse reads your work and returns information, such as transcripts, tracking or a mask you correct, and is allowed. Enhance processes your own work without changing what it shows, such as denoise or upscaling. It is allowed at Human Led if declared, and at Human Made only in its classical, non AI form. Generate creates new pixels, frames, motion, voice, music or text, and is not allowed for anything the audience sees or hears except where Human Led allows it.
Performed avatar
A character made by a person and driven in real time by a verified performer's own movement and voice. A generated or autonomous persona, or a voice changed by a model, is not eligible at either tier.
Contributor
Anyone who makes part of what viewers see or hear, such as an editor, animator, writer or voice actor. Contributors are named in the declaration. A voice actor is checked for consent and a dated raw recording.
Presenter
The real person who appears or is heard as themselves in the work and makes it. A certificate covers one presenter.

Grey areas

Some uses fall between the rules. The auditor decides case by case, writes the ruling down, and the second reviewer must approve it before a certificate is issued. The same line applies to both tiers unless a tier rule says otherwise.

  • Generative fill or object removal on a single frame
  • AI de-ageing or heavy beauty filters
  • Voice enhancement that changes how a voice sounds
  • AI footage inside a clearly fictional or animated segment
  • AI assisted inbetweening or colouring of a person's own keyframes and line art
  • AI written product descriptions or sponsor text
  • A tool that mixes helper and generative features, such as an editor with text based editing

Sampling

  • The audit period is a fixed 90 days. It ends 14 days before we accept the application.
  • The auditor takes an inventory of every video published in the period at the start.
  • A random draw chooses the sample, and the seed is recorded. The creator never chooses and cannot swap a video.
  • The sample is between 5 and 12 videos depending on channel size. A channel with fewer than 5 videos is sampled in full, with a minimum of 3.
  • The auditor adds one extra video wherever something looks out of pattern.
  • The creator has 10 business days to supply the evidence for each request (15 for a re-test). Files up to 200 MB can be uploaded. Larger material, such as raw footage, is sent as a share link. The auditor downloads it and computes its SHA-256 themselves, and the creator keeps the original until the certificate ends.

The ten checks

Formats

The ten checks are the same for every creator. Four of them, C1, C3, C5 and C7, are worded for what viewers mainly see and hear, because the proof is different. A camera original, a screen recording and a layered working file all answer the same question: did a person make this?

On camera
I appear on camera. Talking head, vlog, live stream, music or performance, puppetry or models that I film. Source material: camera originals.
Voice and screen
You hear me and see my screen, gameplay, footage or images. Tutorials, gaming, podcast video, narration, reaction, commentary. Source material: raw screen capture and recordings.
Drawn or animated
I draw, paint or build what you see. Hand drawn, 2D, 3D, motion graphics, whiteboard, stop motion, storytime animation. Source material: layered or scene working files.
Avatar performer
I perform live as a character or avatar. VTubers and other performed avatars driven by my own movement and voice. Source material: avatar model and tracking session.

A channel can have more than one format. The auditor tags each video with its format, the sample includes at least two videos of each declared format where the channel has them, and the certificate lists every format covered. The tier follows the weakest sampled work. An undeclared format found in the inventory is a finding and adds a targeted sample.

Whatever the format, tools are judged the same way. Analyse is allowed, enhance is allowed at Human Led if declared, and generate is not allowed for anything the audience sees or hears. The definitions are above.

C1Identity

The named person is real and is the person in the videos.

Evidence
E4 watched live, E5 documents
How we check
  • Government ID checked by an auditor during the live session. We keep no copy.
  • Selfie matched to the ID.
  • The auditor compares your face and voice across at least three sampled videos.
Passes when
ID is valid, the selfie matches, and the videos show the same person.
Fails when
Altered documents, a mismatch, or an undisclosed person on camera.
Wording for other formats

Voice and screen

The named person who makes this work is real and is who they say.

  • Government ID checked by an auditor during the live session. We keep no copy.
  • Your voice is compared across at least three sampled videos and to the live session.
  • Any face that appears is compared to the ID.

Passes when: ID is valid and the voice, and any face, in the videos is the same person's.

Drawn or animated

The named person who makes this work is real and is who they say.

  • Government ID checked by an auditor during the live session. We keep no copy.
  • In the live session you produce work in the same style as the videos.
  • Any face or voice that appears is compared to the ID and the live session.

Passes when: ID is valid, the live work matches the style of the videos, and any voice or face is yours.

Avatar performer

The named performer is real and is who they say.

  • Government ID checked privately by an auditor. We keep no copy.
  • The performer is shown driving the avatar live in the session.
  • Your public persona may use a pseudonym. The record says persona, performer verified.

Passes when: ID is valid and the person on the live session is the one driving the avatar.

C2Channel control

You control every channel in the application.

Evidence
E4 watched live, E5 documents
How we check
  • We give you a unique code to place in your channel description.
  • You open your studio dashboard live on a call, to a page we choose.
  • We check the channel roles for managers or editors with publishing rights.
Passes when
The code is visible, studio access is shown live, and roles match your application.
Fails when
No access, a removed code, or a hidden manager.

C3Live presence

You can perform as the person on screen, in real time, unscripted.

Evidence
E4 watched live, E5 documents
How we check
  • At least six random prompts you have not seen: turn your head, cover your face, read a random phrase.
  • A check that no virtual camera, face swap or voice changer is running.
  • A short unscripted chat about your own videos, then a callback on a separate channel.
Passes when
Every prompt is completed with no visual or audio glitch, and you can talk about your work.
Fails when
Failed or dodged prompts, detected injection software, or a callback that is not confirmed.
Wording for other formats

Voice and screen

You can make this kind of work, in real time, without help you have not declared.

  • An unscripted voice segment on the call, on a topic we choose.
  • A screen share task in your own tools, such as performing one step of a tutorial or playing a section live.
  • A callback on a separate channel.

Passes when: The task is completed live, the voice matches, and you can explain your own work.

Drawn or animated

You can make this kind of work, in real time, without help you have not declared.

  • A live task of 10 to 15 minutes chosen by the auditor, such as redrawing or changing an element from a sampled shot, with your hands and screen on camera.
  • Opening a sampled working file and changing an element we name.
  • A short unscripted chat about your process, then a callback on a separate channel.

Passes when: The live task is completed in your own tools and style, and you can explain every choice.

Avatar performer

You can perform the avatar in real time, with no help you have not declared.

  • Your webcam and the avatar shown side by side while you follow random expression and head prompts.
  • A list of your running software shows no voice converter or video injection beyond the declared avatar tools.
  • A short unscripted chat in character, then a callback on a separate channel.

Passes when: The avatar follows every prompt live, and the device check shows only declared tools.

C4Script authorship

The ideas and words came from you, not a text generator.

Evidence
E5 documents, E4 watched live
How we check
  • Script drafts with native version history, such as Docs or Notion history.
  • We look for drafts built up over time, not one big paste.
  • We compare the final script to what is actually spoken in the video.
Passes when
Every scripted video has a believable history. Unscripted formats show raw takes instead.
Fails when
A script that appears in one paste, history created after publishing, or no way to discuss it.

C5Original footage

The visuals were filmed by you, not generated or passed off from elsewhere.

Evidence
E1 signed capture, E2 original files, E3 project files
How we check
  • Original camera or phone files for every shot that is not declared stock footage.
  • Metadata checks: device, timestamps, codec, file numbering.
  • Every shot in the published cut is matched to a raw clip.
Passes when
Every shot that is not declared stock traces to an original file. Small gaps are allowed when explained. Stock is declared, licensed and within the cap for the tier.
Fails when
Shots with no source, metadata that contradicts your story, or signs of generated video.
Wording for other formats

Voice and screen

The screen, gameplay, footage and images were recorded by you or declared and licensed, and nothing is generated.

  • Raw screen capture, gameplay or local recording for every shot that is not declared third party material.
  • A list of every third party clip, image and track, with its source and licence.
  • You contribute the script, the voice and the edit. Reaction and commentary add something of their own.

Passes when: Every shot traces to a raw recording or a declared licensed source, and your own contribution is clear.

Drawn or animated

The pictures were drawn, painted, modelled or built by you or declared collaborators, and none are generated.

  • Layered or scene working files for every sampled shot, with save history.
  • A storyboard or animatic, and three to five earlier frames from the same project, rough to final.
  • Every shot that is not declared stock or an asset traces to a working file. Small gaps are allowed when explained.

Passes when: Each sampled shot has working files and history that lead to the published frame, and the live re-export matches.

Avatar performer

The avatar and its scenes were made by people, and the performance was recorded by you.

  • The avatar model, with a commission contract or a record of who made it.
  • A tracking session recording and a local recording of the stream.
  • Overlays, scenes and plugins are listed and declared.

Passes when: The model has human provenance and the sampled shows trace to local recordings.

C6Human editing

A person assembled and finished the video.

Evidence
E3 project files, E4 watched live
How we check
  • Your project file, with its save history and timeline.
  • The timeline is compared cut for cut with the published video.
  • A short live task: open a project and make an edit we pick.
Passes when
Project history is coherent, the timeline matches, and you can operate your own tools.
Fails when
No project file, one created after publishing, or an undisclosed editor.

C7Real voice and face

Every voice and face is a real recording, not cloned, swapped or lip synced.

Evidence
E2 original files, E3 project files, E4 watched live
How we check
  • Your raw microphone track is compared to the published voice.
  • Breath, room noise and mouth sounds are checked for continuity.
  • A live voice sample read on the call.
  • No synthetic dub or voice clone at either tier, including of your own voice.
Passes when
The published voice traces to a raw take, and the live sample matches.
Fails when
An AI avatar, voice clone, face swap or lip sync model, or no raw audio at all.
Wording for other formats

Voice and screen

Every voice you hear is a real recording, never synthetic, cloned or converted.

  • Your raw microphone track is compared to the published voice.
  • No text to speech, voice clone or voice conversion at either tier.
  • A live voice sample read on the call.

Passes when: The published voice traces to a raw take and the live sample matches.

Drawn or animated

Every voice is a real recording, and every character is made by a person, never a generated likeness or voice.

  • Raw voice recordings, from you or from a named, consenting voice actor with dated session audio.
  • No text to speech, voice clone or voice conversion at either tier.
  • Characters are drawn or built, not generated, and are not likenesses of real people without consent.

Passes when: Each voice traces to a raw take and each character to working files.

Avatar performer

The avatar is driven live by the verified performer, in their own voice.

  • Your raw microphone track is compared to the published voice.
  • Only ordinary audio effects (pitch, EQ, reverb) are on the voice chain. No voice conversion model or text to speech.
  • The avatar art is made by a person. A generated or autonomous persona is not eligible at either tier.

Passes when: The voice is yours with only declared audio effects, and the avatar is human made and human driven.

C8Disclosure

All AI tools and helpers are declared, and the declaration is true.

Evidence
E5 documents, E3 project files, E2 original files
How we check
  • A signed declaration listing every tool used for scripts, voice, footage, music and editing.
  • We compare it with your project files and plugins.
  • We check your platform AI labels agree with the declaration.
Passes when
The declaration is complete and matches the evidence.
Fails when
Any undisclosed generative AI. This is a critical finding.

C9Consistency

The videos we sampled represent the whole 90 days.

Evidence
E2 original files, E3 project files, E4 watched live
How we check
  • We compare voice, style, pace and output across every upload, not just the sample.
  • Anything that stands out gets added as an extra sampled video.
Passes when
No unexplained shifts, and the extra video passes.
Fails when
A sudden change in voice, style or volume that the extra video confirms is not compliant.

C10Conduct

You have not faked evidence or impersonated anyone in connection with this mark.

Evidence
E5 documents, E4 watched live
How we check
  • A check of public platform actions tied to impersonation or fake AI claims.
  • We do not judge opinions, topics or quality.
Passes when
Nothing relevant found.
Fails when
Fabricated evidence in this audit, or a confirmed impersonation finding.

Findings and decisions

SeverityMeaningExamples
CriticalContradicts the claim or shows deception.Undisclosed synthetic voice or face, fabricated evidence, a false declaration.
MajorThe claim is not supported for a sampled video.Missing original footage, no project file, no script history for a scripted video.
MinorA gap that does not affect the claim.Missing optional metadata, an undeclared utility tool.
  • One critical finding, or major findings on two or more sampled videos, means a certificate is not issued. So does a result of not met on identity, channel control, live presence or real voice and face.
  • A certificate with notes can carry up to three minor findings. If the Human Made rules are not met but the Human Led rules are, the auditor can offer Human Led. The creator is told the reason, and accepts by signing the declaration again. The history of any change of tier is kept.
  • If the only gaps are major findings on one video, or minor findings, the creator has 15 business days to supply more evidence and is re-tested on a new random sample. This can happen once.
  • Any refusal is reviewed by a second auditor who did not work on the file.
  • The creator sees the findings first and has 7 days to reply before a decision is made.
  • Failed audits are never published.

Validity and removal

  • A certificate is valid for 12 months from issue and cites the audit period it is based on.
  • A spot check of two new videos happens around month 6.
  • The creator must report changes of channel owner, new AI tools that affect the tier, and platform strikes for deceptive content within 14 days.
  • If a Human Made creator starts using AI in a way only Human Led allows, the certificate can be moved down to Human Led. It is signed again and the change is shown on the public record. Moving up needs a new audit.
  • A certificate is removed for undisclosed AI use, a false declaration, an unreported change of owner, or a failed spot check.
  • Removals are listed in the registry with a short reason code, never with free text.
  • A creator can appeal a refusal, a suspension or a removal within 20 business days, in two stages. The second stage includes outside professionals. See the appeals procedure.

Limits of a certificate

A certificate says the evidence reviewed supports the claimed tier for a stated audit period, at a limited level of assurance. It does not say that no AI was used outside the sample, or after the period. We never describe a creator as guaranteed free of AI.

Change history

VersionDateChange
1.028 October 2026Initial release.

Changes to the standard are announced before they take effect, and every certificate names the version it was audited against.