Explain how a synthetic illustration was made and what it does not document

Record what was actually made and reviewed

List the media’s purpose, ingredients and production steps: original source material, generative step, human edits, voice, data labels and final review. Do not claim that a real customer appeared or an interview occurred when the material is an illustration. If a process detail is unknown, preserve that gap instead of adding a precise-sounding tool history.

Partnership on AI’s framework focuses on audiovisual synthetic media and distinguishes creators from distributors. Its creator guidance recommends explaining synthetic elements especially when knowing about synthesis changes perception. This is an identified voluntary framework, not a universal legal rule or a claim that every small text correction requires the same label.

Prepare a complete reader-facing explanation

Fictional proposed disclosure: “Illustrative scene generated with AI; it does not show a real customer or observed event. The accompanying numbers are constructed teaching data. Human review checked the labels against that example; no reader or business outcome was measured.” This is example copy for a fictional process, not a description of an actually generated photo in this packet.

Make the explanation specific enough to resolve the likely mistaken inference. A vague “AI” stamp may not tell readers whether the image, voice, numbers or the entire story is synthetic. Avoid claiming a human review that did not occur, or using a generic disclosure to disguise an unsupported factual result.

Keep visible explanation and embedded history distinct

PAI distinguishes direct disclosure, such as a visible context note, from indirect methods such as metadata or cryptographic provenance. Choose methods appropriate to the actual audience and surface. A reader should not need an unavailable specialist viewer just to learn that an apparent event is a fictional illustration.

Where credentials are used, retain their actual scope and privacy implications. Do not invent a signed manifest or assume that metadata survives cropping, re-encoding or reposting. The media-provenance guide handles interpreting an actual credential; the creation record and visible explanation remain separate artifacts.

Resolve three constructed disclosure decisions

A1 — Constructed generated scene accompanies invented reporting numbers. Decision: identify both the illustrative scene and the constructed dataset, with only the checks actually performed. Outcome: do not let “AI image” imply that the adjacent customer count, interview or improvement was measured.

A2 — Constructed synthetic narration is proposed as a real person’s testimonial without an established source/permission basis. Decision: hold that representation and resolve actual rights, consent and claim support rather than rely on a small AI label. Outcome: disclosure does not create a genuine interview or permission from the depicted speaker.

A3 — Constructed export carries disclosure only in a caption that is later detached. Decision: inspect the delivered asset/context and retain a clear reasonably connected explanation in the relevant copy. Outcome: no claim that a vanished caption, stripped metadata or a tiny unreadable note informed the recipient.

Review the actual public representation

Open the exact export and read it at the intended size with its caption, transcript or alternative. Can the reader tell what is synthetic, what factual source supports any real claim and what remains an example? A label in an editing project is not evidence that the final downloaded or reposted copy contains it.

Review the actual destination’s current relevant rules separately when a real publication is planned. This guide does not assert a universal LinkedIn label control or substitute PAI guidance for law or platform policy. A content note also does not settle reuse rights, privacy or whether an event claim is accurate.

Save the process and actual reader-facing copy

The worksheet records purpose, real/constructed status, ingredients, actual tools/steps, human checks, rights/consent boundaries, visible disclosure, embedded record if any and exact export/destination observation. Keep private source details minimal. Never fill an unperformed review or a missing outcome with a convenient story.

PAI’s actual creator/distributor/scope guidance and relevant C2PA concepts were read on 1 November 2026. All A-cases are constructed. The packet’s figures are standard plots of invented numbers, not generated photos or voices. No synthetic person, recording, actual publication or reader/business result was created or claimed.