Instagram and Facebook AI labels: publisher checklist

This publisher checklist covers organic Instagram and Facebook posts, plus the separate review needed when a post becomes an ad. Official policy sources were checked on October 23, 2026. It is a production decision record, not a promise that every account has identical controls or that a disclosed post will receive a particular reach.

Three labeling questions to answer separately

Start with the unit you are reviewing. Is it one piece of media, a profile presented as a person, or a paid placement? Write that down before choosing a setting. A creator can have a real-person profile and publish a synthetic scene; a fictional-person account can publish an ordinary product photograph. Those are different decisions.

For organic content, Meta’s current election-safeguards page says people must use its disclosure and label tool for digitally created or altered photorealistic video or realistic-sounding audio. Meta also describes AI info labels based on detected industry signals or self-disclosure. Treat that as a requirement to review realistic media, not proof that automatic detection will complete your work.

Meta: current organic AI labeling and disclosure policy

Instagram’s August 31, 2026 update names a separate AI-generated profile label for an account featuring a generated person. Merely using AI in the creative process does not require that profile label. The official instructions describe editing the profile and enabling AI-generated profile; Account Status is the documented place to review recommendation restrictions or appeal a mistaken decision. This is an identity check, separate from the media review below.

Instagram for Creators: AI-generated profile label

For ads, Meta’s June 1, 2026 update describes About this ad in the three-dot menu, including AI info for detected third-party AI use. It notes regional differences. Social-issue, election and political campaigns have additional self-disclosure requirements in specified cases. An organic-post setting is therefore not a complete ad approval checklist, and an automatically detected label is not a substitute for a required advertiser disclosure.

Meta: current GenAI transparency for ads

1. Make a short inventory of the finished asset

Open the export you intend to upload, not a generation preview. Identify every material replacement or addition: a face, body, voice, setting, event, product behavior or spoken claim. Include audio that continues while the picture changes. If a realistic voice was created from text, record it even when the visual is a still image.

For each element, keep four notes: what the original showed, what the export shows, what a viewer might infer, and what context would prevent a mistaken inference. “Generated mountain background” is useful. “Made with AI” is less useful when the product itself is real but the demonstration around it is synthetic.

Check the caption and cover alongside the media. A clearly fictional illustration can become misleading when paired with “our actual customer yesterday.” Conversely, a truthful caption does not fix a fabricated testimonial spoken by a realistic avatar. Review the complete message as a new viewer would encounter it, including the first frame before they open the caption.

Classify ordinary adjustments separately from meaningful alterations. Color correction, a crop and caption placement do not answer the same question as changing someone’s words or inventing a product result. Record the actual change rather than assuming a software name establishes the disclosure decision. If the current rule is unclear for your format, consult the platform guidance before release.

2. Clear permissions and claims before choosing a label

Keep the source permission, likeness or voice permission, and final-use approval separate. A licensed photograph does not establish permission to make its subject deliver a sales script. Likewise, permission from a spokesperson does not verify a factual claim about a product. Assign one reviewer to check each category and retain the approved script or asset version.

Ask whether any customer story, endorsement or demonstration is authentic. Do not give a fictional speaker a fabricated personal experience. If the scene is illustrative, say what it illustrates. Preserve the evidence for real specifications and outcomes outside the synthetic scene: a product record, a real demonstration or a documented customer statement approved for this use.

For a sponsored collaboration or promotional post, review the commercial relationship separately. An AI production disclosure and a sponsorship disclosure answer different questions. Make the relationship understandable to viewers and check the destination’s applicable branded-content or advertising instructions. This guide does not certify a campaign’s legal compliance.

Use the commercial-use rights checklist for permissions, and the face-swap ad review guide when a person’s appearance is replaced in promotional media.

3. Preserve provenance without treating it as a truth certificate

Content Credentials use C2PA provenance to record signed information about an asset and its editing history. That information can be inspected and validated; it does not establish that a depicted event happened or that every claim is true. Missing credentials also do not prove a file was generated. Keep the source and edit record alongside the export.

Inspect the actual final file if provenance is part of your workflow. Do not assume that metadata found in an original survives every editor, download or repost. Record what the inspection found and which file was inspected. If the exported file lacks credentials, retain your internal production evidence and give viewers appropriate context; do not manufacture a provenance claim.

C2PA: current specification and provenance model

4. Review the current posting controls and save the decision

Use the current disclosure control available in the account and format you are publishing through. Read its wording before selecting an answer. Do not reuse a screenshot from another account as proof that your setting is saved. Confirm the decision again after replacing the media, moving from a post to a Reel, or handing the upload to another teammate.

The sources above verify the policy distinction and the profile setting; this article does not assert a universal tap-by-tap menu for every organic format. If the required disclosure control is missing or the wording is ambiguous, pause that release and consult current platform support. Adding a caption may help explain the scene, but it is not proof that a required platform control has been completed.

Save a compact approval note with these fields: destination; account; organic or paid use; export filename; picture changes; audio changes; viewer inference; permissions reviewer; claims reviewer; disclosure decision; caption version; final reviewer. After publication, add the URL and a screenshot of the viewer-facing context. Store the note with the exact file, not just the project name.

For a team, name the person responsible for the final upload. “The editor added a note” and “the social manager selected a setting” can otherwise become two incomplete handoffs. If the asset is reused next month with a new offer or audience, reopen the claim and permission checks rather than inheriting approval from the previous campaign.

Four illustrative publishing decisions

These are hypothetical editorial examples, not tests of Meta’s detection or approval systems. They show what to document and explain. Apply current platform instructions to the final asset; do not treat an example as a guarantee that a label, approval or distribution outcome will occur.

A real product in a generated setting

Suppose a bottle photographed on a desk is placed in a generated alpine scene. Keep the real packshot and record that the setting is illustrative. A useful caption could say “AI-generated setting; product shown from our supplied photograph.” Check that reflections, label text and scale still match the real bottle. Do not imply the scene proves insulation performance or a real expedition.

A consenting spokesperson with generated speech

Suppose a founder approves a realistic avatar reading an approved announcement. Record which words and audio are synthetic and retain the approval for that script. Add precise context such as “AI-generated presentation of an approved script.” Inspect the spoken words, subtitles and gesture together. Approval to use the founder’s likeness does not make an inaccurate price or invented promise acceptable.

An account built around a fictional person

Suppose an Instagram account presents a generated host as its recurring identity. Review the profile identity separately from each post’s media. Do not describe the host as a real employee, customer or independent reviewer. Keep a consistent explanation in the account context, inspect the current profile-label instruction above, and review every new script for invented experience or endorsements.

A stylized illustration reused in an ad

Suppose an illustrated campaign image later becomes a paid placement with a product offer. Recheck the offer, product details, usage rights and destination rather than assuming the original post review covers the ad. Read the actual ad preview and any transparency information available. The caption, landing page and targeting context can change what viewers infer from the same picture.

5. Check the result viewers actually see

Open the published destination and inspect the relevant post or ad, its caption, first frame, crop and audio. For an ad, look for the current transparency destination where available. Compare this with the approval note. A correct export alone does not prove that the upload kept the intended context or that a teammate selected the intended option.

If the posted context differs, record the discrepancy before changing anything: wrong file, missing caption line, misleading crop or a label requiring review. Use the platform’s supported editing, review or support path. Do not promise that removing metadata, changing a filename or reposting will remove a label. Resolve the underlying factual or permission issue first.

Keep measurement separate from policy review. A drop in reach after a disclosed post is not proof that its AI label caused the drop. Likewise, good engagement is not proof that a post was correctly disclosed. Track the actual notice, restriction or review result independently from views and conversions so that a performance question does not rewrite the facts.

A practical pre-publication checklist

Before release, confirm six things: the final media is the approved file; realistic picture and audio changes are inventoried; permissions and claims have named reviewers; organic, profile and paid-placement checks are separate; the required control and explanatory context have been reviewed; and someone owns the post-publication inspection. An unresolved claim or permission is a reason to pause, even if the cover looks finished.

After an edit, revisit the decision when the voice, person, scene, script, commercial relationship or placement changes. A spelling correction normally needs a narrower review than a replacement testimonial. Keep the decision proportionate, but do not copy an old approval to materially different content. The record should explain why this version is ready.

Publishing the same video on TikTok? Review the TikTok AI labels and disclosure checklist for that destination as a separate decision. This Instagram and Facebook checklist does not establish TikTok’s requirements.

Create the illustration, then review the disclosure

If the task is to revise an existing image, prepare one controlled edit with the subject and product facts fixed. Review the downloaded image against the source before choosing your publishing context. The image editor is a production step; Instagram or Facebook handles the platform’s disclosure control and any subsequent label review.

Sources and scope

The current Meta Newsroom and Instagram for Creators pages above support the policy distinctions. Exact organic posting menus were not authenticated or tested for this article. These retained background references provide additional context; the current 2026 primary updates take precedence for this checklist.

Meta transparency background reference

C2PA specification 2.3: retained earlier reference