AI Captions and Disclosure: What a Local Business Must Do

The short answer

There is no single rule that says "AI content must be labelled". There are several narrower rules, and they depend on what the AI produced and where your audience is.

What Meta asks for

Meta has two separate things, and people mix them up. One is a label Meta adds itself. The other is a disclosure Meta asks you to make.

• Labels Meta adds. In April 2024 Meta said it would label AI-generated content using industry signals and self-disclosure. In July 2024 it renamed "Made with AI" to "AI info" after real photos with small AI edits were being labelled. In September 2024 it moved the label into the post menu for content that was only edited with AI, and kept a visible label on content detected as generated (Meta newsroom, with updates).
• Disclosure you make. Meta's help page asks people to disclose when they share realistic video or realistic-sounding audio that was digitally created or altered. That requirement is described for video and audio, not for images, although images can still receive a label if Meta detects AI use (Meta help page). Read the wording on that page before relying on a summary, including this one, because Meta changes it.

In practice: if you use an AI voiceover on a Reel that sounds like a real person, disclose it. If you generate a lifelike video of your kitchen that does not exist, disclose it, and ask whether you should post it at all.

What the EU AI Act asks

If you sell to people in the EU, Article 50 of the AI Act applies to you as a "deployer" from 2 August 2026. Two parts matter for a social account.

• Deepfakes. AI-generated or manipulated image, audio or video that resembles real people, objects, places or events and would falsely appear authentic must be disclosed as artificially generated. A lifelike AI "photo" of your dining room is the kind of thing this describes.
• AI text on matters of public interest. Text published to inform the public on such matters must be disclosed, unless it has been through human review or editorial control and a person holds editorial responsibility. A promotional caption for a haircut is not normally a matter of public interest, and a person who approves every caption fits the exception anyway.

Read the article itself, not a blog summary: Article 50 on the European Commission's AI Act Service Desk. The Commission also publishes a FAQ on the transparency duties. A separate marking duty falls on the companies that build generative AI systems, not on you as a business posting a caption.

The bigger risk: invented praise and facts

Regulators care less about whether a caption was written by a person or a model and more about whether it is true. Three rules make that concrete.

• FTC Consumer Reviews and Testimonials Rule. In force since 21 October 2024. It bans writing, creating or buying fake or false reviews and testimonials, and that includes AI-generated ones. Reposting a review on your own social account turns it into a testimonial (FTC questions and answers).
• FTC influencer guidance. You cannot describe experiences with a product you have not tried (Disclosures 101). An AI that writes "our customers love the new menu" has no experience to describe.
• Your own statements bind you. In *Moffatt v Air Canada* (2024 BCCRT 149, 14 February 2024) a Canadian tribunal held an airline responsible for wrong information its website chatbot gave, and rejected the idea that the chatbot was a separate entity. It is a small-claims decision, not binding precedent, but it shows how a tribunal sees it (McCarthy Tétrault summary).

The US National Institute of Standards and Technology lists "confabulation" as a core risk of generative AI: confidently stated content that is wrong (NIST AI 600-1). A model asked for a caption about your opening hours will give you hours. They may not be yours.

Accuracy checklist for AI-drafted captions

Run this on every AI draft. It takes about a minute and catches the problems that actually cause trouble.

Six questions: label it, change it or post it

1. 1

   Is any image, video or voice generated rather than captured?

   If no, go to question 4. If yes, go to question 2.

2. 2

   Does it look like a real photo, recording or place?

   If yes, label it or replace it with a real photo. Meta asks for disclosure of realistic video and audio. The EU treats misleading realistic content as a deepfake. If it is an obvious illustration, label it anyway if there is any doubt.

3. 3

   Does it show something you do not actually sell, offer or have?

   If yes, do not post it. A realistic picture of a product you do not make is a misleading claim, label or not.

4. 4

   Does the text contain a review, quote, result or statistic?

   If yes, check that it is real, consented, dated and attributable. If you cannot, cut it.

5. 5

   Is the text promotional and about your business?

   If yes, a person approving it is enough for the EU text exception to be a non-issue. If it informs the public on a matter of public interest, read Article 50 or ask a lawyer.

6. 6

   Can you show who approved it and when?

   Keep the approved version and the approver's name. It will not prove compliance on its own, but it answers the first question anyone will ask.

Keep a record that names a person

The text exception in the EU Act and the FTC's interest in who stands behind a claim both point the same way: a real person should be responsible for what goes out. A record of who approved which version, and when, is the simple way to show it.

Keep it plain. A spreadsheet row per post with date, approver, caption version and the photo file name is enough. Tools that publish for you can store the same details automatically. For example, KeepsPosted drafts captions from your own photos and notes, you or a teammate approve them, and it keeps a publishing record of what went out and when. A record is evidence, not a guarantee that a post meets a rule.

What customers say about AI content

Two recent surveys suggest customers care. Treat both as one snapshot, from different samples, not as proof about your customers.

• YouGov and Meltwater, published 21 April 2026, covered nearly 10,000 consumers in seven markets: 86% said AI-generated content should be disclosed, 32% said they would trust a brand less if its content were AI-generated and 15% said they would trust it more (Meltwater press release).
• Clutch, published 8 September 2026, surveyed 601 consumers in August 2026 (country not stated): 53% said they are less likely to buy from brands they know use AI in social content (Clutch report).

The practical reading is modest. Using AI to speed up drafting is not what people object to. Using it to pretend, with fake photos, fake praise or invented facts, is.

Common questions

Sources

• Meta: approach to labeling AI-generated content (with dated updates) (checked 10 Nov 2026)
• Meta help: AI-generated content disclosure (checked 10 Nov 2026)
• European Commission AI Act Service Desk: Article 50 (checked 10 Nov 2026)
• European Commission: FAQ on Article 50 transparency obligations (checked 10 Nov 2026)
• FTC: Consumer Reviews and Testimonials Rule, questions and answers (checked 10 Nov 2026)
• FTC: Disclosures 101 for social media influencers (checked 10 Nov 2026)
• McCarthy Tétrault: Moffatt v Air Canada (checked 10 Nov 2026)
• NIST AI 600-1: Generative AI Profile (checked 10 Nov 2026)
• Meltwater and YouGov: consumer perception of generative AI (checked 10 Nov 2026)
• Clutch: AI in social media report (checked 10 Nov 2026)

Written by the KeepsPosted editorial team from the primary sources listed above, with AI assistance for drafting and a person reviewing every claim. This is general information, not legal advice. Rules change, so check the linked pages before you rely on them.