Social Media Assistant
Journal

Labelling AI content in the UK, and stopping AI drafts inventing claims

Two questions come up whenever a small business starts using AI for social posts. Do I have to say it was made with AI? And how do I stop the tool making things up? Here are plain answers to both.

General information, not legal advice. Checked against the official pages on 4 November 2026.

Is there a UK law that says "label it"?

Not a general one. The ASA says the UK advertising codes have no AI-specific rules, and that there is no blanket legal duty in the UK to disclose AI use in ads. The existing rules apply however an ad is made.1 The main one is that an ad must not mislead.

The ASA suggests two questions instead of a fixed rule. Would people be misled without a label? And if so, does the label actually clear that up, or does it contradict what the ad shows?1

Its examples are useful. Tidying a photo with AI, such as removing something in the background, is much like ordinary editing. But an AI image that shows a skin or beauty result the product does not deliver is likely to mislead, and a small "made with AI" label will not fix it.1

What the platforms ask

Platform rules sit on top of the law, and they are stricter about realistic content.

  • TikTok asks creators to label AI-generated or significantly edited content that shows realistic scenes or people. Unlabelled content of that kind may be removed, restricted or labelled by TikTok.2
  • YouTube asks creators to disclose realistic content that is made or meaningfully changed with AI, such as a real person saying something they never said or a scene that never happened. It does not ask for disclosure of AI help with scripts, titles or captions, or of beauty filters and other minor edits.3
  • Meta (Facebook, Instagram and Threads) has said people must use its disclosure tool when they post photorealistic video or realistic-sounding audio that was digitally made or altered, and that it may apply penalties if they do not.4

Platforms change these pages often, so check them before you rely on them.

A simple house rule

  • Label it if it shows realistic people, places or events that did not happen, or a voice that is not a real person's. Use the platform's own label too.
  • No label needed, usually, for AI help with wording, or light edits that do not change what the photo shows.
  • Never use an AI person as a customer, an expert or a famous face. The ASA has said an AI-made celebrity endorsement that looks real enough to mislead is likely to break the Code.5 The same thinking applies to a made-up customer.

Labelling realistic AI people, voices and scenes every time is the safer default. It is a standing rule in our method notes, which cite Alexis Ellis's The Invisible Hand of AI among their sources for it.

Why AI drafts invent claims

Writing tools fill gaps with whatever sounds likely. Ask for a post about a bakery and you may get "family-run for generations", "award-winning" and a glowing quote from a customer called Sarah. None of it came from you. All of it is a claim you would have to prove.

Invented reviews are the most serious case. Under UK consumer law, businesses must not write fake reviews or get others to write them. The CMA describes a fake review as one that seems genuine but is not.6 A review an AI tool writes in a customer's voice fits that description. Testimonials in ads must come from real people, in their own words, with permission.7

How to stop it

  • Give the tool your facts. A short list of prices, services, hours, areas and proof you hold. Tell it to use only those facts.
  • Make it mark gaps. Ask it to write [CHECK] wherever it would need a fact it does not have, instead of guessing. Then you see the gaps at a glance.
  • Ban the usual inventions. No reviews, quotes, awards, years in business, numbers of customers or results unless they are in your list.
  • Read every claim before approval. You need evidence for a factual claim before it goes out, whoever wrote it.8 The ASA has said the advertiser keeps the main responsibility even when automated tools made the ad.9
  • Mind what you paste in. Customer names, messages and photos are often personal data. The ICO expects any sharing with an AI tool provider to be lawful, and AI output to be checked for accuracy.10

Social Media Assistant's planned design rules include an AI label where one is needed, consent for faces and voices, and a log of every approval. It is being built and tested on Keepwright's own social media first, and is not open to other businesses yet.

Sources

  1. ASA, Disclosure of AI in advertising: striking the balance between creativity and responsibility: asa.org.uk
  2. TikTok, Community Guidelines: Integrity and authenticity: tiktok.com
  3. YouTube Help, Disclosing use of altered or synthetic content: support.google.com/youtube
  4. Meta, Labeling AI-generated images on Facebook, Instagram and Threads: about.fb.com
  5. ASA, AI and deepfakes: four things advertisers need to know: asa.org.uk
  6. CMA, Short guide for businesses publishing consumer reviews and complying with consumer protection law (GOV.UK): gov.uk
  7. ASA, Testimonials and endorsements: asa.org.uk
  8. ASA, Substantiation: asa.org.uk
  9. ASA, Generative AI and advertising: decoding AI regulation: asa.org.uk
  10. ICO, Innovation advice: previously asked questions (generative AI): ico.org.uk
  11. Method: Alexis Ellis, The Invisible Hand of AI (book).