Microsoft Advertising has published dedicated guidance for adverts made or materially altered with artificial intelligence. The main point is straightforward: using an AI tool does not move responsibility away from the advertiser.
The guidance covers generated, manipulated and other synthetic content. It addresses disclosure, permissions, misleading representations, deepfakes and the preservation of information that helps identify how an asset was made.
For businesses in Bath and across the South West, this is less a reason to stop using AI and more a reason to tighten the approval process. AI can speed up creative work, but speed is useful only when somebody still checks the finished advert properly.
Disclosure is not as simple as labelling everything “AI”
Microsoft’s guidance deals with disclosure where it is required. That should not be read as a blanket rule that every advert touched by an AI tool must carry the same label. Requirements can depend on the content, the format and the places where a campaign runs.
Where disclosure is necessary, it needs to be clear and close to the relevant content. Microsoft recommends placing it within image or video assets and also points advertisers towards its disclaimer feature for supported formats.
A note hidden in campaign paperwork is unlikely to do the same job as a notice the audience can actually see or hear. Before launch, advertisers should therefore ask two separate questions: does this creative need a consumer-facing disclosure, and has that disclosure been placed where it will be understood?
Keep the provenance information
AI-generated files may contain metadata, watermarks or other signals recording something about their origin or editing history. Microsoft’s guidance says advertisers should not interfere with that provenance information.
It is worth preserving original exports and keeping a simple record of the tool used, the prompt or brief, later edits and the person who approved the result. If an agency, freelancer or platform feature produced the asset, retain the supplied source files rather than working only from a screenshot or compressed copy.
Provenance and disclosure are related but different. Machine-readable metadata may help document where an image came from, yet it may be invisible to a customer. Equally, a visible AI notice does not prove that every claim in the advert is accurate.
Permission and accuracy still matter
An AI tool can produce a plausible person, product, place or endorsement without having permission to do so. A disclosure does not make an unauthorised likeness acceptable, and it does not rescue a misleading price, performance claim or testimonial.
Microsoft identifies prohibited deepfakes, impersonation, unauthorised use of a person’s likeness or voice, missing required disclosures and interference with provenance signals as possible reasons for an advert to be rejected, restricted or removed.
Local advertisers should be particularly careful with generated “customer” images, voiceovers that resemble a known person, before-and-after visuals and synthetic demonstrations of a product or service. Ask whether the advert could give a reasonable viewer the wrong impression even if no single sentence is obviously false.
The same discipline should apply to location. A generated street scene that looks like Bath may suggest a real local presence, premises or project. If that implication is untrue, the image can mislead without saying anything at all.
A useful pre-launch check for local campaigns
Add a short AI creative review to the normal campaign sign-off:
- Origin: record which elements were generated or materially altered, and keep the original files.
- Rights: confirm permission for logos, music, photographs, voices, likenesses and other protected material.
- Claims: check prices, offers, product details, results and testimonials against a reliable source.
- Disclosure: decide whether one is required for the relevant content, format and campaign geography, then make it understandable.
- Landing page: ensure the page supports the promise made in the advert and does not quietly contradict it.
- Human approval: name the person who made the final decision rather than assuming the platform checked everything.
This review fits naturally within a broader search marketing process. Creative, targeting, tracking and landing pages should be checked together because a compliant image cannot compensate for a confusing offer or a broken enquiry journey.
Businesses experimenting with several AI tools may also benefit from a simple internal policy covering approved services, customer data, source-file retention and sign-off. Practical AI governance for local businesses need not be an enormous manual; a clear one-page checklist is a useful start.
Do not confuse availability with approval
The most useful lesson is that a platform feature being available does not mean every output is ready to publish. Generation tools can help with variations, resizing and early ideas, while the advertiser remains responsible for what reaches the public.
For Bath and South West businesses, the sensible response is calm: keep using helpful tools where they save time, but introduce a clear human checkpoint before an AI-assisted advert goes live. Preserve how the asset was made, check permissions and claims, and make any required disclosure meaningful to the person seeing the advert.
