The ASA ruled that a betting brand's AI-generated video of a footballer, posted on its own X account, was an ad, and treated the likeness as the player himself.
AI Skills for Marketing Teams: One Campaign Carried to Sign-Off, and the Rules It Has to Pass
By James Cotton · Last updated · Sources checked · 12 min read
Part of our topic guide on AI Skills for Business.
By James Cotton, Founder, iO-Sphere
The launch brief
The launch this guide follows is an illustration: a new product from a UK firm, with a landing page, a short product video, paid social ads placed partly by automated campaign tools and an agency, posts on the brand's own social accounts, a reviews panel and a launch offer. AI tools can make, or help make, each of them.
HubSpot's 2026 State of Marketing report says 80% of marketers use AI for content creation and 75% for media production; HubSpot, which sells marketing software, does not say where they work.
In the Agency Census 2025 of the Institute of Practitioners in Advertising (IPA), published on 11 February 2026, 88.3% of the member agencies that submitted returns said AI is having a considerable impact on how they work.
In November 2023 a Department for Education analysis placed marketing and public relations roles among the occupations most exposed to AI.
The Advertising Standards Authority (ASA) rules on ads under the CAP Code, which every ruling here cites; the Committee of Advertising Practice (CAP) advises on it. CAP's advice of 11 June 2026 says: "The Code is media-neutral, meaning the rules apply regardless of how the content was created."
The advertiser, it adds, "remains responsible for any output used in their ads". CAP's May 2025 advice gives its own example of AI work that changes nothing material: "subtle non-material changes", such as removing unwanted background elements or swapping a facial expression.
The experts on the landing page
The landing page could carry a panel of experts, each with a portrait, name, title and quote, all made by an AI tool.
The ASA's ShroomIQ ruling (Birling Shore Ltd t/a ShroomIQ, 22 April 2026) concerned a children's supplement homepage, seen on 26 October 2025, with a section headed "Meet the Experts": five profiles with portraits, names, titles and quotes.
A disclaimer at the bottom of the homepage said: "Some images, including product visuals, names and people shown, are AI-generated representations used for illustrative and branding purposes only." The ASA said this "suggested the individuals shown may not be genuine health professionals".
"Nevertheless," it went on, people "presented as health professionals alongside broader claims about the product's effects" created "the overall impression that those health professionals endorsed the claimed health benefits of the product".
The page breached rules 3.1 and 3.9, on misleading advertising and qualifications, and the health-claim rules 12.18 and 15.6.3. The ASA told the advertiser "not to include health professionals or individuals presented as health professionals to endorse the product".
The health rules applied because the product was a food supplement. What carries to any launch is that a label saying people were AI-generated did not cure the impression the page created.
The Advertising Association's voluntary Best Practice Guide for the Responsible Use of Generative AI in Advertising, published on 5 February 2026 with a working group that included the ASA, says deceptive content "should never be used, whether GenAI-use is disclosed or not", and flags AI testimonials and spokespeople that could pass as real.
A label can help where it corrects an impression. CAP's May 2025 advice says that making clear an influencer is AI-generated, or that a deepfake is only for comedic effect, "could well help to negate an otherwise misleading impression. It will always come down to the specific content and context, and the likely audience interpretation."
For this launch's own label, CAP's test is two questions: would the audience be misled if the use of AI were not disclosed, and if so, would the disclosure clarify the ad's message or contradict it?
A famous face in the launch post
The launch post on the brand's own account could use an AI likeness of someone the audience knows. In the Midnite ruling (Dribble Media Ltd t/a Midnite, 17 September 2025), the brand's X page carried an AI-generated video of the footballer Trent Alexander-Arnold giving a speech, labelled as an "ai-generated parody".
The ASA upheld the complaints under the gambling rules on strong appeal to under-18s. Two findings matter here: the post, in "non-paid-for space online under Midnite's control", was an ad; and the AI imagery was "a depiction of Mr Alexander-Arnold, which consumers would clearly recognise as being him".
CAP's advice of February 2025 forecast that the Code's Section 6 rules on real people were likely to apply to AI likenesses, and rule 6.1 urges written permission before "referring to a person with a public profile" or "implying any personal approval of the advertised product".
In June 2026 CAP added: "If an AI-generated celebrity 'endorsement' looks real enough to mislead consumers, it's likely breaching the Code." Personality rights and intellectual property, it noted, "aren't issues the ASA regulates".
The previous government's Report on Copyright and Artificial Intelligence, published in March 2026 by the Department for Science, Innovation and Technology, the Department for Culture, Media and Sport and the Intellectual Property Office, said existing protections did "not cover all situations where a digital replica is made without consent", and that a new personality right would be considered.
The product video and the launch offer
The product video shows the product at work, and the ASA reads what it shows as a claim. In the WiggyDog ruling (UAB CommerceCore t/a WiggyDog, 25 March 2026), a Meta ad's scenes "appeared to be of real and/or AI-generated puppies" and made a robot-dog toy look highly realistic and lifelike.
The advertiser "provided no evidence regarding the functionality and appearance of the toy", and the ad breached rules 3.1, 3.7 and 3.11. Whether the puppies were real, the ASA left open.
Rule 3.7 requires marketers to "hold documentary evidence" for objective claims before an ad goes out, rule 3.11 bars exaggerating "the capability or performance of a product", and rule 3.2 allows obvious exaggeration, or puffery, that does not materially mislead.
AI images answer to the same rules as any other image. In the SoulTalk ruling (Optimize Business Ltd t/a SoulTalk, 17 June 2026), a TikTok ad with five AI-generated images of female characters breached the social responsibility and harm and offence rules.
CAP asks advertisers to sense-check AI output as "Socially responsible", "Free from harmful, discriminatory stereotypes and misleadingness" and "Appropriate for the audience and placement".
The launch offer is copy an AI tool can produce at scale, and false urgency copy is now fined, though none of the first cases involved AI. On 14 October 2026 the Competition and Markets Authority (CMA), which enforces consumer law, announced its first fines for misleading sales tactics under its new powers, including £4.8 million for Wayfair, whose false time-limited offers ran with countdown timers.
A false claim that a product is available only for a limited time is, in the CMA's words, "considered unfair in all circumstances".
The reviews panel
The reviews panel answers to consumer law as well. Since 6 April 2025 the Digital Markets, Competition and Consumers Act 2024 (the DMCC Act), which the CMA enforces, has prohibited unfair commercial practices, including those its Schedule 20 lists as unfair in all circumstances; the CAP Code's review rules, 3.44 to 3.46, mirror it.
Paragraph 13 of Schedule 20 bans submitting or commissioning a fake consumer review, one that "purports to be, but is not, based on a person's genuine experience", and publishing reviews without reasonable and proportionate steps to prevent and remove fakes. The Act's definition of writing covers creating by any means.
The CMA's fake reviews guidance (CMA208, 4 April 2025) says a review "does not have to be written by a consumer to fall within the definition", treats buying or selling reviews "generated by software applications (such as bots)" that pass as consumers' as banned, and asks publishers to correct AI-generated review summaries that banned reviews have affected.
On 27 March 2026 the CMA opened five investigations into fake and misleading reviews, saying it had reached no conclusions, and told shoppers to "Be alert to AI generated reviews", since fake reviews "can now sound fluent, polished and highly convincing". Its case pages show the investigations open, with the next update due in "Winter 2026 to 2027".
The paid ads, placed by a tool and an agency
The paid social ads go out partly through automated campaign tools and an agency. In the Stripe & Stare ruling (8 March 2023), the advertiser said its agency ran the ad "through Google Performance Max who had selected this image for use".
The ASA held that advertisers "were primarily responsible for ensuring that their ads complied with the CAP Code", and the ad breached the responsible advertising and harm and offence rules.
In the AI Mirror ruling (Polyverse Inc t/a AI Mirror, 6 December 2023), the advertiser said an agency uploaded its ad and "their ad network's algorithm" placed it in a mobile game. The ASA upheld the complaint under the social responsibility and harm and offence rules, and CAP cited both rulings in its June 2026 advice.
An AI agent could run the whole campaign. The CMA's guidance on complying with consumer law when using AI agents (9 March 2026, updated 30 October 2026) has that example: design and train the agent to the DMCC Act, with accurate prices, labelled paid endorsements and genuine offers.
As in each of its examples, "Someone with appropriate experience" should review the campaign regularly, and if the agent is not making influencers label their endorsements, the firm should "quickly refine the relevant prompts and workflows".
The Digital Regulation Cooperation Forum's joint statement of 30 October 2026, from the CMA, the FCA, the Information Commission (the ICO) and Ofcom, says of personal AI agents acting for consumers that an agent's autonomy "does not diminish the responsibility of organisations that deploy it".
Between advertiser and agency, the principles that ISBA (the Incorporated Society of British Advertisers) and the IPA set out on 1 November 2023, for the creative process only, ask both to be transparent with each other about their use of AI and to carry out due diligence on the AI tools they work with.
The ASA's own AI monitoring system processed 60 million ads in 2025, says its Annual Report 2025. The tools and the agency choose under the brand's name, so who in the team holds the check on their choices is settled before any ad is booked.
The training behind the launch
Each asset above maps to something the team must know. The Advertising Association's guide wants training "proportionate to organisational size and employee roles": onboarding on its principles and on legal frameworks including the CAP and BCAP Codes, role-specific training, and "regular updates when regulations, guidance, or organisational policies change".
A smaller organisation can designate one person to follow regulatory changes and brief the team. The guide's measures of whether training worked are staff knowing when to escalate an AI concern, approval and review processes being followed, and fewer problematic AI outputs.
As its own recommendation, the guide ties approval to risk: content making claims "about health, safety, or product efficacy" needs "senior executive approval and multiple review stages", and "Authorised and disclosed use of real people's likenesses" needs "managerial approval and expert review". Its Principle 4 asks for human oversight before publication, with "records of review decisions".
The Chartered Institute of Marketing's Global Professional Marketing Framework lists "Artificial Intelligence" as a skillset under "Digital agility".
Two of Skills England's national standards for marketing roles name AI too. The current Level 3 Advertising and media executive standard (ST0644) includes a duty to follow organisational policies on digital tools such as "generative artificial intelligence".
The revised Level 3 Multi-channel marketer standard (ST1031), available for information only during a lead-in period until 30 November 2026, when starts can begin, names technologies "including artificial intelligence (AI), and approaches to using them responsibly".
For this launch, that means claims and offers for the copywriter, people and likenesses for the panel and the post, the review ban for the reviews, approval levels for the sign-off, and who owns what an AI tool makes for anyone publishing a tool's images.
Who on the team learns first
ISBA's annual Gen AI Survey, published on 22 July 2026 and based on 200 responses from advertisers, found 65% of individuals using generative AI regularly and 28% saying it has meaningfully impacted their day-to-day work.
The previous wave, run in July 2025 and published that September, found 46% of advertisers had not yet started AI skills training, the series' last published figure on training.
Of the 2,350 readers who answered Marketing Week's 2026 Career & Salary Survey, reported on 29 May 2026, 66.5% had identified an AI skills gap within their team over the past 12 months.
Our view is that everyone who touches a launch like this should learn the rules for the assets they make. One person with the experience the CMA describes should own the sign-off, at the level the riskiest asset calls for, and keep its record; one or two should go deeper, into measuring campaigns and leading the team's change in how it works with AI.
Our data and AI fluency training would teach the people who make and approve a launch like this the rules each asset has to pass, in whichever AI tools the team already uses.
Who pays what depends on the employer, and the IPA's 2025 census estimated that over 85% of the levy funds its submitting agencies paid, money for apprentice training, went unused. The person who measures this team's campaigns could take Advanced Data & AI (Level 4), on the Data analyst standard (ST0118), which lists "Marketing Data Analyst" among its job titles, and whoever leads that change, AI Transformation (Level 4).
The checks at sign-off
In the PixVideo ruling (Saeta Tech Ltd t/a PixVideo, 18 March 2026), on an ad for an AI video app, the advertiser told the ASA it was adding "stricter creative guidelines, enhanced internal review, and mandatory compliance checks" to its review and approval processes. The person signing off this launch can run these:
- The video and the copy: every objective claim has evidence on file before publication.
- The landing page: anyone shown as an expert, customer or endorser is real and holds that role.
- The brand's own posts: any real person's face or voice has written permission behind it.
- The reviews panel: every review is a real customer's, incentives are shown, and any AI summary uses only those.
- The launch offer: the deadline and the countdown are true.
- The paid ads: someone at the firm has seen the images and placements the tools and the agency chose.
- The label: where AI use is disclosed, it clarifies what the audience sees and does not contradict it.
- The whole launch: approval sits at the level its riskiest asset calls for, with the decision on record.
Frequently asked questions
Does UK law require AI-generated marketing to be labelled?
The previous government's March 2026 report said there were no UK obligations to label AI-generated content, and CAP advised in May 2025 that there is "no blanket legal requirement in the UK to disclose the use of AI in ads". A written statement to Parliament on 18 March 2026 (HCWS1416) promised a labelling taskforce, with an interim report "in autumn". Campaigns seen in the EU face EU disclosure duties.
Does the advertising code cover business-to-business campaigns?
The CAP Code applies to business-to-business (B2B) advertising, as the ASA's November 2024 report 'AI' as a Marketing Term put it. Of around 16,000 unique AI-related paid online ads its monitoring found between July and September 2024, 68% were for B2B products and services with AI features, the largest group in the count.
Has a UK court decided whether AI tools trained on other people's images break copyright?
Getty Images v Stability AI (High Court, 4 November 2025) left it open: Getty abandoned its training claim for lack of evidence that training happened in the UK. The court found the model was not an "infringing copy", and an appeal on that point was permitted in December 2025. In March 2026 the previous government dropped its preferred data-mining exception, saying it no longer had a preferred option.
Upskilling a whole team?
Tailored data and AI training for organisations, from data literacy to technical upskilling, built around your team's real work.