AI for marketing: how to check UK ad copy

What to take away

How the ASA and CAP regulate UK advertising claims

The Advertising Standards Authority is the UK's independent advertising regulator. It administers the CAP code for non-broadcast ads, which covers most digital, print, poster and direct marketing. Broadcast ads sit under the BCAP code.

Both sets of rules are published in the CAP and BCAP Advertising codes - ASA | CAP, and they apply to the message, not the method used to write it.

The ASA usually acts on complaints from the public, competitors and sector bodies. It also runs monitoring sweeps on topics such as misleading environmental claims and age-restricted products. A sweep can pull in dozens of advertisers at once, so a single badly evidenced draft can become a pattern case.

CAP is the Committee of Advertising Practice, the body that writes the codes. The ASA applies them. That split matters because the rules are not legislation. They are a self-regulatory contract between the industry and the public. Breaching them is not a criminal offence, but it carries commercial consequences.

Complaints are assessed against the code as it stood when the ad appeared. If you change an ad after a complaint, the ASA may still rule on the original. Keep dated copies of every version, the briefs used, and the evidence file behind each claim.

For UK sign-off teams, the practical point is that AI does not create a new regulatory category. It changes how wording is prepared, not who is accountable. The obligations stay with the advertiser whose name is on the ad.

That is the starting position for everything below, and it connects to the wider advertising rules that govern campaign approval.

Substantiation rules for AI-assisted performance claims

The Substantiation rule sits in the CAP code and requires advertisers to hold evidence before publishing an objective claim. If the copy states that a service cuts costs by a stated percentage, doubles conversion or delivers results in a set number of days, that is an objective claim. It needs proof.

The proof must be documentary. It must be in the advertiser's possession at the time the ad runs, not promised for later. It must relate to the claim as a reasonable reader would understand it, not to a narrower internal definition. A confident tone is not evidence.

The ASA has upheld complaints against claims that were technically true but based on unrepresentative samples. It has also upheld complaints where a claim rested on a survey with leading questions. The same logic applies when a summary of a report rounds the numbers up.

Three failure modes recur in AI-assisted copy. The first is invented statistics. The second is a real statistic attached to the wrong product. The third is a claim that is true for one customer presented as typical. All three breach the rule.

A workable evidence file contains the source data, the date, the sample size, the method and the person who verified it. Keep it with the ad version. If a claim cannot survive that test, rewrite it as opinion, or cut it.

The rules on claims, endorsements and testimonials set out how far you can go with comparative and testimonial claims, including claims drawn from customer reviews that may have been summarised or paraphrased. Read the Claims, endorsements and testimonials - ASA | CAP guidance alongside the core code before approving a draft.

Worked example: a claim that fails

An ad says a payroll tool saves finance teams eleven hours a week. The figure came from a blog post about a different sector. The advertiser holds no study of its own customers.

On complaint, the ASA would treat the claim as objective and unsubstantiated. The remedy is to remove the number or replace it with a sourced figure the advertiser can produce on request.

Worked example: a claim that holds

An ad says the same tool cut processing time in a trial of forty UK finance teams, and links to the trial write-up. The advertiser holds the raw data and the trial brief. The claim is specific, sourced and capable of substantiation.

Misleading claims and the CAP code on exaggeration

The Misleading claims rule prohibits ads that deceive or are likely to deceive, even if every individual statement is literally true. The test is the overall impression on a reasonable consumer, not a line-by-line reading.

Puffery is allowed. A claim that a product is the best in its class, without a measurable basis, is usually treated as opinion. A claim that it is faster than a named rival is objective and needs evidence. AI-assisted copy often adds superlatives for fluency, which blurs that line.

Omissions count. If the ad leaves out a material condition, such as a minimum spend, a contract length or a geographic limit, the ad can mislead by what it does not say. A draft rarely volunteers caveats unless instructed, so brief them in.

Qualifiers must be prominent. A footnote that corrects the headline claim is not enough if the headline still misleads. The ASA has upheld complaints where small print contradicted the main message.

Price claims need care. A discount claim must be against a genuine previous price, and a free offer must be genuinely free. Fast drafting often invents urgency, such as a sale ending on a date that does not exist.

Comparative claims must compare like with like. If a draft compares your entry tier with a rival's premium tier, the comparison is misleading even if both figures are accurate.

Testimonials and endorsements must be genuine and current. Copy that reaches for social proof may include plausible quotes from customers who never said them. That breaches the code and may also breach consumer law enforced by the Competition and Markets Authority.

Health, financial and environmental claims attract extra scrutiny. Green claims in particular have been a monitoring priority, and vague terms such as eco-friendly or sustainable are likely to be treated as objective claims needing evidence.

Recognition of marketing in AI-assisted campaigns

The Recognition of marketing rule requires ads to be obviously identifiable as ads. A reader must be able to tell that a message is marketing before they engage with it. That applies to social posts, newsletters, native placements and influencer content.

Two features of AI-assisted copy make this harder. Such copy can read like editorial or a personal recommendation. It can also appear at a volume that pushes disclosure into a template that readers stop noticing.

A paid partnership label must be clear and upfront. A hashtag buried in a block of tags is not enough. If the caption is AI-assisted, the disclosure must appear in the first lines, not appended at the end.

Advertorials need a prominent label. Native content that mimics a news layout must carry a clear identifier, and the label must not be contradicted by a headline that reads like independent reporting.

Influencer briefs should state the disclosure requirement explicitly. An AI-assisted brief should include the label wording, the placement and the timing. The advertiser remains responsible for what the creator posts.

Email marketing must identify the sender and the commercial nature of the message. A friendly personal note written for a bulk send can cross into misleading territory if it implies a one-to-one relationship that does not exist.

This is where a written disclosure policy earns its keep. It tells writers, agencies and creators what must be labelled, where the label goes and who checks it.

Common risks in automated advertising

The ASA publishes its decisions in full, and the Rulings - ASA | CAP database is the best place to see how the code is applied in practice. Real rulings illustrate how enforcement works on ad claims, including claims released at scale through automated systems.

One recurring theme is the use of automated claims in financial and credit advertising. The ASA has upheld complaints against ads that presented indicative rates or approval likelihoods as if they were certain outcomes. The issue was not the technology but the impression created.

Another theme is pricing and availability published at scale. Where automated feeds place outdated prices or sold-out stock in live ads, the ASA has treated the resulting mismatch as misleading. The advertiser is responsible for the accuracy of what the system publishes.

A third theme is social proof. Complaints about reviews and testimonials often turn on whether the quotes were genuine and whether negative material was suppressed. Automated collection does not remove the duty to verify.

A fourth theme is environmental claims. Monitoring sweeps have led to rulings against broad green claims that lacked a clear basis. Copy that reuses a sustainability phrase across a product range can create a systemic problem.

A fifth theme is age-restricted and gambling content. Where automated placement puts ads in front of the wrong audience, the ASA has upheld complaints on targeting and content grounds.

Read the rulings with your own drafts open. If a decision looks close to something you are about to publish, treat it as a warning and fix the copy before it runs.

Using Copy Advice before an AI-assisted campaign runs

The ASA Copy Advice - ASA | CAP service gives confidential, non-binding guidance on specific non-broadcast ads before they are published. It is free for advertisers and agencies, and it is the fastest way to test a risky claim.

Copy Advice is not a pre-clearance promise. It does not bind the ASA if a complaint later arrives. What it does is give you a reasoned view from the code team, which is strong evidence of good faith and a useful record if a complaint follows.

Allow time. Copy Advice is not instant, so build it into the campaign schedule rather than treating it as a last check. For a large campaign, submit the riskiest executions first.

Use the response to fix the whole family of assets. If the advice flags a claim, check every variant, including the ones written for other channels. The same wording often spreads across a campaign.

Keep the advice with the campaign file. If the ASA later investigates, a documented pre-clearance request shows the advertiser took the rules seriously. That does not decide the outcome, but it shapes it.

This sits inside the broader rules and ethics that UK marketers apply to AI tools, from data protection to intellectual property.

Sanctions risk and who signs off AI output

The ASA publishes Sanctions - ASA | CAP against advertisers that do not comply with its rulings. The available measures include requiring the ad to be amended or withdrawn, placing an ad alert against the advertiser, and referring persistent offenders to trading standards or other bodies.

The most immediate cost is the published ruling. It names the advertiser, describes the breach and stays online. For a brand, that is a permanent public record that competitors and journalists can cite.

Ad alerts go to media owners, who may then refuse to run the advertiser's ads. That can pull a campaign off air at short notice. Repeat non-compliance can lead to referral for legal action.

The reputational effect is often larger than any direct penalty. Search results for the brand carry the ruling. Procurement teams find it. Investors find it.

Sign-off should be a named person with the authority to stop the campaign. That person needs three things: the evidence file, the disclosure check and the Copy Advice response where relevant. The question of who should sign off is a governance choice, but a human signature is still required.

Use a short checklist before release.

Run that list once per campaign, not once per asset. It takes minutes and it catches the errors that fast drafting introduces most often.

Common questions