Five AI Engines, One Wagering Creative: Why Generic Tools Miss Compliance Breaches

Three things have quietly been slipping through wagering creative reviews.

Models who read 18 to 24 shown in wagering scenes. Winners portrayed as more attractive after the bet. "Guaranteed" or "definite outcome" framing on multis.

The AANA Wagering Advertising Code rules against all three. Each has been the subject of Ad Standards complaints. And generic AI tools, asked to review that creative, score it as compliant anyway.

So we benchmarked them.

The short version

• One wagering creative, one brief, five AI engines, the same regulatory framework applied to each.
• Checked It scored 91. Microsoft Copilot 61, ChatGPT 59, Claude 44, Gemini 40.
• The gap is not model quality. It is regulatory context, and whether the tool knows which four frameworks apply at once.
• A single Australian wagering ad sits under the AANA Wagering Advertising Code, the National Consumer Protection Framework, the Interactive Gambling Act 2001, and section 18 of the Australian Consumer Law. Generic models pick up one or two.

What did we actually test?

One test wagering creative, built to carry three specific, identifiable AANA breaches.

One brief. The same brief, verbatim, to every engine. Same regulatory framework named in each. No follow-up prompting, no coaching, no second pass.

Score is accuracy, not verbosity. It measures how closely each engine's compliance output matched the actual code breaches present in the creative: breaches correctly identified, clause correctly cited, and no invented breaches counted as hits.

This is a one-shot prompt accuracy test. It is deliberately unkind to general-purpose tools, because that is how they get used in practice. Nobody running a campaign has time to prompt-engineer their way to a defensible compliance answer at 6pm on a Thursday.

The results

Checked It, running Gambling & Gaming v3.2.2, scored 91.

Microsoft Copilot scored 61. ChatGPT scored 59. Claude scored 44. Gemini scored 40.

Every general-purpose engine failed to reach a pass mark on creative built to breach three clauses of a published Australian code.

Which AANA clauses do generic AI tools miss?

The three breaches in the test creative map to three clauses of the AANA Wagering Advertising Code.

Section 2.3. Advertising or marketing communication for a wagering product or service must not depict a person aged 18 to 24 years old engaged in wagering activities.

Section 2.5. It must not state or imply a promise of winning.

Section 2.7. It must not state or imply a link between wagering and sexual success or enhanced attractiveness.

None of these are edge cases. They are the three most reliably breached clauses in Australian wagering creative, because each one describes something a creative director naturally reaches for. Aspirational casting. Confidence in the outcome. The winner who gets the room's attention.

The clause language is also unforgiving. "Depict a person aged 18 to 24" is not a question of the model's actual age. It is a question of how they read on screen.

Why do generic AI tools score wagering creative as compliant?

Generic AI tools, used as-is, are pattern matchers.

Ask ChatGPT or Gemini whether an ad is compliant and it will produce something that looks like a compliance review. Confident tone, sensible structure, plausible headings. What it does not have is the code in front of it, a rule for how each clause is assessed, or any obligation to cite the clause it is relying on.

So it reasons from the general shape of "responsible gambling advertising" it absorbed in training. That shape catches the obvious things: no under-18s, include the helpline, do not target children. It does not catch a 22-year-old reading as a 22-year-old in a betting scene, because nothing in the general shape says that is a breach.

Checked It is built the other way round. Sector regulations first, with detailed guardrails and multi-stage creative-to-regulatory-code assessments, so every finding is tied back to a specific clause. That is the design difference behind the gap, and it is why we shipped Gambling & Gaming v3.2.2 rather than a better prompt.

Why does one wagering ad sit under four frameworks at once?

This is the part that breaks general-purpose tools.

A single piece of Australian wagering creative is simultaneously governed by:

• The AANA Wagering Advertising Code, administered through Ad Standards, covering content and portrayal.
• The National Consumer Protection Framework (NCPF), covering consumer protection obligations across online wagering.
• The Interactive Gambling Act 2001, administered by the ACMA, covering what may be offered and promoted.
• Australian Consumer Law section 18, covering misleading or deceptive conduct, which applies to every claim in the ad regardless of what the codes say.

Each of the four applies. Not the strictest one. All of them, at the same time, to the same thirty seconds of film.

Generic models typically identify one or two, then write a confident summary as though that were the full picture. A review that covers half the applicable frameworks is not half a review. It is a false clear.

What does a compliance failure actually cost?

Two separate cost lines, and they are often confused.

Ad Standards has no power to fine you. The AANA codes are self-regulatory. What an upheld complaint costs is the withdrawal, the reshoot, the wasted media spend, the platform and broadcaster conversations that follow, and a public case report with your brand name on it.

The ACMA does have power to fine you, under the Interactive Gambling Act and related instruments, and it has been using it. Recent Australian examples: Unibet penalised $1 million in 2025 over BetStop self-exclusion breaches. PointsBet penalised $500,800 in May 2025 for spam and self-exclusion breaches. Tabcorp paid $158,400 in February 2026 for taking illegal in-play sports bets.

None of those penalties were for the bet itself. They were for the compliance layer around it. That is the honest read of the current enforcement climate: the regulator is active, the amounts are real, and the failures being punished are process failures, not product failures.

What this means for your next wagering campaign

If you are running a general-purpose AI tool as your first-pass compliance check, you are getting a confident answer with roughly half the applicable law behind it.

Shift the check left. Run it at brief and storyboard stage, when a casting note or a headline is still a sentence in a document and not a shoot day. That is where a clause-cited finding is cheap. After the shoot, it is a reshoot.

And keep the human in the loop. This is a pre-flight pass, not a legal sign-off. The value is that your compliance and legal people stop spending their hours on the first-pass triage a machine can do, and spend them on the judgment calls only they can make.

Frequently asked questions

Can AI check wagering ads against the AANA Wagering Advertising Code?

Yes, but only if the tool has the code built in. Checked It pre-screens wagering creative against the AANA Wagering Advertising Code, the NCPF and related Australian frameworks, returning clause-cited findings. General-purpose engines reason from training data instead, which is why they scored 40 to 61 in this benchmark.

Why do ChatGPT, Claude, Gemini and Copilot miss wagering compliance breaches?

Because they are pattern matchers. They recognise the general shape of responsible gambling advertising and check for the obvious things, but they do not hold the code, apply clause-level tests, or cite the section they relied on. Subtle breaches such as casting age or implied promise of winning pass straight through.

Which AANA Wagering Code clauses are breached most often?

In practice, three. Section 2.3, depicting a person aged 18 to 24 engaged in wagering. Section 2.7, implying a link between wagering and sexual success or enhanced attractiveness. Section 2.5, stating or implying a promise of winning. All three sit close to instincts creative teams reach for naturally.

Which regulations apply to a single Australian wagering ad?

Four at once. The AANA Wagering Advertising Code, the National Consumer Protection Framework, the Interactive Gambling Act 2001 administered by the ACMA, and section 18 of the Australian Consumer Law. They apply simultaneously, not in order of strictness, and a review that covers only one or two is incomplete.

Is an AI compliance check a substitute for legal review?

No. It is a pre-flight pass with a human in the loop. Checked It does the repetitive first-pass triage and cites the specific clause behind each finding. Your legal and compliance team keeps the judgment and the final sign-off, but spends its hours where they actually count.

What penalties apply to non-compliant wagering advertising in Australia?

Ad Standards cannot fine you; an upheld complaint costs withdrawal, reshoot, wasted media and a public case report. The ACMA can fine you under the Interactive Gambling Act. Recent examples include Unibet at $1 million in 2025 and PointsBet at $500,800 in May 2025.

When in the process should creative be compliance-checked?

At brief and storyboard stage, before production spend is committed. A casting note or a headline is cheap to change while it is still a sentence in a document. The same finding after a shoot means a reshoot, a media reschedule, and a conversation with the client nobody wants to have.

If your wagering creative touches the AANA Wagering Advertising Code, run it through Checked It before it reaches legal.

Free trial at checkedit.ai · How Checked It works · Questions: stusheridan@hiaitus.ai