The complete guide to the TGA "overall impression" standard in Australia (2026) — and why a ChatGPT or Claude prompt can't apply it

Every Australian health brand experimenting with AI has tried it: paste the ad copy into ChatGPT or Claude, ask "is this TGA compliant?", and get back a confident sentence-by-sentence assessment. The output looks rigorous. It is answering the wrong question.

Regulators do not assess sentences. In 2023/24 the TGA issued 4,806 advertising removal requests and 147 infringement notices totalling $2.20 million, and the assessments behind those actions were made against whole advertisements — the impression created by copy, imagery, product name and context together. A workflow that grades text line by line can pass every line and still publish a breach.

The Therapeutic Goods Advertising Code 2021 is applied to the advertisement as a whole. The operative question is what a reasonable consumer would take from the complete piece — not whether any individual phrase, viewed in isolation, states a prohibited claim. The same whole-of-conduct logic runs through the other three frameworks: the Australian Consumer Law prohibits conduct that is misleading in its overall effect (ACL s.18) and false or misleading representations however conveyed (ACL s.29); AHPRA's advertising provisions under the National Law assess what the advertising as a whole conveys about a regulated health service, including the prohibition on testimonials (National Law s.133); and FSANZ Standard 1.2.7 governs the health claims a food conveys, not merely the words used to convey them.

Three practical consequences follow. Implication counts: a claim can be made by juxtaposition, imagery or omission without ever being written down. Context resolves ambiguity: the same phrase can be compliant in one setting and a breach in another. And precision is anchored to the register, not the dictionary: for listed medicines, therapeutic claims must reflect the product's permitted indications on the ARTG (TGA s.8(2)) — which means compliance depends on facts about the specific product that no general-purpose model knows.

Consider a supplement brand's Instagram post. The copy reads: "Feeling flat by 3pm? You're not alone. Meet [Product] — magnesium, the mineral your busy life asks a lot of. Get your spark back." Run those sentences through a chat prompt individually and each survives: a relatable question, an empathy line, a factual ingredient statement, an upbeat sign-off with no therapeutic term in it.

Assessed as a whole piece, the post reads differently. The symptom hook ("feeling flat by 3pm") names a fatigue complaint. The product is presented immediately after it. "Get your spark back" implies restoration of a lost function. The overall impression a reasonable consumer forms is that the product treats or relieves fatigue — a therapeutic claim that must match the product's ARTG permitted indications under TGA s.8(2), and that may not be available to this product at all. Add a tired-office-worker image and the impression strengthens further, without a single word changing.

This is the structural blind spot of line-by-line review. The breach does not live in any sentence. It lives in the relationship between the sentences, the image and the product — exactly the layer a sequential text assessment never sees.

The limitation is architectural, not a question of model quality — the same gaps apply whichever model sits underneath.

The assessment unit is wrong. A chat prompt evaluates the text it is given, sentence by sentence. It does not weigh imagery, layout, placement or the cumulative effect of the piece unless every element is described to it perfectly — and the person describing the ad is the person who wrote it, with the same blind spots.

The product context is missing. TGA s.8(2) ties therapeutic claims for listed medicines to the specific indications on that product's ARTG entry. A general prompt working from "TGA knowledge" cannot know whether "supports energy levels" is a permitted indication for this listing. Most enforcement exposure for supplement brands lives precisely there.

The rulebook is frozen and the engine drifts. A DIY prompt is anchored to the Code as it read when the prompt was written — and the Therapeutic Goods Advertising Code is under formal review by the TGA (2025–26). Meanwhile the model underneath updates continuously; outputs shift and nothing measures whether the workflow got better or worse at catching a s.8(2) or s.19(2)(f) issue.

There is no record. If the TGA or AHPRA asks how an advertisement was reviewed, a chat history is not documented evidence of reasonable steps. It is not timestamped against a Code version, not linked to the product profile, and not retained in any regulator-ready form.

None of this means brands should stop drafting with AI. Generative models are exceptional at producing copy quickly. It means the checking function needs different architecture from the writing function.

The workable pattern separates creation from validation. The marketing team drafts at full speed with whichever AI it prefers. Before publication, the complete piece — copy and imagery together, connected to the specific product profile — passes through a governed validation layer that assesses the overall impression against current versions of TGA · AHPRA · FSANZ · ACCC simultaneously and returns a low, medium or high risk assessment with the specific clause behind every finding, rather than a false pass/fail binary.

Governance is what makes that layer trustworthy where a chat prompt is not: a version-controlled regulatory knowledge base that is updated as guidance changes, regression testing against a locked test set after every model or knowledge base change so drift is measured rather than discovered, compliant rewrite suggestions that preserve the marketing intent instead of stripping it, human escalation on high-risk findings, and a timestamped audit record in the Audit Vault retained for the duration of the subscription. The person or tool that creates an asset is never its final sign-off authority.

CompliGUARD provides compliance and creative marketing insights based on TGA, FSANZ, ACCC and AHPRA guidance as current at the time of review. The Therapeutic Goods Advertising Code is currently under formal review by the TGA (2025–26). CompliGUARD assessments reflect the current Code; guidance may be subject to change. This content does not constitute legal advice.

• Publishing without checking the complete advertisement’s overall impression.
• Treating one regulator’s checklist as coverage for every applicable framework.
• Failing to retain evidence showing what was reviewed and approved before publication.

CompliGUARD reviews the complete advertisement against TGA · AHPRA · FSANZ · ACCC requirements, identifies clause-level risks, proposes compliant wording, and retains the result in an audit trail before publication.