Direct answer: when marketing or product teams use AI to draft environmental or sustainability copy for flights, packages, hotels or tours, the claims themselves — and the approval path that lets them go live — must stay outside freeform generation. A model may propose candidate wording. It must not invent eco labels, absolute “green / eco-friendly / sustainable” claims, or publish without a claims register, a substantiation pack, and named human approval. Vague absolute terms without evidence are high-risk under the CMA Green Claims Code and ASA CAP Code Section 11.
That is the operator question Kaize is answering here. We are a UK AI product studio for travel and hospitality. This page is about **marketing claims and advertising compliance for AI-generated green copy** — not ATOL pre-choice disclosure, Package Travel Regulation journey changes, GDPR booking amendments, or distribution layers. If your team is bolting a generative model onto campaign, landing-page or email workflows, the control question is which environmental assertions must remain deterministic product and compliance artefacts.
Primary evidence is first-party UK advertising and consumer-protection material: the CMA Green Claims Code on environmental claims on goods and services (truthful and accurate; clear and unambiguous; does not omit or hide important information; fair comparisons; considers the full life cycle; substantiated); ASA CAP Code Section 11 Environmental claims (11.1 basis clear; 11.2 terms clear; 11.3 absolute claims need a high level of substantiation; 11.4 full life cycle unless stated otherwise; take account of the CMA Green Claims Code); and ASA AdviceOnline: Travel marketing — Environmental claims (17 September 2024), which applies those rules to travel ads and flags absolute wording, aspirational future initiatives, “sustainable aviation”, SAF claims and CO₂ comparisons.
Why this is a claims-control problem, not a prompt problem
Prompt instructions such as “be accurate” or “avoid greenwashing” do not create a defensible claims process. The CMA Green Claims Code expects businesses to ensure environmental claims are truthful, clear, complete enough not to mislead by omission, fair in comparisons, life-cycle aware, and backed by robust evidence held before the claim is made. CAP 11.1–11.4 expect the basis of the claim to be clear, the meaning of terms to be clear to consumers, absolute environmental claims to meet a high substantiation bar, and life-cycle impacts to be considered unless the ad states otherwise.
ASA’s travel-specific advice underlines the same gap: absolute wording such as “environmentally friendly” must not be used unless there is convincing evidence of no environmental damage over the full life cycle. Aspirational or future initiatives must not imply that significant mitigation has already been delivered (for example, the Lufthansa “PROTECTING ITS FUTURE” ruling referenced in ASA advice, 10 March 2023). “Sustainable aviation” can be read by the general public as an absolute claim without explanation (Etihad, 12 April 2023 — insufficient evidence for an absolute claim). SAF claims need material information on remaining emissions (Virgin Atlantic, 7 August 2024 — “100% sustainable aviation fuel” misleading without that context). CO₂ comparisons need rigorous evidence and a clear basis, including factors such as load and seat density (Ryanair 2020 example in the same advice). None of that is solved by a better system prompt.
So the product problem is a **claims gate**: which strings are allowed into live marketing surfaces, what evidence pack attaches to each string, who signs off, and how the model is prevented from inventing new eco labels on the fly.
What AI may draft vs what it must not own
What the model may draft
- Candidate paragraphs that only use **pre-approved claim IDs** from a claims register (no new absolute environmental assertions).
- Plain-language explanations of already-substantiated, scoped claims (for example, a specific verified metric with date, scope and method stated).
- Questions for marketers that surface missing life-cycle, baseline or comparison assumptions before copy is requested.
- Internal drafts clearly labelled non-publishable until human claims approval.
What the model must not own
- Inventing or inventively paraphrasing absolute claims such as green, eco-friendly, environmentally friendly, carbon neutral, net zero delivered, or sustainable aviation without an approved register entry and evidence pack.
- Implying that future SAF, offset or fleet initiatives have already delivered significant mitigation.
- CO₂ or emissions comparisons without an approved basis (scope, methodology, load/seat-density factors where relevant).
- SAF percentage or “100% SAF” wording that omits material limitations and remaining emissions information.
- Publishing to web, email, OTA modules, agent scripts or paid media without named human approval tied to the claim IDs used.
That split matches how Kaize thinks about controlling AI agents that access booking and operational systems: models propose; governed components decide what is allowed to reach a customer-facing surface. Marketing claims need the same discipline as booking actions — different rules, same control pattern. For a wider practical framing of where AI belongs in travel product work, see our AI for travel practical guide.
Build a claims gate as a product and process component
Treat environmental marketing claims as a controlled component, not free text the LLM owns end-to-end.
- **Claims register:** every customer-facing environmental assertion has an ID, approved wording variants, forbidden paraphrases, channel scope, and expiry/review date.
- **Substantiation pack:** evidence, methodology, life-cycle notes, comparison basis, and limitations stored before the claim can be selected — not after a campaign ships.
- **Named human approval:** compliance or marketing owner signs the claim set for that asset; the model cannot self-approve.
- **Generation constraint:** prompts and tools may only interpolate approved claim IDs; freeform “make it greener” is blocked at the API boundary.
- **Channel coverage:** web, app, email, paid social, agent scripts, partner feeds and OTA modules all consume the same register — not a separate chatbot policy.
- **Audit trail:** log which claim IDs appeared in which asset, who approved, and which evidence version was current.
Wire the gate into every path that can emit sustainability copy: CMS blocks, generative landing-page tools, email assistants, and agent-facing macros. If a surface can invent a new absolute green claim, it is not gated.
Life-cycle and substantiation — what “outside freeform generation” means
CAP 11.3 and 11.4 and the CMA Code push operators toward the uncomfortable part of travel marketing: aviation, accommodation and tour products have material environmental impacts across the life cycle. Absolute claims therefore need a high level of substantiation; life-cycle impacts should be considered unless the ad makes the limited basis clear. ASA travel advice is explicit that absolute environmental friendliness claims need convincing evidence of no environmental damage over the full life cycle — a bar most flight, package and tour propositions will not meet as freeform AI copy.
Operationally, that means substantiation is a **data and document problem** before it is a language problem: scope of the claim, baseline year, measurement method, what is excluded, whether offsets are used and how they are described, and whether a comparison is fair. The model can help draft surrounding narrative once those facts are locked. It must not invent the facts.
How this fails when AI is bolted onto marketing
The failure mode looks like other travel AI that looks fine in testing and breaks in production: demo copy uses carefully reviewed claims; production prompts ask the model to “emphasise sustainability” across hundreds of destinations; absolute phrases appear in titles, meta descriptions and agent macros that nobody re-substantiated. SAF and net-zero roadmap language drifts from aspirational to implied present tense. Comparisons lose their basis. The brand discovers the gap when ASA, a competitor challenge, or a retailer takedown arrives — not when the prompt was written.
Instrument the publish path, not only the prompt log. Log claim IDs rendered, whether the substantiation pack was current, and whether named approval existed for that channel. Adjacent booking-disclosure work (for example ATOL AST 1.4 pre-choice disclosure for AI booking assistants) is a different control surface: that article is about consumer choice disclosures, not green advertising claims. Do not collapse them into one “compliance prompt”.
What Kaize will and will not claim
This article is product and process guidance for UK travel operators, OTAs, hotels, tour operators and travel-tech teams implementing AI-assisted marketing. It is **not legal advice**. Operators remain responsible for advertising compliance, ASA/CAP obligations and consumer-protection duties — check counsel and your compliance owners before changing live claims. Kaize does not claim to have audited your green claims, delivered enforcement outcomes, or guaranteed ASA clearance.
If you want a structured review of where AI belongs in marketing and product workflows — and which claim controls must stay deterministic — start with Kaize’s AI Opportunity Review.
FAQ
Can we let the model invent “eco-friendly” variants if we add a disclaimer?
Disclaimers do not repair absolute environmental claims that lack convincing life-cycle substantiation. ASA and the CMA Code expect the claim itself to be clear, accurate and evidenced. Keep absolute variants out of freeform generation; only emit register-approved wording.
Are SAF and “sustainable aviation” safe for generative landing pages?
Not as freeform absolutes. ASA travel advice treats “sustainable aviation” as potentially absolute without explanation, and highlights SAF claims that omit material information on remaining emissions. Use only pre-approved, scoped wording with required limitations — or do not generate those claims.
Does this replace ATOL or Package Travel disclosure work?
No. This page is about environmental **advertising and marketing claims**. ATOL AST 1.4 pre-choice disclosure and Package Travel journey controls are separate product requirements. AI systems need both kinds of gates where relevant — they are not interchangeable.
Is this legal advice?
No. Advertising and consumer-protection compliance remains with your marketing, legal and compliance owners. This page explains how AI marketing tooling should be designed so claims control stays outside freeform generation.