Claims and compliance
Can you legally say that in an advert?
Best in the area, guaranteed results, clinically proven — which marketing claims need evidence you can produce, and what happens to the ones that do not.
How MarketWar handles it
An unsubstantiated claim is never publishable here: every claim is classified into 1 of 4 states — verified, confirmed by you, inferred and pending, or prohibited — and a prohibited superlative does not reach an advert. AI-generated assets ALWAYS carry a disclosure and a C2PA-style provenance record with creator consent attached, because a synthetic image passed off as a photograph is a disclosure failure before it is a creative choice.
Most small-business advertising breaks the rules by accident, in the same four words: best, guaranteed, number one, proven.
They do not feel like claims. They feel like enthusiasm. They are claims, and the test applied to them is whether you can produce evidence — not whether you believed it when you wrote it.
The four states a claim can be in
Every claim is classified, and the classification decides what can happen to it:
Verified — evidence exists and is attached. Publishes.
Confirmed by you — you have asserted it as the business owner and taken responsibility for it. Publishes, with your name against it. "We have been trading since 2009" belongs here: you know, and no external source is going to tell us.
Inferred and pending — the model produced it and nothing stands behind it. Does not publish; goes to validation.
Prohibited — an unsubstantiated superlative, a health claim nobody can make, a guarantee with no mechanism. Does not publish at all.
The distinction between the third and fourth matters. "Pending" is a workflow: find the source, and it moves. "Prohibited" is a refusal.
Why a superlative is harder than it looks
"Best plumber in Leeds" is a comparative claim about every plumber in Leeds. To substantiate it you need a basis — a measure, a source, a date — and that basis has to be the kind a regulator would accept.
The practical alternative is almost always stronger anyway. "Best" is what everybody writes, so it carries no information. "We answer the phone within two rings between 7am and 7pm, and if we cannot come today we will tell you today" is specific, checkable, and far more persuasive — and it is the kind of sentence that gets you found and cited by an AI assistant, which now reads for specifics rather than adjectives.
Synthetic content carries a label, always
Any AI-generated asset gets a disclosure and a provenance record, with creator consent recorded.
This is not us being cautious on your behalf. Disclosure requirements are tightening across platforms and jurisdictions, and the expensive version of this problem is discovering in two years that a back catalogue of unlabelled synthetic assets needs auditing. Labelling at the point of creation costs nothing; retrofitting it costs a project.
The regulated categories
In health, finance and the other regulated contexts, the threshold moves. A sentence that is merely optimistic in one industry is a regulatory matter in another, and the engine blocks an inferred or unsourced claim in those contexts rather than softening it.
Softening is the trap. "May help support" keeps the unsupported claim and removes the only word a reader could have checked.
What this does not do
It is not legal advice, and it cannot substantiate a claim for you. It also cannot read your mind about what you can prove — if you hold the evidence, confirm the claim and it publishes with your name on it.
What it removes is the specific failure of publishing a claim that nobody, anywhere, can back up — which is the one that generates the complaint.
The question to ask
Take your current homepage and ask, sentence by sentence: if somebody asked me to prove this, what would I send them?
Every sentence with no answer is one to rewrite as something specific. The pages that result are better marketing and lower risk, which is an unusual combination.
What this does not do
It is not legal advice and it cannot substantiate a claim for you. If you say your product is the fastest, somebody has to hold the evidence — the engine's job is to stop that sentence being published while nobody does.
Common questions
Can I say I am the best in my area?
Only with evidence that stands up — a superlative is a factual claim about every competitor in that area. Without substantiation it is classified as prohibited and does not publish.
What about "guaranteed"?
A guarantee is a contractual promise, not an adjective. If you mean a refund policy, say the policy. The engine treats an unbacked guarantee as prohibited for the same reason it treats a superlative that way.
Do I have to say an image was AI-generated?
Here, yes — AI assets always carry a disclosure, with a provenance record. Platforms and regulators are converging on this, and retrofitting labels across a back catalogue is far more expensive than labelling as you go.
What are the regulated categories?
Health, finance and the other categories where an ordinary claim becomes a regulated one. A claim in those contexts that is inferred or unsourced is routed to validation rather than hedged into something unfalsifiable.
Related
How do you publish AI content without inventing statistics?
AI writes a convincing statistic and a plausible citation with equal confidence. What an evidence classifier blocks, and what it lets through.
Are you allowed to use that photo in a paid advert?
Organic permission is not paid permission, a stock licence is not a model release, and the 17 rights a campaign asset actually needs.
How do you research a competitor without writing knocking copy?
Research naming a rival's weakness is easy to turn into advertising that attacks them, which usually loses. What to do with it instead.