AI can draft a week of marketing in minutes. The catch is that it can publish something false, off-brand, or non-compliant just as fast — and under your name, not the model's. "Can you trust AI-generated marketing?" is really the whole question, so this post lays out the real risks and a checklist you can use to judge any tool before you let it near your brand.
The four real risks
- Made-up claims. Language models state things confidently whether or not they are true — "the leading provider of X," a precise-sounding statistic, an award you never won. This is not just embarrassing: under advertising rules (the FTC in the US, and equivalents elsewhere) you are liable for a false or unsubstantiated claim in your marketing, even if a machine wrote it.
- Off-brand and tone misreads. The model gets the subject right most of the time and the voice — or the moment — wrong some of the time. Cheerful launch copy scheduled for the morning your service goes down is the classic example.
- Provenance and disclosure. A growing number of jurisdictions expect AI-generated media to be disclosed, and customers increasingly ask. If a regulator or client asks "is this AI, and can you prove where it came from?", you need an answer better than a shrug.
- Your data and who owns the output. Two quiet risks: is your brand and content used to train someone else's model, and do you actually own — and can you export — what the tool produces?
A checklist to judge any AI marketing tool
Trust isn't a vibe; it's a set of mechanisms you can ask about. Run any tool through these:
- Does it check claims against anything? Or is every draft a fresh chance to publish an unsupported statement?
- What can it publish without you? "Review available" and "review required" are different products. The safest default is that nothing goes live without a human yes.
- Is there a real audit trail? Can you prove who approved a post and when — a tamper-evident record, not an editable activity feed?
- Can you verify AI imagery? Do generated images carry content credentials you can check independently?
- What happens to your data? Get the training answer and the deletion timeline in writing.
The single most important safeguard
If you take one thing from this: the highest-leverage safety feature is a human approval gate — a required moment where a person says "yes, publish this" before anything goes live. Every other safeguard (claim-checking, content credentials, brand-voice memory) only means something if there is a point where a human confirms the result. We wrote up why we made that the non-negotiable default — including what we learned from teams who asked us for autopilot — in why Pappus requires you to approve every post.
How Pappus is built for this
These risks are the reason Pappus exists in the shape it does. Every claim in a draft is checked against a Claims-DB of facts you have approved, and anything unsupported is auto-hedged or flagged before you ever see it (we call it Substantiation Mode). Every AI-generated image carries a C2PA content credential you or a regulator can verify independently. Every approval is recorded in a signed audit trail, and nothing publishes without one. Your content is never used to train models, and there is a documented deletion timeline when you leave.
The short version of the trust posture lives on the home page's trust section, and the approval workflow itself is described on the marketing approval workflow page. Trust you can check beats trust you are asked to take on faith — that's the whole idea.