Claims Discipline: How to Market AI Without Lying
A practical standard for honest AI marketing claims. How to separate what your software does from what you wish it did, and write copy you can defend.
Most AI products lose trust at first contact. Not because the software is bad, but because the marketing promised something the software cannot do. The buyer tries it once, hits the gap, and never comes back. The headline did the damage.
I sell assurance for a living, so claims discipline is not a nicety for me. It is the product. Here is the standard I hold my own copy to, and the one I would hand to anyone shipping AI.
Separate what it does from what you wish it did
Every claim falls into one of three buckets. Write them down before you write a word of copy.
The first bucket is what the software does today, reliably, that you could demo to a skeptic in the room. The second is what it does sometimes, under good conditions, with caveats. The third is what you hope it will do once you finish building it.
Only the first bucket belongs in a headline. The second can appear in the body if you name the conditions honestly. The third is a roadmap, not a claim, and it goes nowhere near the page until it moves to bucket one.
The trap is that the third bucket is always the most exciting. It is the version of the product you see in your head. Marketing it is lying, even when you fully intend to build it.
Write copy you can defend
The test for any sentence on a product page is simple. If a buyer reads it, signs up, and points to it during a support call, can you defend it without flinching?
That test kills most hype on contact. "Fully automated" rarely survives it, because something almost always needs a human. "Understands your business" does not survive, because the model does not understand anything. "Never misses" does not survive, because everything misses sometimes.
What survives is specific and bounded. "Drafts the first reply so you edit instead of starting cold." "Flags the three records that look wrong so you check them first." Narrow claims sound smaller. They also hold up, which is the entire point.
This is the same constraint I build into the software itself. A system that will not fabricate a number or a citation is one whose marketing cannot either, because the marketing has to match what the system actually refuses to do. I wrote more about why that constraint is the moat over on Agency Script.
Treat overclaiming as a cost, not a shortcut
The argument for overclaiming is always that honest copy converts worse. Sometimes the inflated version does pull a higher signup rate on day one.
Then the cost arrives later and larger. The mismatched buyer churns inside a week. The support load climbs because people expected the headline, not the product. Worst of all, the refund email and the angry post both quote your own marketing back at you. You armed them.
Honest claims pre-qualify. The person who signs up already understands what they are getting, so they activate faster, stay longer, and tell the truth about you to the next buyer. Lower top of funnel, higher everything after. For a solo operator who cannot absorb a churn spike, that trade is not close.
The standard, in one line
Say only what the software does today, say it specifically enough that a buyer could check, and never let a roadmap item wear the costume of a feature.
Honesty is not the brake on the sale. With AI, where every buyer has already been burned by an overclaim, it is the fastest way to earn the one thing capability cannot buy: a buyer who believes the next sentence you write. If you want to see how this runs across a whole portfolio, that is the work.