Public trust in AI fell this year, and that changes how every company should describe the AI inside what it sells. The word that helped sell software two years ago now lowers the odds of a sale with many buyers, so the job is to sell the outcome and keep "AI" for the places where it earns trust.
The numbers have turned. In the Bentley University and Gallup survey published in July, 39% of Americans said AI does more harm than good, up from 31% a year earlier, and only 27% said they trust businesses a lot or some to use it responsibly. Among adults under 30, the share with at least some trust fell from 30% to 20%. Pew Research Center found that four in ten adults expect AI to have a negative effect on society over the next twenty years, 63% think it is advancing too quickly, and about six in ten are not confident that American companies will develop and use it responsibly. Familiarity rose over the same period. People know more about AI and like it less.
The advertisers closest to the problem have already adjusted. This week Marketing Brew asked the agencies that sell AI products to consumers how they approach the work, and the answers were strikingly modest. Mother LA's head of strategy described the goal as "a couple notches above apathy," and estimated that at least 90% of the online conversation about AI carries what she called an "ick" quality. VML's Howard Laubscher said the connotation of the general term is "completely negative," and that the message should be about what the product does for people, without the buzzwords at all.
When the people paid to make AI appealing are aiming for neutral, the rest of us should take the word out of the headline.
The Word Has a Price
The evidence says the label costs sales, not only sentiment. In experiments with more than 1,000 American adults, researchers led by Mesut Cicek at Washington State University showed people identical descriptions of products and services, with and without the phrase "artificial intelligence." Across eight categories, the version that mentioned AI drew lower purchase intent, because it lowered what the researchers call emotional trust, and the penalty was larger for risky purchases such as expensive electronics, medical devices and financial services. That study was published in 2024, when attitudes were warmer than they are today.
The mechanism is not mysterious. A buyer who reads "AI-powered" has to do extra work: decide whether to believe it, wonder what it will get wrong, and ask what happens to her data. A buyer who reads "your invoices reconcile themselves overnight" has to decide only whether she wants that. The first sentence asks for trust before it offers anything; the second offers something and lets trust follow.
This is why the label hurts most where the stakes are highest. Few people worry about the model inside a photo filter. A great many worry about the model inside their insurance claim.
Business Buyers Are Not Exempt
It is tempting to assume business audiences are different, that a buying committee wants to hear about AI because its board does. Part of that is true; a serious buyer will ask what the AI does and how it is controlled. But wanting the capability is not the same as trusting the claim, and regulators have noticed the difference. By DLA Piper's count, the Federal Trade Commission's May case against three marketing firms over an "Active Listening" product was its thirteenth AI-washing action since 2024, and seven of the last eight involved claims made to other businesses rather than to consumers. Holland & Knight, reviewing the same record, notes that the advertising industry's own review body keeps flagging the same faults: features announced before they exist, demonstrations that imply performance the product cannot repeat, and productivity claims that rest on how users feel rather than on measured results.
So the business version of the problem is sharper, not softer. A vague "AI-driven" in the headline adds suspicion; a specific capability claim without evidence adds legal risk. The only safe position is a claim about the result that the company can prove.
Where the AI Belongs in the Pitch
None of this means hiding the technology. A buyer who learns later that a product relies on AI nobody mentioned will trust it less than one who was told plainly at the start. The discipline is about order and proportion.
The result goes first: the hours saved, the errors avoided, the faster answer, stated as precisely as the evidence allows. The AI comes second, as the explanation of how the result is produced, at the point where a careful buyer would ask; that is usually a product page, a security review or a sales conversation, and rarely a headline. The controls come with it: what a person checks, what the system is not allowed to do, and where the data goes. For a skeptical buyer, those answers are the persuasive part.
A marketing leader can settle most of this within a quarter. Run the main headline with and without the word on the pages and ads that carry real volume, and judge the versions by qualified pipeline rather than by clicks. Ask product marketing for the evidence behind every present-tense AI claim on the website, and remove the ones nobody can document. Read the sales team's call notes for the objections buyers raise when AI comes up, because they are the most honest research the company will get this year.
The companies that do well in this period will not be the ones that say "AI" most often. They will be the ones whose customers notice, some months in, that the product simply works, and only then ask how.


