👋 Hi, it’s Kyle and welcome to Growth Unhinged, my weekly newsletter exploring the hidden playbooks behind the fastest-growing startups.

Last year my friends at Irrational Labs ran a groundbreaking buyer psychology study that asked: do people actually care whether your product is “AI-powered”? Now “AI-powered” has been replaced by “AI agent”, and it’s EVERYWHERE.

Today I’m bringing you new Irrational Labs research on AI messaging led by Isabel Macdonald. Irrational Labs is a team of behavioral scientists who work inside product and growth teams, and has worked with the likes of Microsoft, Canva, LinkedIn, and Intuit (you can get in touch here).

Something strange is happening in AI marketing.

A year ago every product was “AI-powered.” Now every product has “an AI agent.” And the ambitious ones have a whole team of them: a research agent, a drafting agent, a scheduling agent, each with its own job title.

Irrational Labs wanted to know whether this trend in product marketing matters to actual buyers. So we ran an experiment.

We showed groups of U.S. software buyers marketing materials for real AI products. Each group saw the same website designs with one small difference: whether or not the ads described the product as an “agent.” For example, one group saw a website promoting a social media tool that could help you create and schedule content. Another group saw almost the exact same website and product features, but was told it was a social media agent and this agent could help them create and schedule content.

Then we asked each group whether they’d buy the product and how much they’d pay.

The data was consistent: calling a product an “agent” did not make people more willing to buy. If anything, it slightly decreased interest, though this effect was not statistically significant (p = 0.084). It also didn’t seem to change people’s perception of helpfulness, ease of use, or trustworthiness, as measured by our survey.

But one framing did move people: describing the product as a “team of agents.” People were 6.3 percentage points more likely to want to buy when the same product was described as a team of agents rather than one agent (p = 0.022). The team framing also drew more interest than the no-agent version, though that gap wasn’t statistically significant. And in every version the features and page design were identical, so the only thing that changed was the wording.

A behavioral science concept called the unpacking effect may explain how a label alone made more people willing to buy. First, the experiment.

Related: How to build an AI brain for your positioning

Our core research question was whether calling an AI tool an “agent” on marketing materials would increase user interest and willingness to pay.

To answer this, we surveyed 1,014 U.S. software buyers. Each participant was shown a marketing page for a real AI product and asked how much they would pay to use the tool each month. We also asked users to rate how helpful, easy to use, and trustworthy they thought the product would be.

The marketing pages were manipulated to fit our three conditions:

We created these three conditions for three different AI products aimed at small businesses: a calendaring service, a social media toolkit, and an accounting software. Each participant saw only one design, which displayed one product and one framing condition.

Here’s an example of the mockups people saw for the social media toolkit:

You can see that the designs are almost identical, with the exception of the header and subheader copy, as well as the labels above each feature. There is no difference in the features presented or the general design.

Keeping the designs and features consistent allows us to test whether a simple marketing change to how the product is described (i.e., using the term “agent”) will change buyer interest.

Desire to buy

Multiple-agent framing drove the most interest from buyers, followed by no-agent framing. One-agent framing had the least interest. Only the difference between the one-agent and multiple-agent framing is statistically significant (p = 0.022). This effect holds up under a variety of different modeling assumptions (robust SEs, logit, demographic controls, and excluding speeders, p = .016 to .033).

This suggests that multiple agents beat one agent, but it’s not conclusive how these compare to the no-agent framing. This pattern was consistent across all three products (calendaring, social media, and accounting tools).

Willingness to pay

The dollar amount people would pay each month was also highest for multiple-agent framing, followed by no agent, and finally one agent, but none of these differences is statistically significant.

Ease of use

Surprisingly, people thought the tool would be harder to use when framed as multiple agents relative to the one-agent framing (p = 0.057) and the no-agent framing (p = 0.023). In general, we’d expect people to buy and pay more for tools they find easier. But in this case, the framing that drove the most interest (multiple agents) had the lowest perceived ease.

Helpfulness and trustworthiness

We also asked participants to rate the perceived helpfulness and trustworthiness of the tools, and to identify the number of tasks they thought the tool could perform from a list of realistic and non-realistic tasks. We found no significant difference in these measures across the conditions.

Related: How Notion, Rippling, and Profound turned product releases into a powerful growth lever

One possible explanation is a pattern behavioral scientists call the unpacking effect.

When you break something into its parts, people tend to judge it as bigger, more likely, or more valuable than the same thing described as a whole. This is the psychology behind the advice that you shouldn’t wrap all your Christmas presents in one box: several separate gifts feel like more than the same items in one package.

Tversky and Koehler first showed this with probability: people rate "death resulting from heart disease, cancer, or some other natural cause" as more likely than "death resulting from natural causes," even though they mean the same thing. Later research found the same pattern in how people value things (Van Boven and Epley 2003).

You’ve probably experienced the unpacking effect yourself. For example, which of these vacation packages would you choose?

For most of us, the second one sounds more appealing. Unpacking “water sports” into separate components makes us feel like we’re getting more, making us more willing to buy and more willing to pay a higher price.

The same mental phenomenon could explain why a team of agents piques our interest. Describing each task as a separate agent (analysis agent, creative agent, admin agent, growth agent) creates more psychological separation between the tasks, leading to a higher perceived value relative to one agent that can manage everything.

It’s important to caveat that this is only one potential story for our results. We asked people to share perceptions and estimate willingness to pay, but we didn’t see their actual buying behavior. A survey like this can’t always pin down mechanisms or rule out alternative explanations.

But these findings do show that a relatively minor change in framing can impact the way people perceive AI products. More research with real purchasing outcomes can “unpack” this further.

So should all AI products re-brand as a team of AI agents? We certainly hope not.

The bigger lesson here is about making individual features visible. People often value the sum of the parts more than the whole. Products look more valuable when buyers can see each separate job they do. Lumping everything into one system or one agent hides those benefits.

This only holds for jobs buyers actually want, though. If one feature feels irrelevant to them, they may reject the whole package, a pattern behavioral scientists call idiosyncratic fit. So unpack the jobs that matter and leave out the filler.

A team of agents is one way to keep the jobs visible. But marketing and product teams have many other options here, including:

One warning: more features or agents may also seem like more work to the user. In this study, people valued the products with multiple agents while also expecting them to be harder to use.

When things seem effortful, people are more likely to procrastinate and drop off, causing retention issues downstream. So if you split your product into parts, making the setup seem simple and showing how everything fits together will be key to long-term value.