These Respondents Have Never Bought Anything—And They’re Here to Judge Your Product
By David Schneer
7-Minute Read
You just completed a study among hundreds of prospective customers to understand their needs and reactions to your product. You’re super excited because the results took only 3 minutes to render for under $100. That’s because you used synthetic sample. That’s right, sample that exists only in algorithmic form. None of the information you collected is from a sentient being—just an advanced formula designed to mimic one.
Let’s meet your synthetic respondents. Your sample composition was evenly split by gender. Let’s start with the profile of one such respondent.
We’ll name her Synthetic Sally. Since Sally can’t emote, she’ll pick her favorite commercial, but she won’t know why because it wasn’t her choice—it was some random human’s choice. Neither does Sally get annoyed by advertising wear-out, due to excessive repetition. To her, “Act Now!” gives her something to do. She can’t smell or taste coffee, but she is still qualified to evaluate your espresso campaign. Based on a similar approach in a toothpaste ad, Sally voted the espresso campaign a winner. In the space provided for an open-ended comment, she wrote: “Satisfactory emotional resonance detected.” Sounds so human, doesn’t it? You see, Sally lives in a digital world where no one ever skips an ad or is distracted by kids, a text, or a phone call. And the beauty of it all is that Sally’s synthesized opinion is fast and doesn’t require a Starbucks gift card. Oh, the irony of it.
And the Men. They are a wealth of data just like their female counterparts. Consider this.
Meet Byte Bob. He has never stepped inside a Target but can serve up all the metadata about the chain. Like Synthetic Sarah, Byte Bob doesn’t eat cereal. However, if he did, he’d likely prefer Cap’n Crunch’s Cinnamon Crunch based on the purchase behavior of 55,000 other consumers with similar IP addresses. Bob was also able to help price your product at $5.89, even though he’s never had to buy anything. Nevertheless, while Bob can’t feel excitement, doubt, guilt, or loyalty—he can tell you the average click-through rate of banner ads for dish soap.
Bob is ready to evaluate your new concept…if it doesn’t require buying it because Bob doesn’t have a heartbeat or a VISA card.
Okay, let’s turn off the sarcasm and inject a little logic with a point-counterpoint didactic approach regarding synthetic sample.
POINT
Proponents of synthetic sample are selling fast, cheap, and data-rich respondents.
COUNTERPOINT
Synthetic sample is “cheap” because you’re not paying for human insight—you’re licensing an algorithm’s output. These aren’t buyers; they’re data proxies with no purchasing power, emotions, or context. This means you’re optimizing for irrelevance by asking the opinions of people who will never purchase your product or service. If your goal is real-world relevance, synthetic respondents are noise masquerading as signal. You’re not just cutting costs—you’re cutting the cord to your actual customer.
POINT
Proponents of synthetic sample indicate that while respondents do not actually buy products, they can mimic the buying behavior of others and deduce intelligence from that.
COUNTERPOINT
And while they may approximate demographics or psychographics, synthetic respondents don’t navigate real-life tradeoffs like budgeting, brand loyalty, peer pressure, or convenience—all of which shape real purchase behavior. You’re not gathering feedback from consumers; you’re getting recycled assumptions from yesterday’s data.
While synthetic respondents can imitate shopping behavior, they don’t understand why. Mimicking behavior is not the same as understanding behavior. Human purchasing decisions are often emotional, irrational, and influenced by personal context. Synthetic respondents may be able to replicate surface-level patterns, but they lack the conscious awareness to explain why a message resonates, a product delights, or a price point causes sticker shock.
More critically, synthetic models are trained on past behaviors and established categories. They falter when tested against new products or unfamiliar messaging—precisely where the richest insights are needed. Predicting tomorrow’s success using yesterday’s habits is a strategic risk, not a shortcut.
Real buyers make trade-offs. They have budgets, doubts, loyalty quirks, and hesitation when spending their own money. How can Synthetic respondents replicate the tension and scrutiny that affects real-world purchasing?
POINT
Yes, synthetic sample is fast, and I’ve never heard a client say, “take your time.” So, no argument here.
COUNTERPOINT
But if you’re launching a multimillion-dollar product based on inauthentic feedback, fast could come at the cost of a market flop. Are your communications or products targeted toward humans or bots? Speed is seductive—but only if the insights are sound. Launching a product based on fast but fabricated feedback is like building a house on quicksand. Fast doesn’t equal factual. If you’re testing messaging, pricing, or innovation with a sample that has no capacity to feel, decide, or buy—then what, exactly, are you validating? Synthetic sample might deliver results in minutes, but you’ll pay for it later in missed cues, bad launches, or wasted budgets.
CONCLUSION
Every piece of marketing, branding, and product development hinges on this: Will it move people to purchase? If your respondents don’t buy—or can’t buy—you’re flying blind.
We Talk to the People Who Swipe the Card
At Merrill Research, when we conduct strategic communications or new product research, we target real buyers. These are the people who influence market outcomes, not just click through surveys. You wouldn’t launch a campaign without audience targeting— why accept research sample that ignores it? Real respondents. Real purchases. Real decisions.
In new product development research, the gold nuggets often lie in surprising contradictions—unexpected feedback, odd use cases, and emotional pushback. A synthetic model can’t simulate that unpredictability. It smooths it out.
Marketers don’t launch to bots—they launch to real people with messy lives, imperfect logic, and emotional needs. You need that richness in your research sample if you want results that resonate.
They can’t tell you how a message makes them feel, only how it might perform statistically.