How a Toothpaste Company Quietly Killed the Focus Group: The Dawn of Synthetic AI Consumers

If you want to see the future of business, you do not always look at Silicon Valley software releases. Sometimes, you look at toothpaste.

A quiet, absolutely earth-shattering disruption just occurred in the market research industry, and almost nobody is talking about it yet.

Colgate, in collaboration with PyMC Labs, recently published a research paper with a dense academic title: LLMs Reproduce Human Purchase Intent via Semantic Similarity Elicitation of Likert Ratings. Behind that mouthful is a terrifying reality for traditional research agencies, brand strategists, and anyone who has ever sat in a focus group watching a one-way mirror.

We no longer need to wait months to know what consumers want. We can just ask the AI.

And when I say "ask," I do not mean the clumsy way most people have been trying to use AI for market research. I mean a specific, elegant hack that turned a language model into a synthetic consumer so accurate it predicted real-world purchase intent with 90% fidelity.

Here is how a toothpaste company cloned the consumer mindset, why the method matters more than the headline, and what it means for the timeline of every product launch from here on out.

The Problem: AI Is a Terrible Survey Taker

Let us start with the failure point everyone has encountered.

If you take a standard Large Language Model and prompt it with, "Please rate this new product concept on a scale of 1 to 5," the results are useless. The AI will spit out a safe, middle-of-the-road, diplomatic response. It will hedge. It will qualify. It will give you a 3.5 with a paragraph of caveats that holds absolutely zero predictive value for real-world sales.

This is why many business leaders—smart ones, experienced ones—mistakenly concluded that AI could never replace the nuanced, emotional, irrational decision-making of a human focus group. They tried the obvious approach, got garbage, and wrote the whole idea off.

They were wrong. They just were not asking the machine the right way.

The problem is not that AI cannot think like a consumer. The problem is that you cannot ask a consumer to rate something on a numbered scale and expect honesty either. Real humans do not walk around with Likert scales in their heads. They have gut reactions, contradictions, memories, biases, and stories. The number is just the compressed output of a much richer internal narrative.

Colgate's researchers understood this. So they stopped asking for numbers and started asking for the story instead.

The Hack: Roleplay and Semantic Similarity Rating (SSR)

The researchers at Colgate and PyMC Labs realized that to get human-like data, you have to bypass the AI's logical guardrails entirely. You cannot ask an LLM to be analytical when real consumers are emotional. You cannot ask it to be objective when real consumers are subjective.

They invented a method called Semantic Similarity Rating—SSR.

Here is how it works, step by step, because the details matter:

Step 1: Persona Injection. Instead of treating the AI as a neutral judge, they forced it into a highly specific roleplay. They fed the LLM a detailed demographic profile—age, income bracket, geographic location, lifestyle habits, brand loyalties, even anxieties and aspirations. They did not ask the AI to simulate a consumer. They asked the AI to become one.

Step 2: Visceral Elicitation. They showed the AI a product concept—not as a data sheet, but as a real consumer would encounter it. Then they asked for raw, unfiltered, diary-like reactions. Not ratings. Not pros and cons lists. They asked: "What is your immediate gut reaction? What does this remind you of? What would you tell your friend about this in a text message?"

The AI produced narrative entries. Consumer diary fragments. Emotional reactions. Irrational associations. The kind of messy, qualitative texture that traditional surveys usually sterilize out of existence.

Step 3: Semantic Translation. Here is the elegant part. The researchers did not manually read thousands of AI diary entries and guess at scores. They used a secondary semantic model to translate those qualitative, text-based thoughts into hard numerical scores. The semantic model measured how closely the AI's language patterns aligned with the language patterns of real consumers who had previously rated products in known ways.

In other words, they did not ask the AI to rate the product. They asked the AI to think like a human, then used another AI to measure how human-like that thinking was.

The Staggering Results: 90% Accuracy Against 9,300 Real Humans

When the team compared the AI's synthetic responses to 57 actual corporate surveys encompassing 9,300 real human responses, the results were chilling.

The synthetic AI consumers matched real human purchase intent with 90% accuracy.

They did not just guess right on average. They captured the nuanced economic psychology of the human market. The AI personas accurately replicated how different age groups and income brackets would react to sudden price changes—price elasticity, in economist terms. They mirrored the irrational aversion to certain price points. They replicated the way luxury buyers shrug at a 20% markup while budget buyers abandon the cart.

Even more incredibly, the qualitative feedback generated by the AI was found to be deeper, more thorough, and far more critical than the feedback provided by actual humans.

Think about why. Real humans rush through paid surveys. They are distracted. They are thinking about lunch. They give the minimum viable answer to collect their incentive and move on. The AI, given a persona and asked to think deeply, has infinite patience. It does not get bored. It does not check its phone. It produces the kind of thorough, critical analysis that only your most engaged customers would give you—if you could find enough of them and pay them to care.

Why This Changes Everything: The Death of the Focus Group Industrial Complex

Traditional market research is slow, expensive, and structurally biased.

A typical product validation cycle looks like this: hire an agency, recruit participants, schedule sessions, run focus groups, transcribe the recordings, analyze the themes, compile a report, present findings to stakeholders, debate the implications, and finally—six to eight weeks later—make a decision. The cost runs from tens of thousands to hundreds of thousands of dollars. And by the time you act, the market may have shifted.

The Colgate paper blows that timeline apart.

Imagine simulating 1,000 highly targeted, demographically accurate customer interviews overnight. Not generic surveys. Deep, persona-driven, narrative-rich explorations of how your product lands in the mind of a 34-year-old mother in suburban Houston, a 58-year-old retiree in Miami, a 22-year-old student in Seoul.

Imagine running infinite A/B tests on pricing strategies before writing a single line of production code. You could test twenty price points across five demographic segments in a day, see exactly where demand drops off a cliff, and know your optimal pricing architecture before you even manufacture the product.

Imagine validating a brand name, a packaging design, or a feature set not with a roomful of strangers eating free sandwiches behind a one-way mirror, but with a synthetic panel that thinks like your exact target customer and tells you the unvarnished truth in minutes.

This is not a marginal improvement. This is an existential threat to the traditional market research industry. And it is an existential opportunity for the brands that adopt it first.

The Mercury Play: From Insight to Action in One Breath

At Mercury Technology Solution, our core ethos is to Accelerate Digitality. We empower brands to improve operations, elevate marketing, and boost efficiency through the strategic implementation of cutting-edge technology.

This Colgate paper is the ultimate manifestation of that ethos. It does not just shatter the cost structure of market research. It collapses the distance between knowing and doing.

Think about the operational leverage when synthetic consumer intelligence plugs directly into your business systems:

Feed the insights into your ERP. The Mercury Business Operation Suite can immediately adjust purchase management strategies to maintain optimal stock levels based on predicted demand. If the synthetic research shows that a price drop at a specific threshold triggers a 40% demand spike, your procurement system can pre-position inventory before the campaign even launches.

Inform your sales pipeline. Synthetic sales forecasts can shape real-world pipeline priorities. If the AI consumers in a specific demographic show resistance to a feature, your sales team knows to lead with a different value proposition for that segment.

Trigger content generation automatically. Agents like Mercury Muses AI can take the exact psychological triggers uncovered by synthetic research and generate high-quality, perfectly targeted blog content, email campaigns, and social copy—tailored to the specific language patterns, anxieties, and aspirations of each persona.

The loop becomes: synthesize, validate, execute, measure, iterate. In hours, not quarters.

The Deeper Truth: Speed Is the Only Moat Left

We are entering an era where the companies that learn to distill human behavior into AI models will iterate faster, launch cheaper, and pivot instantly.

The traditional moats—scale, distribution, brand awareness—are still important. But they are no longer sufficient. A competitor with a faster feedback loop will outlearn you, outmaneuver you, and eat your lunch before your quarterly review even lands on the calendar.

The era of guessing what the customer wants is over. The era of knowing—with mathematical confidence, overnight—is here.

But here is the catch. The tool is available to everyone. The LLM that Colgate used is not proprietary. The SSR method is publishable. Your competitors can read the same paper you just read.

So the competitive advantage is not in having access to synthetic consumers. It is in what you ask them, how you interpret the answers, and how fast you act on what you learn.

That is where the human element survives. The strategist who knows which questions matter. The operator who knows how to reconfigure the supply chain when the data demands it. The creative who knows how to turn a synthetic insight into a story that moves real humans.

The AI gives you the map. You still have to decide where to go.

Stay ahead of the curve.

— James

Frequently Asked Questions

What is the Colgate synthetic consumer research?A 2026 research paper published by Colgate in collaboration with PyMC Labs demonstrating that Large Language Models can act as synthetic consumers and predict real-world purchase intent with 90% accuracy when using a method called Semantic Similarity Rating (SSR).

What is Semantic Similarity Rating (SSR)?A research methodology that bypasses direct numerical scoring by forcing an LLM into a detailed demographic persona, eliciting raw narrative reactions to a product, then using a secondary semantic model to translate those qualitative text responses into predictive numerical scores based on alignment with real human language patterns.

Why does simply asking AI to rate products fail?Standard LLMs, when asked for numerical ratings, produce safe, middle-of-the-road, diplomatic responses that lack predictive value. Real consumers do not think in Likert scales—they think in stories, emotions, and associations. SSR captures that qualitative richness and translates it mathematically.

How accurate was the AI compared to real consumers?The synthetic AI consumers matched real human purchase intent with 90% accuracy when tested against 57 corporate surveys and 9,300 human responses. The AI also replicated complex economic behaviors like price elasticity and produced deeper, more critical qualitative feedback than human survey participants.

What is a synthetic consumer?An AI model programmed with a specific demographic persona to simulate the attitudes, preferences, emotional reactions, and decision-making patterns of a real consumer segment. Unlike traditional market research simulations, synthetic consumers generate narrative, qualitative responses that mirror human psychology.

Why is this an existential threat to traditional market research?Because it collapses the timeline and cost structure of consumer insight. What previously required months, tens of thousands of dollars, and physical focus groups can now be simulated overnight with 1,000 demographically precise personas. Speed and scale advantages will accrue to brands that adopt synthetic research first.

How can businesses use synthetic consumer insights operationally?By integrating synthetic research directly into ERP systems for demand forecasting, into sales pipelines for segment-specific strategy, and into AI content agents for psychologically targeted marketing. The goal is a closed loop where insight becomes action in hours rather than quarters.

What is the difference between synthetic consumers and traditional AI surveys?Traditional AI surveys ask models for direct ratings or opinions, yielding generic, unhelpful data. Synthetic consumers use persona-driven roleplay and narrative elicitation to generate emotionally realistic, qualitatively rich responses that can be semantically mapped to real human behavior patterns.

Will synthetic consumers replace human focus groups entirely?Not immediately for all use cases. Physical product testing, sensory experience evaluation, and emergent cultural trend detection still benefit from real human interaction. But for digital products, pricing strategy, messaging validation, and concept testing, synthetic research is already faster, cheaper, and in some cases more insightful than human panels.

What is the competitive advantage if everyone can use synthetic consumers?The tool is democratized, but the advantage lies in strategic application: knowing which questions to ask, how to interpret nuanced responses, and how rapidly to operationalize insights across product, pricing, and marketing. The winners will be the organizations with the tightest insight-to-action loops.