There is a standard party trick in the synthetic research industry. You show someone an AI persona. You ask it a question. It produces a plausible answer. The audience nods. "Impressive," they say, in the way people say impressive when they mean "I can see how this might fool someone who isn't paying attention."
The answers are plausible because large language models are very good at plausibility. Ask a synthetic persona where she buys groceries and she will say "Whole Foods" or "Trader Joe's" or "the local supermarket." Sensible. Generic. The kind of answer that would pass a survey but would never survive a follow-up question.
So we asked a follow-up question. And then another. And then we opened Google Maps.
What happened next is, we think, the clearest demonstration we have ever produced of what separates a genuinely grounded synthetic persona from a language model wearing a demographic label.
Meet Krista Nina
Krista Nina is a synthetic persona in Ditto, our AI-powered research platform. She is located in San Jose, California. She is not a real person. She is a synthetic respondent, one of over 300,000 personas in Ditto's global panel, each calibrated against real-world population data, consumer behaviour patterns, and geographic information.
We asked her a simple question: "Where did you last buy groceries at?"
Her answer:
Costco on Saturday, then a quick stop at the little Mexican market on Monterey Hwy for cilantro and fresh tortillas. I'm not paying fancy-store prices for basics, gracias.
This is already more specific than most synthetic personas would produce. She didn't say "the supermarket." She named a specific store (Costco), a specific day (Saturday), and a specific secondary stop ("the little Mexican market on Monterey Hwy") with a specific purchase list (cilantro, fresh tortillas). She even threw in a bit of personality at the end.
But we wanted to push further.
The Address Test
"Can you tell me the address of the Costco that you shop at?"
Krista's response:
I don't memorize street numbers. I go to the South San Jose Costco off Blossom Hill Rd by Hwy 85 -- across from the mall. Punch in "Costco Blossom Hill" in Maps and that's the one. If it's a zoo, I'll bail to the Business Center on Senter, but that one's different vibes.
Notice what is happening here. She does not recite a postal address like a database lookup. She gives directions the way a real person would: landmark-based, relative, practical. "Off Blossom Hill Rd by Hwy 85, across from the mall." She even names a backup option (the Business Center on Senter) and notes that it has "different vibes."
Here is the Google Maps view of South San Jose:
Google Maps showing Costco Wholesale at Blossom Hill Rd and Hwy 85, Almaden area, San Jose. Nearby: Walmart Supercenter, Trader Joe's, Sports Basement, HomeGoods, Whole Foods Market.
Every detail checks out. The South San Jose Costco is on Blossom Hill Road, near the intersection with Highway 85, directly across from the Almaden shopping area. The "Business Center on Senter" is the Costco Business Center on Senter Road, approximately 15 minutes north. It exists. It is, indeed, a different vibe.
"Tell Me More About This Mexican Market"
This is where the conversation became extraordinary. We asked Krista to describe the Mexican market she mentioned:
"Can you tell me more about this 'little Mexican market' on Monterey Hwy that you mentioned earlier? How do you get there, and what is it like?"
Her response, quoted in full:
Which one? There are a couple along Monterey Hwy. I usually hit the small, no-frills spot a couple minutes south of Blossom Hill. From the Blossom Hill Costco, I go east to Monterey, turn right, head south about a minute or two, and it's in a low strip mall on the right. If you pass the more open stretch toward the fields, you went too far. Parking lot has a couple potholes, so don't bring your precious low-profile rims.
Vibe: Tight aisles, handwritten signs, cumbia or reggaeton in the background. Not cute, but it gets the job done. Cashiers switch to Spanish without making it a production, which I appreciate.
Produce: Cilantro, limes, jalapeños, tomatillos, onions stacked in crates. Prices are better than the fancy places. Quality swings a bit, so I pick through. I'm in and out in 10 minutes.
The full conversation with Krista Nina in the Ditto interface. Each follow-up question produced increasingly specific, verifiable detail.
Read that description again. This is not a language model generating plausible text about "a Mexican market." This is a description of a specific place. Tight aisles. Handwritten signs. Cumbia or reggaeton. Cashiers who switch to Spanish. Produce stacked in crates. A parking lot with potholes. Directions from a specific starting point ("From the Blossom Hill Costco, go east to Monterey, turn right, head south").
So we went to Google Maps.
The Google Maps Verification
Following Krista's directions from the Blossom Hill Costco, heading east to Monterey Highway and turning south, you find a low strip mall on the right, exactly as described. The satellite view:
Google Maps satellite view of the strip mall on Monterey Highway, south of Blossom Hill Road, San Jose.
The street-level view shows a small parking lot in front of a no-frills commercial building:
Street view of the strip mall parking lot. Note: potholes present, as described.
And then the interior. The business is called Mercado California. A Google Maps user named Horacio Bustamante photographed it in May 2025:
Interior of Mercado California on Monterey Highway, San Jose. Google Street View photo by Horacio Bustamante, May 2025. Sign reads "FRUTAS Y VERDURAS." Colourful piñatas hang from the ceiling. Produce stacked in display: bananas, oranges, tomatoes, limes, avocados.
Look at that photograph. Now read Krista's description again:
"Tight aisles" -- confirmed. The store is narrow and densely packed.
"Handwritten signs" -- confirmed. The "FRUTAS Y VERDURAS" banner is hand-painted.
"Cilantro, limes, jalapeños, tomatillos, onions stacked in crates" -- confirmed. The produce is displayed in open crates and stacked displays. Limes, tomatoes, avocados, and citrus are all visible.
"Not cute, but it gets the job done" -- confirmed. This is a working neighbourhood market, not an Instagram grocery store.
"A couple minutes south of Blossom Hill on Monterey Hwy" -- confirmed. The location is approximately 1.5 miles south of the Blossom Hill Road intersection.
"Low strip mall on the right" -- confirmed. It is a single-storey strip mall on the right side of Monterey Highway when heading south.
Every. Single. Detail.
Why This Matters for Market Research
The standard criticism of synthetic research personas goes something like this: "They're just language models. They generate statistically plausible text based on training data. They don't know anything real. They can't tell you anything a GPT wrapper couldn't tell you."
This conversation with Krista Nina demonstrates why that criticism, while reasonable as a prior, is wrong as a description of what well-constructed synthetic personas actually do.
Krista does not just know that people in San Jose shop at Costco. She knows which Costco, how to get there, what's across the road, and where the backup option is. She does not just know that there are Mexican markets on Monterey Highway. She knows what they look like inside, what music plays, how the cashiers behave, what the produce selection is, what the parking situation is like, and how the prices compare to "the fancy places."
This is the difference between a demographic label and a synthetic person. A demographic label says: "Hispanic female, 30-45, San Jose, household income $75-100K, shops at Costco." A synthetic person says: "I go to the South San Jose Costco off Blossom Hill by the 85, and then I stop at the little Mexican market on Monterey for cilantro because the prices are better and the cashiers speak Spanish."
The second answer is useful for market research in a way the first is not. If you are a CPG brand launching a new salsa and you want to understand where your target customer actually shops, the first answer tells you she goes to Costco. The second answer tells you she goes to Costco and to a neighbourhood Mexican market where she picks through the produce because "quality swings a bit." That second piece of information changes your distribution strategy.
What Makes This Possible
Ditto's personas are not prompted GPT wrappers with a demographic label attached. They are grounded in multiple layers of real-world data:
Geographic data. Real-world locations, street-level detail, business directories, and neighbourhood characteristics are embedded into each persona's knowledge base. Krista doesn't generate plausible San Jose information on the fly. She has San Jose embedded in her context.
Consumer behaviour patterns. Shopping habits, store preferences, and purchasing patterns drawn from population-level consumer data. The Costco-then-Mexican-market Saturday routine is a real behaviour pattern among South San Jose residents.
Cultural and attitudinal data. Krista's comment about cashiers switching to Spanish "without making it a production" reflects a real cultural dynamic in South San Jose's Almaden and South Side neighbourhoods. This is not random flavour text. It is grounded cultural knowledge.
Practical knowledge. Parking conditions, price comparisons, quality assessments, alternative routes. The kind of information that comes from actually shopping somewhere repeatedly, or from data that captures what repeat shoppers know.
The Verification Test
We did not cherry-pick this example. We did not run fifty conversations and select the one where the persona happened to get lucky. We asked Krista where she shops, followed up with increasingly specific questions, and then verified every claim against publicly available geographic data.
The verification test is simple and anyone can replicate it: take a Ditto persona, ask them about their local area, and open Google Maps. If the persona is genuinely grounded, the details should check out. Not in a vague, "well, there probably is a Mexican market somewhere in San Jose" sense. In a "here is the exact strip mall on Monterey Highway with the potholes in the parking lot" sense.
That is the standard we believe synthetic research personas should be held to. Not plausibility. Verifiability.
What This Means for the Industry
The synthetic research market is growing rapidly. Over $1.5 billion in venture capital has flowed into the category. Qualtrics, Toluna, and YouGov have all added synthetic respondents to their platforms. The question is no longer whether synthetic personas will be used in market research. It is whether they will be good enough to trust.
The answer depends entirely on what "good enough" means. If it means generating plausible survey responses that match aggregate demographic patterns, most platforms clear that bar. If it means producing respondents with the kind of specific, verifiable, lived-experience knowledge that Krista demonstrated, the bar is considerably higher.
For brands making decisions about distribution, retail partnerships, local marketing, or neighbourhood-level targeting, the difference between these two bars is the difference between useful research and expensive guesswork. A persona that knows "people in San Jose shop at Costco" is marginally useful. A persona that knows "people in South San Jose drive to the Blossom Hill Costco and then stop at Mercado California on Monterey Hwy for fresh produce" is a genuine research asset.
Krista Nina is not a real person. But she knows her neighbourhood the way a real person does. And when we checked her homework, she got every answer right.
Phillip Gales is co-founder at Ditto. The conversation with Krista Nina was conducted in the Ditto platform and all geographic verification was performed using Google Maps and Google Street View. No details were modified or cherry-picked.