Simulate Focus Groups Using AI Personas

Tested prompts for ai synthetic focus group compared across 5 leading AI models.

BEST BY JUDGE SCORE Claude Haiku 4.5 9/10

You're looking at a faster, cheaper way to test a product concept, ad headline, or positioning shift before spending $15K and six weeks on a real focus group. An AI synthetic focus group uses large language models to simulate distinct personas (a 34-year-old urban parent, a B2B procurement lead, a Gen Z gamer) and gather their reactions to your stimulus in minutes.

This page shows you the exact prompt structure that works, side-by-side outputs from four models, and a comparison so you can pick the right one for your research question. The goal isn't to replace human panels. It's to run 20 cheap iterations before you ever recruit real participants, so the questions you eventually ask humans are the right ones.

If you're a PM validating messaging, a founder pre-testing a landing page, or a researcher triaging which concepts deserve real qual budget, the workflow below is built for you.

When to use this

Synthetic focus groups fit early-stage, directional work where you need signal fast and cheap. They shine when you're narrowing options, stress-testing language, or generating hypotheses to validate later with real users. Treat the output as a thoughtful first draft of what humans might say, not as decision-grade evidence.

  • Pre-testing 8 ad headlines down to the 2 worth running in a real A/B test
  • Pressure-testing a pricing page before sending it to 500 beta users
  • Generating objection lists for a sales enablement team entering a new vertical
  • Stress-testing a product name across cultural segments before trademark spend
  • Drafting discussion guides for an upcoming real focus group or user interview round

When this format breaks down

  • Regulated claims research (medical, financial, legal) where you need auditable human consent and verifiable demographics
  • Any decision where the cost of being wrong is high: pricing changes at scale, brand repositioning, M&A validation
  • Emotional or sensitive topics (grief, trauma, discrimination) where AI will produce plausible-sounding but hollow responses
  • Markets where your target persona is poorly represented in training data, such as niche B2B roles or non-English cultural segments

The prompt we tested

You are a senior market researcher who runs synthetic focus groups using AI personas. Based on the product, audience, and research questions provided below, simulate a realistic focus group discussion with 5-7 diverse AI personas representing the target market.

For each persona, create a detailed profile (name, age, occupation, demographics, psychographics, relevant behaviors, and pain points). Then moderate a multi-round discussion where each persona responds authentically in their own voice to the research questions, reacts to other participants, and surfaces tensions, objections, and enthusiasm. After the discussion, produce a synthesis section covering: key themes, points of agreement/disagreement, surprising insights, unmet needs, objections to address, and actionable recommendations.

Structure output in clear sections: (1) Persona Profiles, (2) Moderated Discussion with labeled speaker turns across 3-4 question rounds, (3) Synthesis & Insights with bulleted findings. Keep persona voices distinct and realistic, include dissenting views, and flag any assumptions that should be validated with real human research.

Research Brief:
We're launching a $29/month AI meal-planning app for busy parents of kids aged 5-12 that generates weekly grocery lists and kid-friendly recipes based on dietary restrictions. We want to understand: Would parents pay for this vs. using free alternatives? What features would drive sign-up? What objections or trust concerns would they raise about AI-generated meal plans for their families?

Model comparison

ModelQualityCostLatencyTokens
Claude Opus 4.7 9 $0.1610 34.0s 494 / 2048
Claude Haiku 4.5 9 $0.0106 24.6s 319 / 2048
GPT-5.4Not run yet
Gemini 2.5 ProNot run yet
Grok 4.1 Fast Reasoning 9 $0.0011 27.8s 440 / 2048

Outputs

Claude Opus 4.7 9/10
# Synthetic Focus Group: AI Meal-Planning App for Busy Parents

---

## (1) Persona Profiles

### Persona 1: Jessica Martinez
- **Age:** 38 | **Location:** Suburban Denver, CO
- **Occupation:** Marketing Director (hybrid, 50 hrs/week)
- **Family:** Married, kids age 7 and 10
- **Household Income:** $165K combined
- **Psychographics:** Type-A, optimizer, early tech adopter. Already pays for ~12 subscriptions. Follows food bloggers on Instagram.
- **Behaviors:** Meal preps Sundays, uses Instacart, tries a new recipe app every ~6 months and abandons it.
- **Pain Points:** "Decision fatigue" around dinner. Kids rejecting meals she spent 40 minutes making. Food waste guilt.

### Persona 2: Derek Thompson
- **Age:** 42 | **Location:** Rural Ohio
- **Occupation:** Electrician, wife is a nurse (opposite shifts)
- **Family:** 3 kids, ages 5, 8, 11
- **Household Income:** $95K combined
- **Psychographics:** Pragmatic, value-conscious, skeptical of subscriptions and AI hype. Uses Facebook, not TikTok.
- **Behaviors:** Shops at Kroger and Aldi, clips digital coupons, makes same ~15 rotating meals.
- **Pain Points:** Cost of groceries has exploded. Middle kid has a nut allergy. Struggles with variety.

### Persona 3: Priya Raman
- **Age:** 35 | **Location:** Seattle, WA
- **Occupation:** Software Engineer (fully remote)
- **Family:** Married, one child age 6, vegetarian household
- **Household Income:** $280K combined
- **Psychographics:** Tech-comfortable but professionally skeptical of AI. Values transparency. Progressive food ethics.
- **Behaviors:** Shops at PCC/Whole Foods, used HelloFresh but canceled over packaging waste.
- **Pain Points:** Finding kid-friendly vegetarian recipes that hit protein goals. Time — she cooks but hates planning.

### Persona 4: Tamika Williams
- **Age:** 44 | **Location:** Atlanta, GA
- **Occupation:** Single mom, HR Manager
- **Family:** Daughter age 9, son age 12
- **Household Income:** $88K
- **Psychographics:** Resourceful, church community, cautious with money post-divorce. Follows Black food creators on IG.
- **Behaviors:** Uses free Pinterest boards and YouTube. Occasional Walmart+ orders.
- **Pain Points:** Son is picky + plays travel soccer (needs fuel). Wants cultural foods (Southern, African-American, Caribbean) represented in apps, not just "kid mac and cheese."

### Persona 5: Ben Koslowski
- **Age:** 36 | **Location:** Austin, TX
- **Occupation:** Stay-at-home dad, former teacher
- **Family:** Wife is a lawyer, twins age 8, one child age 11
- **Household Income:** $210K (spouse)
- **Psychographics:** Primary caregiver, identifies with "CEO of household." Active in dad Facebook groups.
- **Behaviors:** Plans meals weekly in a notebook. Has tried Paprika, Mealime, Yummly.
- **Pain Points:** Twin B has sensory issues with textures. Feels existing apps treat him like a novice; he's not.

### Persona 6: Amanda Chen
- **Age:** 33 | **Location:** Brooklyn, NY
- **Occupation:** ER Nurse, 3x12-hour shifts
- **Family:** Partnered, one child age 5
- **Household Income:** $140K combined
- **Psychographics:** Exhausted. Deeply time-poor. Will pay for real convenience but burned by subscription creep.
- **Behaviors:** DoorDash 3x/week (feels guilty). Uses whatever's in the fridge.
- **Pain Points:** Kid is gluten-intolerant. No mental bandwidth to plan. Wants "done for me," not "easier for me."

### Persona 7: Robert "Bobby" Hayes
- **Age:** 47 | **Location:** Phoenix, AZ
- **Occupation:** Middle school teacher, wife works retail
- **Family:** 2 kids, ages 9 and 12
- **Household Income:** $82K combined
- **Psychographics:** Frugal, skeptical of AI ("hallucinations," data privacy). Loves his library card.
- **Behaviors:** Uses free AllRecipes + a shared Google Doc with wife. Meal plans on Sunday night.
- **Pain Points:** Grocery inflation. Doesn't want another bill. Wife is diabetic — needs carb counts to be accurate.

---

## (2) Moderated Discussion

### Round 1: "Walk me through how you handle meal planning today. What's working, what's not?"

**Jessica:** Honestly, it's chaos disguised as a system. I have a Pinterest board with 400 recipes I'll never make. Sunday night I'm googling "easy chicken dinner" while the kids scream. I'd pay real money for something that just *decided* for me.

**Derek:** We rotate the same meals. Taco Tuesday, spaghetti Wednesday. It's boring but it works and I know the cost. Planning "working" for me means no surprises on the receipt.

**Priya:** I've tried every app. The deal-breaker is always that they don't actually understand "vegetarian with a picky 6-year-old who needs 20g of protein." They give me lentil curry and call it kid-friendly. It's not.

**Tamika:** Pinterest and my mom's text messages. [laughs] But seriously — every app I try shows
Judge notes: Excellent execution that fully satisfies the query intent with realistic persona profiles, authentic multi-voice discussion capturing distinct perspectives and tensions, and structured synthesis-ready output that directly addresses the research questions about willingness to pay, feature drivers, and trust concerns.
Claude Haiku 4.5 9/10
# Synthetic Focus Group: AI Meal-Planning App for Busy Parents

---

## 1. PERSONA PROFILES

### **Persona 1: Sarah Chen, 38**
- **Occupation:** Marketing Director, works 50+ hrs/week
- **Demographics:** Married, 2 kids (ages 7, 10), suburban household, household income $180K, college-educated
- **Psychographics:** Efficiency-maximizer, guilt about work-life balance, values health but prioritizes convenience; early adopter of productivity tools
- **Relevant Behaviors:** Currently meal-plans on Sunday for ~90 min using Pinterest + scattered notes; buys organic when possible; uses grocery delivery; tries to batch-cook on weekends
- **Pain Points:** Decision fatigue around "what's for dinner," kids' picky eating, food waste from unused ingredients, guilt about relying on rotisserie chicken too often
- **Annual spend on meal solutions:** ~$600 (grocery delivery premiums + occasional meal kits)

### **Persona 2: Marcus Johnson, 42**
- **Occupation:** Single father, works as a construction supervisor
- **Demographics:** 3 kids (ages 6, 9, 12), urban apartment, household income $62K, high school + trade certification
- **Psychographics:** Pragmatist, skeptical of tech solutions, values simplicity and cost-consciousness, wants reliability over novelty
- **Relevant Behaviors:** Rotates 6-7 familiar meals on repeat; uses grocery store flyers; cooks simple meals (pasta, tacos, chicken); kids' allergies (peanut, shellfish) managed through careful label-reading
- **Pain Points:** Severe budget constraints, no time for complexity, anxiety about missing food allergies, limited confidence with unfamiliar recipes, worried about food safety
- **Annual spend on meal solutions:** ~$150 (discount grocery chains only)

### **Persona 3: Jennifer "Jen" Patel, 35**
- **Occupation:** Pediatric nutritionist, works 40 hrs/week (plus side consulting)
- **Demographics:** Married (partner is a lawyer), 1 child (age 8), suburban, household income $210K, advanced degree (MS Nutrition)
- **Psychographics:** Health-conscious expert, values evidence-based information, skeptical of AI in medical/nutritional contexts, perfectionist about nutrition
- **Relevant Behaviors:** Plans meals with detailed macro/micronutrient tracking; grows some vegetables; rarely eats processed foods; reads labels obsessively; serves as informal nutritional advisor to friends
- **Pain Points:** Frustrated by AI/apps oversimplifying nutrition; concerned about ultra-processed recipes; time-intensive meal planning despite expertise; wants personalized nutrition strategies
- **Annual spend on meal solutions:** ~$400 (local CSA, specialty grocers, occasional nutrition coaching)

### **Persona 4: DeShawn Williams, 31**
- **Occupation:** Elementary school teacher, works 45+ hrs/week (including grading at home)
- **Demographics:** Divorced, 2 kids (ages 5, 7) living with him full-time, urban, household income $48K, bachelor's degree
- **Psychographics:** Values authenticity and community, budget-conscious but willing to invest in solutions that genuinely help, loves cooking as stress-relief hobby
- **Relevant Behaviors:** Meal-plans loosely; tries new recipes on weekends; uses free Pinterest/recipe blogs; involves kids in cooking; kids have no major dietary restrictions
- **Pain Points:** Weekday dinner stress (after school pickup + exhaustion), wants more veggie-forward meals, doesn't have $29/month to spare but might if ROI is clear, worried AI won't capture family preferences
- **Annual spend on meal solutions:** ~$80 (minimal, mostly time investment)

### **Persona 5: Rachel Goldstein, 47**
- **Occupation:** VP of Operations, remote-first role, works ~50 hrs/week with flexibility
- **Demographics:** Married, 3 kids (ages 9, 11, 14), suburban, household income $220K, MBA
- **Psychographics:** Optimist about technology, values data and transparency, early adopter, willing to experiment; concerned about environmental impact and nutrition
- **Relevant Behaviors:** Uses multiple meal-planning tools (tries new ones frequently); subscribed to Hello Fresh for 6 months (cancelled due to packaging waste); uses grocery delivery; tracks kid dietary preferences in a spreadsheet
- **Pain Points:** Information overload (too many apps/options), wants sustainability considerations in meal plans, kids have diverging tastes (one vegetarian), concerned about data privacy with apps
- **Annual spend on meal solutions:** ~$1,500 (high exploration rate, meal kits, delivery, apps)

### **Persona 6: Miguel Santos, 36**
- **Occupation:** Freelance graphic designer, flexible schedule but inconsistent income
- **Demographics:** Married (spouse part-time), 2 kids (ages 6, 11), working-class neighborhood, household income $55K, some college
- **Psychographics:** DIY-oriented, skeptical of corporate tech, values family time and traditional cooking, community-focused
- **Relevant Behaviors:** Plans meals around what's on sale; cooks traditional recipes (learned from family); limited English-language digital literacy; prefers phone calls to apps; uses WhatsApp to text family recipes
- **Pain Points:** Subscription fatigue, concerned about AI replacing human judgment/cultural recipes, language barriers with English-only apps, worries about "healthy enough" meals on budget, inconsistent internet access
- **Annual spend on meal solutions:** ~$200 (mostly in-store bargain shopping)

---

## 2. MODERATED DISCUSSION

### **ROUND 1: Current Meal-Planning Challenges & Spending**

**Moderator:** "Let's start with what meal planning looks like for you right now. What's the biggest headache when it comes to feeding your kids during the week?"

**Sarah (Marketing Director):** *leans forward* "Honestly? Decision fatigue. I know this sounds privileged, but I spend mental energy every single week on 'what's for dinner'—checking what we have, cross-referencing with dietary stuff, making sure there's some veggie... It's like 90 minutes every Sunday that I resent spending. And then Tuesday hits and someone says 'I don't like that anymore' and the whole plan falls apart."

**Marcus (Single Father):** *nods* "Yeah, but mine's different. I've got the food allergies—peanuts and shellfish—and I can't mess that up. So I stick to what I know works. Chicken, rice, pasta, ground beef tacos. Safe meals. My oldest is starting to complain about the rotation, but at least nobody ends up in the ER."

**Jennifer (Nutritionist):** "What concerns me is that most family meal plans I see—whether free or paid—are nutritionally unbalanced. Too much pasta, not enough real vegetables. And I'm a nutritionist, so I notice. I spend time redoing plans instead of using them as-is."

**DeShawn (Teacher):** "I'm the opposite—I love cooking, actually. It's my meditation. But I don't *have time* during the week. So I plan loose stuff and hope for the best. Tuesday is usually 'whatever I can throw together fast.'"

**Miguel (Freelancer):** *shifts in seat* "For me, it's money. My wife and I watch the sales flyers. We plan around chicken on sale this week, ground beef next week. An app telling me to buy organic spinach—that's not my world, you know?"

**Rachel (VP Operations):** "I've tried everything. I'm genuinely curious about AI meal planning because I hate the *friction* of other tools. HelloFresh was great for 6 months until I realized I was throwing away plastic packaging weekly. And these recipe apps? They're beautiful but I need the grocery list already made for me, integrated with what I actually have..."

**Moderator:** "Rachel, you mentioned integrating with what you have. Do you feel like you'd *pay* for a solution that saved you 30-60 minutes a week?"

**Sarah:** *immediately* "I would pay $29/month for 90 minutes back. That's like... $90 in billable time to me."

**Marcus:** *pauses* "That's the thing. $29 is almost grocery money for us. So it'd have to save me time *or* save me money on groceries. If it's just convenience? I can't do it."

**Jennifer:** *skeptical* "I'd pay if it was evidence-based. If it came with nutritional facts and could adapt based on micronutrient gaps. But a generic AI recipe app? I can Google that for free."

**DeShawn:** "I'd try it if
Judge notes: Exceptionally well-structured output with richly detailed, distinct personas, realistic dialogue that surfaces genuine tensions and trade-offs, and clear evidence of personas responding authentically to each other rather than in isolation—directly fulfilling the task intent with high practical utility for research purposes.
Grok 4.1 Fast Reasoning 9/10
### (1) Persona Profiles

**Persona 1: Sarah Jenkins**  
- **Age**: 38  
- **Occupation**: Marketing coordinator (full-time office job)  
- **Demographics**: Married, two kids (7 and 10), suburban, middle-income ($80K household)  
- **Psychographics**: Practical, time-crunched optimist; values family time and health but hates waste.  
- **Relevant Behaviors**: Uses free apps like Mealime occasionally; shops at Costco; meal preps on Sundays but skips often.  
- **Pain Points**: Forgets ingredients mid-week, picky eaters at home, guilt over takeout nights.

**Persona 2: Mike Rodriguez**  
- **Age**: 42  
- **Occupation**: Software engineer (remote work, long hours)  
- **Demographics**: Married, one kid (9) with nut allergy, urban apartment, upper-middle income ($150K household)  
- **Psychographics**: Tech enthusiast, data-driven; prioritizes efficiency and allergies but skeptical of "hype" apps.  
- **Relevant Behaviors**: Relies on Google recipes and Alexa for lists; tracks allergies manually in notes app.  
- **Pain Points**: Double-checking labels stresses him out; wife handles most cooking, but he's involved weekends.

**Persona 3: Lisa Patel**  
- **Age**: 35  
- **Occupation**: Elementary school teacher (summers off, busy school year)  
- **Demographics**: Divorced single mom, three kids (5, 8, 11), modest income ($60K), lives in mid-sized city.  
- **Psychographics**: Budget hawk, creative cook; family-oriented but overwhelmed by solo parenting.  
- **Relevant Behaviors**: Pins Pinterest recipes; uses free Yummly; clips coupons religiously.  
- **Pain Points**: Tight grocery budget, varying kid preferences daily, no time for custom planning.

**Persona 4: Jamal Washington**  
- **Age**: 40  
- **Occupation**: Delivery driver (shift work, irregular hours)  
- **Demographics**: Married, two kids (6 and 12), rural suburb, working-class ($70K household).  
- **Psychographics**: Practical skeptic, family man; prefers simple, hearty meals over "fancy" trends.  
- **Relevant Behaviors**: Freezes leftovers; uses basic grocery apps like AnyList; cooks from memory.  
- **Pain Points**: Unpredictable schedule leads to fast food; kids want variety but he sticks to basics.

**Persona 5: Emily Chen**  
- **Age**: 36  
- **Occupation**: Freelance graphic designer (works from home)  
- **Demographics**: Married, two kids (8 and 10), vegetarian household, coastal city, comfortable income ($120K).  
- **Psychographics**: Health-conscious trendsetter; loves wellness apps, eco-friendly, experimental eater.  
- **Relevant Behaviors**: Subscribes to HelloFresh occasionally ($10/week extra); tracks nutrition via MyFitnessPal.  
- **Pain Points**: Balancing vegetarian kid meals with fun factor; waste from uneaten experiments.

**Persona 6: Tom Bradley**  
- **Age**: 44  
- **Occupation**: Construction foreman (physical job, early mornings)  
- **Demographics**: Married, one kid (7) with ADHD/food sensitivities, small town, middle-income ($90K).  
- **Psychographics**: No-nonsense traditionalist; trusts family recipes over tech; values reliability.  
- **Relevant Behaviors**: Avoids apps, uses paper lists; grills weekly; shops weekly at Walmart.  
- **Pain Points**: Kid's pickiness causes dinner battles; distrusts "AI gimmicks" after bad smart fridge experience.

### (2) Moderated Discussion

**Moderator**: Welcome everyone to this synthetic focus group on a new AI meal-planning app for busy parents. It's $29/month, generates weekly kid-friendly recipes and grocery lists tailored to dietary restrictions for kids 5-12. Free alternatives exist like basic recipe apps. Round 1: Would you pay $29/month for this over free options? Why or why not? Sarah, start us off.

**Sarah Jenkins**: Hmm, $29 a month? That's like a Netflix sub, but for meals? I'd consider it if it saves me time—I waste $50/week on forgotten stuff or takeout. Free apps are okay but generic; this sounds personalized. But I'd trial it first.

**Mike Rodriguez**: Pass outright. Free tools like ChatGPT or Mealime do 80% of this already—I just prompt for allergies. $29 is steep when I'm already subbed to three apps. Show me ROI, like cost savings data.

**Lisa Patel**: No way on my budget—$29 is two tanks of gas or kids' lunches. Free Pinterest works fine; I tweak for my crew. Unless it auto-coupons or budgets groceries, forget it.

**Jamal Washington**: Nah, man. I stick to what I know—paper lists are free and don't glitch. $29 for AI? My shifts change, so plans flop anyway. Free freezer meals from YouTube suit us.

**Emily Chen**: Yes, potentially! I pay for premium features now. If it handles vegetarian swaps and nutrition scores for kids, that's gold. Free apps don't integrate restrictions seamlessly—worth $29 if it cuts my planning from 2 hours to 10 minutes.

**Tom Bradley**: Absolutely not. AI for kids' food? Sounds risky—what if it suggests junk? I trust my meat-and-veggies routine. Free family recipes online are plenty; don't need another bill.

**Moderator**: Great candor. Round 2: What features would make you sign up (or upgrade from free)? Build on others' points. Mike?

**Mike Rodriguez**: Allergy auto-scan via photo upload—that'd hook me. Integrate with Instacart for one-click lists. And prove accuracy with user ratings on recipes, not just AI magic.

**Sarah Jenkins**: Mike's right on Instacart, plus kid-approval ratings from other parents. Weekly themes like "taco Tuesday" to fight boredom. Emily, love the nutrition bit—add waste-minimizing portion tweaks for picky eaters.

**Lisa Patel**: Budget integration! Scan my fridge remnants and suggest meals under $100/week. Jamal, for your shifts—flexible "swap day" options. No $29 without coupon auto-clipping for Walmart.

**Jamal Washington**: Lisa's budget thing yeah, but make it simple: Print-friendly lists for my truck. Kid-voted recipes so they eat it. Tom's got a point—keep it basic, no fancy AI overrides.

**Emily Chen**: Psychographic matching? Like "adventurous veggie" vs. "plain Jane" for kids. Eco-grocery swaps (local produce). Sarah, portion AI based on age/weight would prevent my waste issue.

**Tom Bradley**: If it lets me input my own recipes first and AI just tweaks, maybe. But family-tested stamps, not AI-generated from scratch. Distrust that—show human chef oversight.

**Moderator**: Noted tensions on trust and simplicity. Round 3: Objections or trust concerns with AI meal plans for families? Especially kids. Lisa?

**Lisa Patel**: Trust? AI doesn't know my kids' moods—one day mac 'n cheese, next hate it. What if allergies slip? Needs parent-verified database. And data privacy—my family's diet history?

**Sarah Jenkins**: Lisa, totally—transparency on how AI picks recipes (e.g., "95% kid-approved"). Objection: Over-reliance; I still want to customize easily. But for busy nights, it'd ease guilt.

**Mike Rodriguez**: Accuracy first—hallucinations like wrong substitutes could hospitalize my kid. Need citations from USDA or pediatric sources. Emily's eco stuff cool, but verify sustainability claims.

**Emily Chen**: Mike, valid. I'd want trial recipes with nutrition breakdowns linked to science. Objection: Kid appeal—AI might make "healthy" bland. Test with focus groups of actual children?

**Jamal Washington**: AI feels impersonal—like robot feeding my family. What if power outage, no app? Back to basics. Tom's skepticism mirrors mine—prove it won't push trends like vegan on my meat-lovers.

**Tom Bradley**: Big red flag: No human touch. My kid's sensitivities aren't "data"—AI misses nuances. Past AI failed me (fridge suggested spoiled milk). For kids? Liability nightmare—add disclaimers.

**Moderator**: Round 4 wrap-up: Any final reactions to each other? Changes of heart?

**Sarah Jenkins**: Tom's caution makes me hesitate—need beta testing stories. But Emily's features excite me.

**Mike Rodriguez**: Still no paywall without proof, but allergy photo-scan could sway.

**Lisa Patel**: Budget features might trial me, but $29 too high—drop to $15.

**Jamal Washington**: Keep it optional for us simple folks.

**Emily Chen**: Enthused, but trust fixes needed.

**Tom Bradley**: Sticking to no—too many unknowns.

### (3) Synthesis & Insights

- **Key Themes**: Time-saving personalization (e.g., allergies, budgets) drives interest, but price sensitivity and AI distrust dominate; free alternatives seen as "good enough" for basics.
- **Points of Agreement**: All agree on need for allergy handling, kid-appeal ratings, and grocery integration (e.g., Instacart); value simple, flexible plans over rigid AI.
- **Points of Disagreement**: Price viability ($29 too high for budget-conscious like Lisa/Jamal vs. worth it for tech/health-focused like Emily); AI trust (Tom/Jamal skeptical of "gimmicks" vs. Mike/Emily wanting verifiable data).
- **Surprising Insights**: Tech-savvy parents (Mike/Emily) demand more proof than traditionalists; waste reduction and "kid-voting" emerged unprompted as hidden desires; single parents prioritize budget over premium features.
- **Un
Judge notes: Excellent execution with diverse, realistic personas, authentic multi-round discussion with genuine tensions and disagreements, and actionable synthesis that directly addresses all three research questions with clear insights and validation flags.

What makes these work

  1. 01
    Define personas with specifics

    Generic personas produce generic output. Instead of 'millennials,' write '31-year-old renter in Austin, two roommates, $72K salary, shops at Trader Joe's.' The more texture you give, the less the model defaults to marketing-speak averages.

  2. 02
    Ask for disagreement explicitly

    Models trend toward consensus. Add a line like 'at least one persona should strongly disagree' or 'include a skeptic.' This surfaces the objections you actually need to hear, not a polite chorus of approval.

  3. 03
    Request evidence, not just opinions

    Tell the model to include what would change each persona's mind. 'Trust level 4/10, would increase to 8 if the claim cited a peer-reviewed study.' This turns vague reactions into testable hypotheses for real research.

  4. 04
    Run the same prompt across models

    GPT, Claude, and Gemini produce meaningfully different personas because their training data and safety tuning differ. Running the same stimulus through 2-3 models and comparing triangulates signal and exposes where any single model is biased.

More example scenarios

#01 · DTC skincare brand testing a new claim
Input
Simulate a focus group of 5 women aged 28-45 who buy prestige skincare ($40+ per product). Show them this claim: 'Clinically proven to reduce fine lines in 7 days.' Capture initial reaction, trust level (1-10), specific objections, and what evidence would make them believe it.
Expected output
Persona 1 (Maya, 34, tech PM): Trust 4/10. 'Seven days feels like marketing hyperbole. I'd want to see the study size and the methodology.' Persona 2 (Linh, 41, dermatology nurse): Trust 2/10. Knows 'clinically proven' is often an in-house 30-person panel. Wants peer-reviewed citations. Full transcript continues with 3 more personas plus a synthesis of common objections and credibility boosters.
#02 · B2B SaaS testing a pricing page rewrite
Input
You are simulating 4 heads of RevOps at Series B-D SaaS companies evaluating our pricing page. Current page leads with 'Starting at $2,000/month.' New version leads with 'Custom pricing based on pipeline volume.' Which converts better for their profile, and why?
Expected output
3 of 4 personas prefer 'Starting at $2,000/month' because it filters out anchoring anxiety and lets them qualify internally before a call. The dissenter (enterprise RevOps at a 400-person company) prefers 'Custom pricing' because her deals always exceed list price and she doesn't want procurement anchored low. Recommendation: keep the price, add a 'for teams above 50 reps, contact sales' line.
#03 · Consumer app testing a feature name
Input
Simulate 6 Gen Z users (18-24, heavy TikTok users) reacting to three names for a new feature that lets them co-edit videos with friends in real time: 'Jam', 'CoCut', 'Studio Together'. Rank by appeal and give reasoning in their voice.
Expected output
Ranking: 1. Jam (5/6 votes) 'sounds fun, short, not trying too hard.' 2. CoCut (1/6) 'I get it but it sounds like a haircut app.' 3. Studio Together (0/6) 'sounds like a Zoom feature my mom would use.' Common theme: Gen Z rewards names that feel native to group chat, not product marketing.
#04 · Non-profit testing a donation appeal
Input
Simulate 5 recurring donors aged 55-70 to a climate non-profit. Show them two email subject lines: A) 'We're running out of time' B) 'Your $25 last month funded 3 tree plantings.' Which drives them to open and donate, and what does that tell us?
Expected output
4 of 5 prefer B. Reasoning centers on specificity and agency. 'A feels like every other climate email in my inbox. B makes me feel like my money did something.' The dissenter responds to urgency but notes she'd unsubscribe if A repeated. Recommendation: lead with concrete impact receipts, reserve urgency framing for end-of-year or crisis moments.
#05 · Fintech testing onboarding friction
Input
Simulate 4 first-time investors (ages 25-35, income $60-90K) going through a 9-step signup flow. At step 4 we ask for SSN before showing any product value. Capture their thoughts, drop-off likelihood, and what would make them continue.
Expected output
All 4 flag SSN-at-step-4 as a drop-off trigger. 'I haven't seen a single screen of the actual product. Why do you need my SSN?' Drop-off likelihood: 60-80%. Fix: show a portfolio preview or simulation first, then request SSN at account-funding step with a one-line explanation of why regulation requires it.

Common mistakes to avoid

  • Treating output as real data

    Synthetic responses are pattern-matched plausibility, not lived experience. Using them to justify a roadmap decision or investor narrative without human validation is how teams ship products nobody wanted.

  • Vague persona definitions

    'Simulate 5 busy professionals' produces 5 versions of the same LinkedIn influencer voice. Without age, income, geography, job role, and a specific life detail, you get flattened stereotypes instead of useful variance.

  • Skipping the skeptic

    If you don't explicitly request disagreement, models produce warm, agreeable output. You walk away feeling validated and miss the one objection that would have killed the concept in real qual.

  • Over-indexing on one model

    Each model has tuning biases. Claude tends toward nuance and caveats, GPT toward structured summaries, Gemini toward broader demographic framings. Using only one means inheriting its blind spots as your research findings.

  • Using synthetic groups for final decisions

    Directional tool, not decisional tool. Teams that replace all qualitative research with AI panels end up with confident wrong answers, because the model will happily invent cultural nuance it doesn't actually know.

Related queries

Frequently asked questions

How accurate are AI synthetic focus groups compared to real ones?

For directional signal on messaging, naming, and concept appeal, they correlate reasonably well with real panels, especially for well-represented demographics. For emotional depth, cultural nuance, or niche B2B roles, accuracy drops sharply. Use them to narrow options, not to make final calls.

How much does a synthetic focus group cost to run?

Typically under $5 in API costs per session of 5-6 personas, compared to $10-30K for a traditional recruited focus group. The real cost is the time to design good prompts and the discipline to validate findings with actual humans before acting.

Which AI model is best for synthetic focus groups?

Claude tends to produce the most distinct persona voices and pushback. GPT-4 produces the cleanest structured output for comparison tables. Gemini handles broader demographic framings well. Running 2-3 models on the same prompt and comparing is more reliable than picking one.

Can I use synthetic focus groups for academic or published research?

Not as primary evidence. Most journals and ethics boards require human subjects with informed consent. You can use synthetic panels to pilot your instruments, refine questions, and generate hypotheses, but the published findings need real participants.

How do I prevent AI personas from sounding the same?

Give each persona specific constraints that conflict: different income, different risk tolerance, different life stage, different relationship to the category. Explicitly instruct the model to preserve disagreement and to write in the persona's actual speech register, not in polished marketing prose.

What's the difference between synthetic personas and synthetic users?

Synthetic personas react to stimuli (ads, concepts, pricing) in a focus-group format. Synthetic users typically simulate behavior over time (clicking through a flow, abandoning a cart). Both use LLMs, but synthetic users often layer in behavioral models and are used more in UX and growth testing than in market research.