ditto-product-research（产品研究同上）

Ditto Product Research

Run customer research, pricing tests, and product validation studies using Ditto's 300,000+ synthetic personas — directly from the terminal.

Full documentation: https://askditto.io/claude-code-guide

What Ditto Does

Ditto maintains 300,000+ AI-powered synthetic personas calibrated to census data across USA, UK, Germany, and Canada. You ask them open-ended questions and get qualitative responses with the specificity of real interviews.

• 92% overlap with traditional focus groups (EY Americas validation)
• 95% correlation with traditional research
• Harvard/Cambridge/Stanford/Oxford peer-reviewed methodology
• A 10-persona, 7-question study completes in 10-12 minutes
• Traditional equivalent: 4-8 weeks, $10,000-50,000

Quick Start (Free Tier)

Get a free API key — no credit card, no sales call:

curl -sL https://app.askditto.io/scripts/free-tier-auth.sh | bash

Or visit: https://app.askditto.io/docs/free-tier-oauth

Ask a question immediately:

curl -s -X POST "https://app.askditto.io/v1/free/questions" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question": "What frustrates you most about grocery shopping?"}'

Free keys (rk_free_): ~12 shared personas, no demographic filtering. Paid keys (rk_live_): custom groups, demographic filtering, unlimited studies.

API Essentials

Base URL: https://app.askditto.io Auth header: Authorization: Bearer YOUR_API_KEY Content-Type: application/json

The Standard Workflow (6 Steps)

IMPORTANT: Follow these steps in order. Questions MUST be asked sequentially — wait for all responses before asking the next.

Step 1: Recruit Your Panel

curl -s -X POST "https://app.askditto.io/v1/research-groups/recruit" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "US Adults 30-55",
    "group_size": 10,
    "filters": {"country": "USA", "age_min": 30, "age_max": 55}
  }'

Save the uuid from the response (5-15 seconds).

CRITICAL: Use group_size not size. Use group uuid not id. State filter uses 2-letter codes ("MI" not "Michigan").

Step 2: Create Study

curl -s -X POST "https://app.askditto.io/v1/research-studies" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "title": "Product Concept Validation",
    "objective": "Understand customer pain points and validate pricing",
    "research_group_uuid": "UUID_FROM_STEP_1"
  }'

Save the study id. Response nests under data.study — access via response["study"]["id"], NOT response["id"].

Step 3: Ask Questions (One at a Time)

curl -s -X POST "https://app.askditto.io/v1/research-studies/STUDY_ID/questions" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question": "Your open-ended question here"}'

Returns job_ids (one per persona). Poll until complete before asking next.

Step 4: Poll Until Complete

curl -s "https://app.askditto.io/v1/jobs/JOB_ID" \
  -H "Authorization: Bearer $DITTO_API_KEY"

Polling strategy for 10-persona study:

• Wait 45-50 seconds before first poll
• Then poll every 20 seconds
• Poll ONE job_id as proxy — all jobs from the same question finish together
• Status: queued → started → finished (or failed)

ALL jobs must show finished before asking the next question.

Step 5: Complete the Study

curl -s -X POST "https://app.askditto.io/v1/research-studies/STUDY_ID/complete" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"force": false}'

Triggers AI analysis: summary, segments, divergences, recommendations (20-40s). Use "force": true to re-run analysis on an already-completed study (avoids 409).

Step 6: Get Share Link

curl -s -X POST "https://app.askditto.io/v1/research-studies/STUDY_ID/share" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"enabled": true}'

Returns a public URL. Use share_link field (preferred over share_url). To check existing share state without changing it:

curl -s "https://app.askditto.io/v1/research-studies/STUDY_ID/share" \
  -H "Authorization: Bearer $DITTO_API_KEY"

Workflow Patterns

Standard 7-Question Study

The default. Recruit 10 personas → create study → ask 7 questions → complete → share. ~10-12 minutes total.

Over-Recruit & Curate

For niche audiences or when participant quality matters most (e.g., VC diligence):

1. Recruit 15-20 participants: "group_size": 15
2. Create study from that group
3. Ask Q1 as a screening question ("Tell me about your role and how often you...")
4. Review Q1 responses — score each agent: High / Medium / Low relevance
5. Remove low-relevance agents from the study (not the group): curl -s -X POST "https://app.askditto.io/v1/research-studies/STUDY_ID/agents/remove" \ -H "Authorization: Bearer $DITTO_API_KEY" \ -H "Content-Type: application/json" \ -d '{"agent_ids": [123, 456, 789]}' agent_ids must be list[int] — never strings or UUIDs.
6. Ask Q2-Q7 to curated panel only. Keep minimum 8 personas.

Quick Question (No Study)

For 1-5 quick answers without creating a full study:

To a specific agent:

curl -s -X POST "https://app.askditto.io/v1/research-agents/AGENT_ID/questions" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question": "What matters most when choosing a coffee brand?"}'

To an entire group:

curl -s -X POST "https://app.askditto.io/v1/research-groups/GROUP_ID/questions" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question": "What matters most when choosing a coffee brand?"}'

Free Tier

~12 shared personas, no demographic filtering. Good for quick tests:

curl -s -X POST "https://app.askditto.io/v1/free/questions" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question": "What frustrates you most about meal planning?"}'

Three-Phase Iterative

For deeper validation, run 3 sequential studies (can share the same group):

Each phase's questions should be informed by the previous phase's findings. ~30-45 minutes total.

Resume Stalled Study

If a study was interrupted mid-way:

1. Get study details: GET /v1/research-studies/STUDY_ID
2. List existing questions: GET /v1/research-studies/STUDY_ID/questions
3. Resume from the next unanswered question
4. Complete when all questions are asked

Zeitgeist Survey

Quick single-question survey with predefined answer options:

curl -s -X POST "https://app.askditto.io/v1/zeitgeist/surveys/create" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "title": "Brand Awareness Pulse",
    "question": "Which brand comes to mind first for sustainable coffee?",
    "research_group_uuid": "GROUP_UUID",
    "answer_options": ["Brand A", "Brand B", "Brand C", "Other"]
  }'

Get results: GET /v1/zeitgeist/surveys/SURVEY_ID/results Delete: DELETE /v1/zeitgeist/surveys/SURVEY_ID

Natural Language Requests

Let the system design your study or group from a plain-text brief:

Study request:

curl -s -X POST "https://app.askditto.io/v1/research-study-requests" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"request_text": "I want to understand why millennials are switching from coffee to matcha"}'

Group request:

curl -s -X POST "https://app.askditto.io/v1/research-group-requests" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"request_text": "US women aged 25-40 who drink matcha daily"}'

Check group request status: GET /v1/research-group-requests/REQUEST_ID

AI-Assisted Recruitment

Provide an objective and let the system design the group:

curl -s -X POST "https://app.askditto.io/v1/research-groups/interview" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "objective": "Understand premium pet food purchasing decisions",
    "group_size": 10
  }'

The 7-Question Framework

Design studies with this proven sequence (adapt to your domain):

1. Establish relevance — "Walk me through how you currently handle [task]. What does a typical week look like?"
2. Identify pain — "What's the most frustrating part of [task]? What makes you want to throw your laptop out the window?"
3. Quantify impact — "Roughly how much time/money do you lose to [problem] per week?"
4. Current solutions — "What tools or workarounds do you currently use? What works? What doesn't?"
5. Past attempts — "Have you tried switching to something new? What happened?"
6. Magic wand — "If you could fix ONE thing about [task], what would it be and why?"
7. Adoption barriers — "Imagine a tool that solves [problem]. What would make you hesitant to try it?"

Introduce brand/product at Q3 earliest — not Q1 (avoid anchoring bias). The magic wand question (Q6) is consistently the most revealing.

Complete API Reference

Research Groups

⚠️ Note: append uses group_uuid (not group_id) — inconsistent with other endpoints.

Research Studies

Questions

Jobs

Sharing

Media Attachments

Agents

Natural Language Requests

Zeitgeist Surveys

Free Tier

Demographic Filters

Filters go inside the filters dict when recruiting:

NOT supported: income, employment, ethnicity, political_affiliation.

If 0 agents returned: Broaden filters — remove state, remove industry, widen age by +/- 10 years. Try up to 10 relaxation attempts.

Industry proxies: Auto mechanics → Automotive Manufacturing, restaurant owners → Food & Beverages, pet owners → use age filter (not Veterinary), security professionals → Cybersecurity.

Media Attachments

Upload screenshots, ad creative, or PDFs before asking questions:

# Upload image
curl -s -X POST "https://app.askditto.io/v1/media-assets" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "data_url": "https://example.com/screenshot.png",
    "filename": "landing-page.png",
    "mime": "image/png"
  }'

Or use base64: "file_data": "base64_encoded_content", "encoding": "base64". Allowed types: PNG, JPEG, GIF, WEBP, PDF.

Save the media_asset.id and pass as attachments when asking questions:

curl -s -X POST "https://app.askditto.io/v1/research-studies/STUDY_ID/questions" \
  -H "Authorization: Bearer $DITTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "question": "Look at this landing page. What would make you trust this product?",
    "attachments": [MEDIA_ASSET_ID]
  }'

Polling Strategy

All jobs from the same question finish together — no need to poll each one. If a job returns failed, report partial failure. Do NOT auto-retry.

Response Structure

Study creation: response["study"]["id"] (nested under study key)

Question responses (from GET /v1/research-studies/{id}/questions):

• response_text — the persona's answer (may contain HTML: <b>, <ul>, <li>)
• agent_name, agent_age, agent_city, agent_state, agent_country
• agent_occupation, agent_summary

Completion results: overall_summary, key_segments, divergences, shared_mindsets, next_questions, stats.

Share link: Prefer share_link field. If absent, use share_url.

Reading Responses

After polling is complete, retrieve all Q&A data:

curl -s "https://app.askditto.io/v1/research-studies/STUDY_ID/questions" \
  -H "Authorization: Bearer $DITTO_API_KEY"

Use agent demographics to segment findings by age, location, occupation.

What You Can Test

• Pain points — Do customers actually have this problem?
• Pricing — What price is a bargain? A stretch? A dealbreaker?
• Positioning — Which tagline/value prop resonates most?
• Features — What do customers prioritise vs. ignore?
• Landing pages — Upload screenshots and get qualitative reactions
• Ad creative — Test headlines, images, and messaging variants
• Competitive switching — What would make someone leave their current solution?
• Deal breakers — What kills the sale even if the product is good?

Common Mistakes

• Using size instead of group_size in recruitment
• Using numeric id instead of string uuid for research_group_uuid
• Using group_id with the append endpoint (it requires group_uuid)
• Using response["id"] instead of response["study"]["id"] for study creation
• Passing agent_ids as strings/UUIDs instead of list[int]
• Polling every 10-15s (too aggressive — use 45-50s first, then 20s)
• Batching questions (ask one, poll to completion, then ask next)
• Using full state names ("Michigan") instead of 2-letter codes ("MI")
• Including income or employment filters (not supported)
• Skipping the complete step (you miss AI-generated analysis)
• Asking closed-ended yes/no questions (use open-ended instead)
• Asking leading questions ("Don't you think X is great?")
• Designing questions before researching the product/market
• Introducing the brand in Q1 (introduce at Q3 earliest to avoid bias)

Limitations

Ditto personas have NOT used your specific product. For:

• Actual UX feedback from real users → use real user testing
• Legal/compliance decisions → use human research
• Safety-critical decisions → use human validation
• Exact quantitative metrics (NPS, conversion) → use real data

Recommended hybrid: Ditto for the fast first pass (80% of insight), then human research for the remaining 20% requiring real customer nuance.

Error Handling

Further Reading

• Full API guide: https://askditto.io/claude-code-guide
• Question design: @question-playbook.md
• Worked examples: @examples/carequarter.md
• API endpoint reference: @api-reference.md