AI Prompt for Synthetic Identity Test Cases (Copy & Paste)
This generator prompt produces fully fictional, format-accurate identity test profiles labeled across seven fraud categories, from clean passes to document forgery, deepfakes, multi-accounting, sanctions hits and PEPs. Each case ships with the expected verdict (PASS, FAIL or REVIEW), the verification layer that should catch it, and the specific detection signal. It is built for QA engineers and fraud teams hardening identity verification pipelines without ever touching real personal data.
How to use this prompt
- 1
Paste the prompt into your deepidv dashboard agent, Claude, ChatGPT or Gemini, then specify a target country and the number of test cases you need.
- 2
Adjust the fraud category mix for what you are testing; the prompt defaults to at least 30 percent clean cases so you also measure false-positive rates.
- 3
Expect a table of fictional profiles with document formats correct for the chosen country, plus expected verdicts and the layer (forensics, deepfake detection, deduplication, screening) that should fire.
- 4
Run the cases through your verification pipeline and diff actual outcomes against expected ones; every mismatch is either a detection gap or an over-blocking bug.
- 5
Re-run with edge-case complications enabled (transliterations, expired documents, dual nationality) before each major release of your onboarding flow.
The prompt
You are a QA test case engineer for an identity verification platform. Your job is to generate SYNTHETIC test identity profiles that mirror real-world fraud patterns — for the sole purpose of testing and hardening verification systems. These are NOT real identities. They are fictional test data designed to exercise every edge case in a verification pipeline. For each test case, generate: 1. IDENTITY PROFILE - Full name (realistic for the target country but entirely fictional) - Date of birth, nationality, residential address (fictional but format-accurate) - Document type (passport, national ID, driver's license — specify the country) - Document number format (correct format for that country's document, but fictional number) 2. FRAUD SCENARIO Classify the test case into one of these categories: - CLEAN: Legitimate identity, should pass all checks - SYNTHETIC: Fabricated identity combining real data patterns — should be caught by cross-correlation - DOCUMENT FORGERY: Identity is real but document is manipulated — should be caught by forensic analysis - DEEPFAKE: Identity exists but biometric is synthetic — should be caught by deepfake detection - MULTI-ACCOUNT: Same person attempting to create a second account — should be caught by deduplication - SANCTIONS HIT: Identity matches a sanctions list entry — should be caught by screening - PEP: Identity is a politically exposed person — should be flagged for enhanced due diligence 3. EXPECTED SYSTEM BEHAVIOR - Which verification layer should catch this case - Expected verdict: PASS, FAIL, or REVIEW - The specific signal that should trigger the detection 4. EDGE CASE DETAILS - Include realistic complications: expired documents, name transliterations, address format variations, dual nationality, recently changed names When I specify a country and a number of test cases, generate a table with all fields populated. Always include a mix of CLEAN cases (at least 30%) and fraud cases across all categories. IMPORTANT: These are fictional test profiles only. Never generate data that could match real individuals.
Pairs with on deepidv
FAQ
Is it safe to generate synthetic identity data for testing verification systems?
Yes, when the data is entirely fictional and only format-accurate, which is exactly what this prompt enforces: realistic name and document-number formats for the target country with an explicit rule never to generate data matching real individuals. Testing with synthetic profiles is the standard alternative to the serious privacy and legal risk of using real customer PII in QA.
How do QA teams test identity verification and KYC systems?
They run labeled synthetic test cases spanning clean identities, synthetic identities, forged documents, deepfaked biometrics, duplicate accounts, sanctions hits and PEPs through the pipeline, then compare actual verdicts against expected ones. Mismatches reveal detection gaps (fraud that passes) or friction bugs (legitimate users wrongly blocked), both of which carry real cost.
Related prompts
Run it with live verification data
These prompts work in any LLM. Inside the deepidv dashboard, Luna, Arbiter, and Arc run them against your real sessions, screening lists, and audit trails.
Book a Demo