Reference checks create anxiety partly because they're unpredictable, but practicing with AI-generated scenarios lets you prepare for common questions and objections without imposing on real references. You build resilience by experiencing variations of tough questions before they arrive in an actual call.
Synthetic data is artificially generated data mimicking real-world examples without using actual sensitive information. For reentry candidates, this is valuable for practicing reference checks: you can simulate realistic scenarios (skeptical questions from hiring managers, curveballs about your background) without repeatedly contacting your actual references for practice or potentially revealing that you're nervous about checks.
Reference checks are a critical vulnerability point in reentry hiring. A reference—your former supervisor, program coordinator, mentor—might inadvertently undermine you if they're unprepared for questions. Even well-meaning references can falter if asked unexpected angles: "Was there ever concern about reliability?" or "How did they handle conflict?" or (the real fear) "Would you hire them again?" Your actual references need to be coached, but you can't coach them ad infinitum without becoming annoying.
Synthetic reference scenarios let you practice how to talk about your background, your references' likely concerns, and potential objections—without burdening real people with repeated practice sessions.
You provide an AI with information about your background and references: "My reference is Jane Smith, my former reentry program coordinator. She knows I was incarcerated for 5 years, completed vocational training, and worked in her program's job training track." The AI then generates realistic reference check dialogues—what a hiring manager might ask, how Jane (synthesized version) might respond, what concerns might surface.
These dialogues are synthetic—not actual quotes from Jane, but realistic simulations based on typical reference dynamics. The AI uses patterns from thousands of reference conversations to construct plausible scenarios: questions hiring managers ask, answers references typically give, and follow-up probes.
First, you'd work with real references to establish what they're comfortable saying. Then, you'd translate that into a synthetic reference profile: "Jane will emphasize my reliability and growth, will be honest about my criminal history, and may express concern about my long absence from the workforce." The AI then generates scenarios reflecting those parameters.
You'd practice responding: imagine you're the hiring manager calling Jane, or imagine Jane being asked tough questions. Either way, you're mentally rehearsing how your background will be presented and what concerns might surface. This prepares you to handle concerns proactively in your own interviews rather than being blindsided by questions hiring managers might raise based on reference feedback.
Large language models are particularly good at generating synthetic dialogues because they've been trained on diverse conversational data. You'd use a prompt like: "Simulate a reference check conversation. Hiring manager: [role description]. Reference: [your description of the reference]. Generate 5 realistic questions the hiring manager might ask about this candidate's background, reliability, and criminal history."
The model generates coherent, realistic questions without knowing your actual reference or creating legally problematic content—because it's working from generalizable patterns about reference conversations, not your specific data.
Synthetic data is useful for practice but can diverge from reality. A synthesized Jane might give an answer your real Jane wouldn't; the synthetic scenario might miss nuances of how your specific references think about your background. That's why the final step before interviews is real conversation with actual references: "Here's how I'll be describing you in interviews. Are you comfortable with that narrative?"
Also, synthetic data risks overconfidence. If you practice with synthesized scenarios and they're quite positive, you might overestimate real reference strength. Use synthetic practice to identify concerns and build resilience to tough questions, not to assume references will perfectly advocate for you.
Because you're not providing actual information about real references to the AI, this approach protects privacy. You're using the AI to generate plausible scenarios, not actually analyzing your real reference relationships. That's important both ethically (respecting references' privacy) and strategically (references never need to know you've been practice-interrogating their likely answers).
Try this: Describe one of your actual references to ChatGPT: their role, what they know about your background, their general communication style, and realistic concerns they might have. Ask: "Generate 10 realistic questions a hiring manager might ask this reference about my background, reliability, and how I've changed." Practice answering as if you were your reference, or imagine you're the hiring manager probing. Then compare the synthetic scenarios to what you think would really happen—you'll notice what the AI nailed versus where reality diverges, and that gap is valuable insight.
Peri can explain this concept, give practical examples, help you decide whether it applies to your situation, or recommend a journey if appropriate.
Explore related journeys or tell Peri what you're working through.