Cold email

Synthetic buyer personas: what AI simulated prospects can and cannot tell you

AI simulated prospects can test your targeting, copy and offer before any real send, or they can flatter you. What synthetic personas are, what the research says about their accuracy, and how to use them without fooling yourself.

Synthetic buyer personas: an overhead view of a large crowd of people crossing a busy city intersection, each one a different kind of buyer

Synthetic buyer personas are simulated prospects generated by AI from a description of your audience: each one has a role, a company, a buying stage, pains and objections, and reacts to your message the way such a person plausibly would. Used well, they let you test targeting, copy and offers before you contact a single real buyer, and see why people would pass. Used badly, they flatter you. Research suggests well conditioned AI personas can mirror the response patterns of real groups, but they are best for catching problems and comparing options, not for final proof. An AI email simulator that tests your campaign on simulated prospects is one practical way to use them for cold email.

This guide explains what synthetic personas are, what they are good and bad at, and how to use them without fooling yourself. It is part of our series on how to test a cold email before you send it.

What are synthetic buyer personas?

A classic buyer persona is a written profile: "Operations Olivia, 42, runs a team of 12, worries about missed deadlines." It is a document you read. A synthetic persona is a profile an AI model takes on, so you can ask it questions or show it a message and get an answer in that person's voice.

You will see several names for the same idea: synthetic users, simulated prospects, AI personas, silicon samples, AI focus groups and digital twins. They differ in detail, but all of them use a language model to stand in for a type of person.

For cold email, the useful version is not one persona but a population: thousands of simulated prospects that vary the way a real list varies, across titles, seniority, company sizes, regions, buying stages and attitudes, so that the overall pattern of responses means something.

Key numbers

FindingWhat it meansSource
"Algorithmic fidelity"Language models conditioned on real people's backgrounds reproduced the response patterns of many human subgroups in US surveys.Argyle et al., Political Analysis, 2023
Hypotheses, not decisionsSynthetic users are useful for desk research and generating hypotheses, but not for final decision making.Nielsen Norman Group, 2024
Too kindSynthetic users tend to give overly favorable feedback and are less reliable for niche groups.Nielsen Norman Group, 2024
95%Share of potential B2B buyers who are not ready to buy today. A realistic simulated list must reflect that most people will not reply.LinkedIn B2B Institute, the 95-5 rule
3.43%Average real cold email reply rate. A simulation that predicts far more than this for an ordinary campaign is too optimistic.Instantly, Cold Email Benchmark Report 2026

What can simulated prospects tell you?

Simulated prospects are good at the questions where the answer depends on how a type of person reads a message:

  • Relevance. Would a head of operations at a 50 person logistics firm see this as their problem? If many simulated prospects say "not relevant to my role", your targeting or angle is off.
  • Clarity. Is it obvious what you offer and what they would get? "Unclear what I would get" is a common and fixable reason to pass.
  • Friction. Is the ask too big for a first contact? A 45 minute demo request reads differently from "worth a look?".
  • Objections. Which objections come up most: no budget, already have a vendor, no proof, bad timing?
  • Comparisons. Which of several versions do the same simulated prospects respond to best? Because each version meets the same population, the list cannot bias the comparison.
  • Segments. Which titles, industries, sizes or regions respond best, which tells you where to focus the real list.

What simulated prospects cannot tell you

Be just as clear about the limits:

  • Your real list's quality. A simulation assumes the people you describe. If your real list is full of wrong titles and dead addresses, the real result will be worse.
  • Timing inside a specific company. Whether this account has budget this quarter, or just signed with a competitor, is not something a simulated persona can know.
  • Exact numbers. A simulation gives a range and a direction. Treat any single predicted figure as an estimate, not a forecast you can bank.
  • Brand new or very niche markets. Where the model has little to draw on, persona behavior is less reliable, as the Nielsen Norman Group notes.
  • Politeness bias. A model asked "would you like this?" tends to say yes. A useful simulation is built to be skeptical, and its predictions are checked against real benchmarks.

How to tell a good simulation from a flattering one

Question to askGood signWarning sign
Does it predict realistic reply rates?Ordinary campaigns land in low single digits, close to published benchmarksMost drafts get 15% or more replies
Does it give reasons for passing?Specific reasons you can act on, with how common each isOnly scores, or only praise
Does it vary the audience?Many prospects across roles, sizes, stages and attitudesA handful of personas that all think alike
Does it admit uncertainty?Ranges, confidence levels and "too close to call"Always names a winner, however small the gap
Does it punish weak inputs?A placeholder or generic email scores badlyEven "test" gets a decent score

How to use synthetic personas for cold email, step by step

  1. Describe the real audience, specifically. Titles, industries, company sizes, regions, the tools they use, their pains, the objections you hear and their buying stage. Vague input gives vague output.
  2. Describe the offer honestly. What you sell, the result with a number, your proof, your price band and what makes you different. If you have no proof yet, say so.
  3. Test the real sequence. The emails you would actually send, follow-ups included.
  4. Read the reasons, not just the score. The main reason people pass tells you which link is weak: targeting, copy or offer.
  5. Fix one thing and run it again. Compare the two runs on the same audience.
  6. Compare real alternatives. Use the same population to compare two to five genuinely different versions.
  7. Confirm with real buyers. Send the strongest version to a first real batch and check the result against the prediction.

For what to look for in the copy itself, our cold email copy checklist pairs well with a simulated test, and the subject line tester is a quick first check on the subject.

Synthetic personas vs real customer research

Synthetic personas do not replace talking to customers. They make the conversations you do have more useful. A sensible split looks like this: use real interviews and your won deals to learn who buys and why, write that down as your ideal customer profile (our guide on ideal customer profile and lead scoring shows how), then use synthetic personas built from that profile to test many messages quickly before you spend real sends. When real replies come in, feed what you learn back into the profile.

Where ZicReach helps

The ZicReach AI Simulator builds a population of AI agents from the audience you describe, filled in from your AI Training: job titles, industries, the niche, department, seniority, regions, company sizes, tools, buying stage, pains and objections. You choose the size, from 10,000 to 1,000,000 agents. The agents react to your sequence, and the report shows who the agents were, who replied and why the others passed.

It is built to be skeptical. Results are deliberately conservative and held to the benchmarks of your market, a placeholder email is flagged as not a real email, and when two versions are close the A/B report says they are too close to call. You get Prospect fit, Copy and Offer verdicts, a reply heatmap by title and industry, "Weak areas and fixes" ranked by the replies each would add, and a confidence note on the method. Nothing is emailed to anyone, and the result is a prediction, not a guarantee.

Questions people ask

What are synthetic personas?

Synthetic personas are AI generated stand-ins for types of people, built from a description of an audience. You can show them a message or ask them questions and get answers in that person's voice, which makes them useful for early testing and research.

Are synthetic users accurate?

They can be useful but are not exact. Research such as Argyle and colleagues (2023) found well conditioned models reproduced real group response patterns, while the Nielsen Norman Group warns they tend to be too positive and are weaker for niche groups. Use them for hypotheses and comparisons, then confirm with real people.

What is the difference between synthetic personas and digital twins?

A synthetic persona represents a type of person, such as "a head of operations at a mid size logistics firm". A digital twin usually models one specific real person or customer from their own data. For cold email testing, personas are the practical choice.

Can AI predict how prospects will respond to a cold email?

It can estimate how a described audience is likely to respond and why, and compare versions on the same simulated audience. It cannot know the situation inside a specific company, so treat the result as a range and a direction, and confirm it with a real send.

How many simulated prospects do I need?

Enough to cover the variety in your real audience and to make differences between versions stable. A few personas give anecdotes; thousands give a pattern. In ZicReach the smallest test size is 10,000 agents.

The short version

Synthetic buyer personas let you test targeting, copy and offers on a simulated version of your audience before any real send. They are good at relevance, clarity, objections and comparisons, and weak at exact numbers and company specific timing. Describe your audience precisely, choose a simulation that is skeptical and admits uncertainty, act on the reasons people pass and confirm the winner with real buyers.

To test your next campaign on simulated prospects built from your own audience, start ZicReach for free. The pricing page shows what each plan includes.

Cover photo by Ryoji Iwata on Unsplash.

Put this into practice with ZicReach

Find leads, write the sequence and send from warmed inboxes in one place. Free to start, no card needed.

Start for free

ZicReach Team

Outbound and deliverability

The people who build ZicReach. We run our own outbound on it every day and write up what works, what breaks and how to fix it.

Put this week's reading to work.

Start with 100 free credits, find your first list and send from warmed inboxes. ZicAgent sets up the rest.

100 free credits when you sign upNo card requiredFree Email Warmup, always

100free credits on every new account, no card required