Home › Blog › Customer persona workflow

A 30-minute AI workflow for customer personas

Persona workshops often burn half a day and land on "professionals in their thirties". With AI you can reach a testable hypothesis in thirty minutes. What you must not do is mistake it for an answer.

4 steps 30 minutes Validation included

The short answer

The value of an AI-generated persona is specificity, not accuracy. The goal is to quickly get a concrete statement you can prove right or wrong, instead of a conclusion like "professionals in their thirties" that enables no decision at all.

Which is why the last step of the workflow is always validation. Skip it and put the output straight into ad targeting and landing page copy, and you are spending budget on the basis of an untested guess.

Step 1 — organise the inputs (10 minutes)

Product description — Write it as "whose problem does this solve, and how" rather than as a feature list. Feed in features only and you get a feature-centred persona.

Target market — B2B or B2C, which industry, what size of company. Leave this out and the output drifts into generalities.

What you already know — If you have existing customers, include their characteristics. If you do not, say so explicitly. Pretend you do and the AI will invent to match.

Step 2 — generate three personas, not one (5 minutes)

Asking for three rather than one matters. Generate a single persona and it hardens into the answer. Put three side by side and they read as competing hypotheses, which creates the motivation to find out which is real.

Four items must appear in each: profile (role and situation), real concerns (in everyday language, unrelated to your product), purchase trigger (what has to happen before they start looking), and messaging angle (what phrasing they would likely respond to).

Step 3 — for B2B, split the buying structure (5 minutes)

A single persona fails in B2B. The person who uses it, the person who approves it and the person who resists it are different people with different concerns.

The practitioner asks whether their own work gets easier. The approver asks about cost, benefit and risk. The resister is usually in information security or owns an existing system, and asks whether it complies with policy. Each axis needs different messaging — and omitting the third is what blocks adoption at the final stage.

Step 4 — validate (this is the actual work)

Three customer interviews — The most reliable method. Read the persona’s "real concerns" statement aloud and ask whether it resembles their own situation.

Check against sales records — If you already have sales activity, compare real enquiries against the persona. Where they diverge, the persona is wrong.

Small paid-ad test — Run small budgets per messaging angle and compare click-through. The fastest way to test the hypothesis.

Check search demand — See whether the phrasing this persona would actually search has any volume. If it has none, either the concern does not exist or it is described in different words.

Where an AI persona must not be trusted

The most dangerous output is numbers. When you get a sentence that looks statistical — market size, what percentage of an age group — do not use it. There is no source behind it.

The second is generating without feeding in the customers you already have. Leave real data out and you get a plausible persona that differs from your actual customers, and that then destabilises how you serve the customers you have.

Frequently asked questions

How many personas is the right number?

Three at the start, narrowed to one or two after validation. Organisations that maintain five or more generally use none of them.

Is the product information I enter stored?

It varies by tool. Some free generators destroy it immediately after generating. Worth checking before entering unreleased product information.

Can we run ads without validating first?

A small test is itself validation. But before committing real budget, going through either interviews or a sales-record comparison is the better call.

Can it connect to our CRM data?

Not at the level of a generation tool. If you need live segments connected to real customer data, that moves into building a data pipeline.

Three from a product description

Hypothesis in thirty minutes, validation after

Build three hypotheses with the free persona generator, and if the validated result needs to become real segments, talk to us about a build.

Get in touch