If your business serves buyers with different requirements, you need your AI research to reflect those differences. Assistants answer conditionally. Supply a different description of who is asking and the same question returns a different set of names. A persona is how you control that variable instead of leaving it undefined.
Why the same question has several correct answers
Ask which tool a team should use to manage client work, with no context, and the assistant must guess at the situation. Ask as a two-person consultancy and the answer will favor simplicity and low cost. Ask as a two-hundred-person firm with compliance requirements and it will favor administration, permissions and support.
Both answers are reasonable. They are answers to different questions, and if you track only the un-personalized version you are measuring an average buyer who does not exist.
The practical consequence: a brand can be strongly present for one buyer description and absent for another, and a single visibility number averages that into something that describes nobody.
What a persona is here
A campaign holds three personas besides a neutral one. Each is a written description of ten to fifty words, in the first person, and the model answers as that person. The neutral persona is not counted toward the three and its description is not measured.
The campaign will not lock until those descriptions exist, which is deliberate. A campaign without them would produce research about a market with no buyer in it.
Writing one that changes the answer
A persona is useful when it names the constraints that would change a recommendation. Job title alone rarely does.
Weak, because nothing in it would alter an answer:
I am a marketing manager at a mid-sized company. I care about results and efficiency.
Better, because every clause is a constraint an assistant can respond to:
I run marketing alone at a 40-person accounting firm. I have no developer, our website is on WordPress, and I have to justify any new subscription over $200 a month to a partner who does not follow marketing.
The second description produces answers that weigh setup effort, technical requirements and price, because those are the conditions it names. That is what makes the resulting measurement worth something.
The elements worth including: what the person is responsible for, what they lack, what constrains the decision, and what would disqualify an option.
Choosing which three
You have only three slots, so use them for descriptions that produce different answers.
The useful test is whether you would give two personas different advice. If you recommend the same product to a 40-person firm and a 400-person firm for the same reasons, use one persona for both. Reserve another slot for a buyer with different needs.
Common productive splits are by size, by whether the buyer is technical, by whether they are replacing something or buying for the first time, and by which department holds the budget.
Reading a persona comparison
The comparison across personas is the output worth acting on, and it takes one of four shapes.
| Pattern | What it suggests |
|---|---|
| Present for all personas | Established for this question. Watch for change rather than acting. |
| Present for one, absent for others | Your positioning has landed with one buyer type. The material that achieved it exists and can be adapted for the others. |
| Absent for the persona you sell to most | The most actionable finding available, and a content and placement problem rather than a formatting one. |
| Present only for the neutral prompt | You appear when no constraints are applied and drop out when real conditions are named. Usually means your material describes what you do but not who it is for. |
The last pattern is common and easy to miss without persona tracking, which is the main argument for doing it at all.
What a persona result is not
It is not evidence about real people. Nobody with that job has been surveyed. What is measured is how an assistant answers when given that description, which is a statement about the assistant and the material it draws on.
That is still useful, because it is the same mechanism your actual buyers encounter. It is not market research, and a persona finding should not be reported as what customers think.
Answers also vary between runs regardless of persona, so compare across a series rather than reacting to one contrast.
When setting up a buyer comparison, describe the requirements that matter to the decision. You can then investigate which recommendations differ for those conditions.
See comparing visibility by persona, how to select a prompt set, or campaigns.