Start with a question you need answered
If you tell the assistant, “Create a Buyer Persona,” it may give you a name, age, job, and interests, making you feel you understand your audience. But if those details come from its guesses, you still do not know the people you are speaking to. You have an initial picture to test, not a verified description of your customers.
Start with something that matters to your work: why do people ask about the price? What do they not understand about the service? Which alternatives do they compare you with? The assistant can help organize questions, analyze responses, and compare sources. Its help becomes more useful when it knows the question and sees the material that can answer it.
Learning example: A trainer offers an Excel workshop for employees and wants to reach the people who actually need it.
In the workshop example, the question is: “What stops an employee who uses Excel at work from signing up for training?” This is narrower and more useful than “Analyze the education market.” The issue might be timing, level, or uncertainty about what they will be able to do after the workshop. We are still researching, so we cannot choose the answer at the start and steer every question toward proving it.
Gather material from people and the market
You have two types of material that complement each other. Existing information: reports, service pages, published prices, and posted questions. And information you collect yourself: interviews, surveys, or customer questions in your work. A general report might explain a market trend, but it will not necessarily explain why your particular customer hesitates.
- Interviews: Useful for understanding someone's experience and asking for details. Instead of “Surely timing is the problem?”, ask, “Tell me about the last time you considered learning something. What happened?”
- Surveys: Help you collect structured responses and compare them. Test the questions on a small sample first. Vague questions or incomplete answer options produce numbers that are hard to interpret.
- Messages and comments: Contain people's language and questions, but the people who write them are not the whole audience. Also listen to people who bought and people who did not complete the process, if you can reach them.
Gather material you are allowed to use, and remove names and contact details if the analysis does not need them. Keep the date and source for each part, and check the transcript against the recording if a word could change the meaning.
Have it organize evidence before suggesting content
In our example, we prepared a file of numbered hypothetical responses: R1, “Workshops happen while I am at work”; R2, “I do not know if the level is right for me”; and R3, “I need to learn using spreadsheets like mine.” Now you have material the assistant can classify instead of imagining the audience's motives.
Ask it to give you the supporting text, response number, possible interpretation, and remaining gaps for each observation. “Response R1 mentioned a scheduling conflict” is an observation. “All employees want recordings” is a conclusion far beyond the evidence we have.
If you count frequency, clarify whether you are counting people or responses, and whether the same person expressed the idea more than once. If a question allows multiple answers, the percentages may add up to more than 100%. Have it explain the counting method, and check some results against the original responses yourself. It might classify a negative response as agreement or merge two different objections.
You can follow the steps in whichever assistant you use. Open the tool and upload the inputs listed beside the prompt. File, search, and connection support varies by account. If there is no connection to the source, provide the material yourself.
Copy the prompt as it is, or tailor it with your details first.
Tailor the prompt to your workOptional: add your details and the prompt adjusts itself
View full prompt
Create a useful audience description from what you learned
A Buyer Persona is a description that helps you understand someone with a particular need: what they are trying to do, what gets in their way, what they ask about, and what they need to know to benefit from your service. You do not have to start with a fictional name, picture, and lots of interests unrelated to the work.
Our initial description might be: “An employee who regularly works with spreadsheets, wants training they can connect to their tasks, and still needs to check the level and schedule.” Each part must stay linked to its supporting evidence. If we have no evidence about age or income, we leave them unknown.
If you sell to companies, you may also need an ICP, meaning the characteristics of the company best suited to the service: its business type, team size, or the problem you address. Describing the company differs from describing the person who will use the training or approve its purchase. Ask about that difference if it would change the message.
The assistant can suggest hypotheses about an audience you have not yet met, provided it labels them as hypotheses and suggests how to test them. Keep the description open to updates as new information comes in.
Look at alternatives without confusing research with copying
Choose alternatives that actually serve the same need. For the Excel workshop, there is another trainer, free content, a helpful colleague, or putting off learning altogether. A competitor does not have to be a page offering exactly the same thing.
Compare using clear questions: who are they teaching? What do they explain about the level? Is there a practical example? What questions do people still ask? Keep the link and date accessed, and separate the description on the page from your opinion of it.
Watching a successful video may draw your attention to an important question, but it is not enough evidence of demand or people's willingness to pay.
Turn an observation into something to test
After classifying the responses, you might suggest that the workshop page needs to clarify the level with a sample exercise. That is a suggestion we can test. We cannot yet say that adding the sample will increase sign-ups by a particular percentage.
Ask the assistant to suggest a follow-up question or small experiment for each hypothesis. In our example, ask an interested person, “What would you need to see to know that the level is right for you?”, or show a sample lesson and see whether questions about the level decrease. Decide what to monitor before the experiment, and record any other changes during the same period.
Copy the prompt as it is, or tailor it with your details first.
Tailor the prompt to your workOptional: add your details and the prompt adjusts itself
View full prompt
Choose one question about your audience
Prepare a small set of messages or responses you are allowed to use, and classify them using the first prompt. Review the evidence yourself and come away with one supported observation and one hypothesis that still needs testing. You do not need to describe the entire market in one session.