AI Persona Generator (FREE TOOL)

HubSpot defines a buyer persona as "a research-based profile of your ideal customer", with their goals, challenges and buying behaviour. In ad work it answers one question: who is this ad for, and what will make them stop?

A persona is a person, not a market. "Women 25 to 34 who like wellness" fits a million people and gives an ad nothing to say. "Washes oily hair every morning and is tired of it" gives the ad its first line.

A brand often sells to more than one kind of buyer. Each buys the same product for a different reason, so each needs its own ad.

The tool works from what you type, so its personas are educated guesses. Nielsen Norman Group says proto personas are useful because they make a team's assumptions explicit. The next step is to check them against real buyers.

• Read your reviews. Find the words buyers use for the problem. If no review sounds like a persona, cut it or rewrite it.
• Look at who bought last month. Check your last orders, or ask whoever talks to customers. Which persona do they match?
• Run one ad per persona. Same product, same budget, a different hook for each. The ad that wins tells you which persona is real.

Selzee does this step for every brand it makes ads for. It reads your website and suggests up to 3 personas, each with pains, the outcome they want, objections and what they use today. You pick, edit or add your own in your brand settings.

Then it writes one ad angle for each product, persona and awareness stage, so the same product gets different ads for different people.

Last checked: 2 November 2026. The definitions come from Nielsen Norman Group and HubSpot; the comparison from the two tool pages, read the same day.

• The tidy rules were worked by hand before the code: bullets and numbering stripped, repeats and items with a link or an @ handle dropped, at most 3 items a list, hooks over 140 characters dropped, and a persona with an empty field left out. The code agrees with every case.
• When fewer than 3 complete personas come back, the code asks the model once more. A persona still incomplete after that is left out, and the page says so.
• Ten runs on the "Try an example" input were checked for the rules above, then graded by a second AI model for how specific the personas are, how different they are from each other, and anything invented.
• Typical, boundary and hostile inputs ran through the real page: an empty form, a two word product, a non-English product and an instruction typed into the product box.