User personas are short, fictional profiles of your main customer groups, built from research, that describe what each group is trying to get done, what worries them, and how they decide.
For conversion work they help when they lead to a testable change, like separate paths or messages for groups that really differ, and they do little when they stay in a slide deck.
In our database, the tests that lifted conversions acted on one real difference between groups.
Demographic detail had little to do with it.
This page covers where personas came from, the case against them, a lean persona template built for testing, and the A/B tests that show what happens when teams act on persona research.
MarketingExperiments
Clean CommitBuyer personas: what they are and where they came from
The software designer Alan Cooper introduced personas as a design tool in the late 1990s, in the book The Inmates Are Running the Asylum.
Cooper used them to keep product teams focused on a specific user's goals instead of an "average user" (background on Cooper).
Marketing later borrowed the idea as the "buyer persona," which adds buying triggers, objections and who signs off on a purchase.
In everyday use, "user persona" and "buyer persona" overlap.
A user persona describes someone using the product.
A buyer persona describes someone deciding to pay.
In B2B the user and the buyer are often different people, and your pages may need to speak to both.
Three kinds of persona, by how much research is behind them
Nielsen Norman Group sorts personas into three types by evidence (Laubheimer, 2020):
| Type | Built from | Suggested sample | Use for CRO |
|---|---|---|---|
| Proto persona | Team assumptions, no new research | None | A starting list of hypotheses to check, not facts |
| Qualitative persona | Interviews, usability tests, field studies | 5-30 users | Best fit for most teams: real language, real objections |
| Statistical persona | Qualitative research plus a large survey with clustering | At least 100, ideally 500+ respondents | Sizing segments so you know which one is worth a test |
For conversion work, sizing matters. A persona that describes 3% of your traffic can't move your sitewide conversion rate much, even if the change for that group works.
The case against personas
The strongest criticism comes from researchers at Microsoft.
Christopher Chapman and Russell Milham argued in 2006 that personas are hard to verify, because nobody can say how many real people a persona describes (Chapman and Milham, 2006).
A 2008 follow-up tested this on six survey datasets ranging from 268 to 10,307 respondents.
It found that the share of real people who match a persona-like description falls quickly as attributes are added (Chapman et al., 2008).
That is the problem with the classic persona: "Kathleen, 33, lives in Seattle, two kids, drives a Volvo."
Each extra detail makes the persona more vivid and less representative.
The authors' advice was to check personas against data before assuming they describe real groups.
There's a second, more practical problem. A persona can be accurate and still useless for testing if it doesn't point to anything you'd change on the page.
Many persona documents never get past this point.
What A/B tests show when teams act on personas
Research on personas tells you about their accuracy.
It says nothing about whether they lift conversions.
For that we can look at tests where a team acted on a difference between visitor groups.
Note that these are tests of specific page changes, not of personas as a method.
Persona paths can work when visitors arrive with different needs. A large healthcare company, tested by MarketingExperiments (MECLABS), replaced a single-objective homepage with separate paths for different visitor personas, each leading to the same "find a treatment center" page.
Clickthrough to that page rose 52% (healthcare homepage persona paths).
POSist split its homepage into paths for restaurants switching point-of-sale systems and first-time buyers, together with logos and testimonials.
Contact Us visits rose about 5% in a one-week test, and since several changes shipped together, the paths can't take all the credit (POSist segment paths).
The obvious persona split isn't always the right one. Animoto showed industry-specific sample videos (real estate, photography, online retail) on a landing page, which is the textbook persona approach.
A variant with use-case videos that fit any industry (company overview, product description, customer testimonial) produced a double-digit lift in conversions (Animoto use-case videos).
The job visitors were doing mattered more than their industry, which fits the jobs-to-be-done view.
Engagement lifts don't always reach revenue. The Global Leadership Summit tested a landing page that focused on each visitor's local host site and asked them to pick their location first.
Clickthrough rose 32%, but registrations didn't change (localized landing page).
Abila tailored its homepage banner to customers, partners and prospects and reported 171% more time on site.
That was a before-after comparison of engagement metrics, not leads (Abila banner segmentation).
Splitting by visitor type can help one device and hurt another. Clean Commit gave new visitors an educational homepage and returning visitors an offer-led one.
Overall the result was flat: +0.5% conversion rate and +6.7% revenue per visitor, with a 58.6% probability on revenue per visitor.
Underneath that, mobile conversion rose 17.7% and desktop fell 24.4% (Clean Commit new vs returning homepage).
Letting visitors pick their own persona is often simpler than guessing it. DRIP Agency had mobile shoppers of a beauty-tech brand choose a priority (application time, value for money or overall package) and highlighted the matching product.
Purchases rose 6.5% (priority product selector). This is a persona that visitors choose for themselves, which removes the targeting guesswork.
Our product quiz tests follow the same idea.
A persona template built for testing
Most persona templates ask for a name, a stock photo, age, income, hobbies and a favorite brand.
For conversion work, cut most of that. Start with what the person is trying to do and what stops them.
The template below keeps only fields that can turn into a page change or a targeting rule.
| Field | What to write | Where the evidence comes from |
|---|---|---|
| Segment name | A plain label tied to behavior, like "switching from a competitor," not a made-up name | Interviews, survey clusters |
| Share of traffic or revenue | A number, even a rough one | Analytics, survey weighting |
| How to spot them on the site | Traffic source, landing page, new vs returning, a quiz answer, a self-selected path | Analytics, ad data |
| Job to be done | The progress they want, in their words | Customer interviews |
| Top objections | The two or three reasons they almost didn't buy | Post-purchase and exit surveys |
| Words they use | Exact phrases to reuse in headlines and CTAs | Interview transcripts, reviews, support tickets |
| Decision process | Who else signs off, what they compare, how long it takes | Sales calls, interviews |
| Test idea | One change for this group, with the metric it should move | Your team |
The last row matters most. If you can't write a test idea for a persona, the persona isn't ready, or the group doesn't differ enough from the others to need its own treatment.
Write each idea as a proper hypothesis naming the change, the group and the metric.
How to fill it in
- Start with proto personas, labeled as guesses. List the groups your team thinks exist and what you think separates them.
- Interview 5-30 customers. Ask about the moment they started looking, what else they considered and what nearly stopped them. Keep the exact quotes.
- Size the groups. Add one or two screening questions to a post-purchase survey so you can see what share of buyers fall into each group.
- Find the signal on the site. If you can't tell which group a visitor belongs to, let them tell you with a path choice or a quiz step, as in the healthcare and DRIP tests above.
- Run one test per persona and measure the outcome that matters. The Global Leadership Summit result shows why clickthrough alone can mislead.
When personas are the wrong tool
- Low traffic. Splitting visitors into three groups splits your sample too. A persona-specific test needs enough traffic within that group to reach a result.
- Groups that want the same thing. If interviews show every group has the same job and the same objections, one better page beats three tailored ones.
- No way to identify visitors. Targeting rules based on guessed traits often misfire. A self-selection step is more reliable. If you need neither, use plain segmentation by source or device.
- Demographic-only profiles. Age and income rarely change what a page should say. Chapman's research suggests piling on such details mostly makes a persona describe fewer real people.
Personas work in apps too. Applica rebuilt Five Minute Journal's onboarding on jobs-to-be-done research, with personalized content, early testimonials and clearer benefits.
ARPU rose 20% (p = 0.04), trial starts about 10% and trial conversions 14%.
Many changes shipped at once, so the research shaped the variant but can't be isolated as the cause (Five Minute Journal onboarding).
For the full set of research methods that feed a persona, see our customer research guide, or browse all tests to find changes that worked for audiences like yours.