Gains:
- Ability to produce evidence-based persona drafts based on research data with artificial intelligence and eliminate stereotypes and clichés
- Ability to draft the stages, actions, thoughts, emotions and pain points of the user journey map with the support of artificial intelligence
- Ability to flag assumptions that are not based on data by testing made-up personas with real segment data
The persona and user journey map translate the insights from the research into tangible tools that the team will look at throughout the day. A persona is a fictional but data-driven profile that represents a real user segment: their goals, their barriers, their context. A user journey map is a visual narrative that shows step by step what a user does, what he thinks, what he feels, and where he gets stuck while achieving a goal. AI produces drafts of both in minutes. But herein lies the most dangerous trap: artificial intelligence can create "real"-sounding personas without data. This unit teaches you to stick to evidence without sacrificing speed.
Persona: must be based on evidence, not fantasy
The only difference between a good persona and a bad persona is the attitude. The bad persona puts the assumptions in the designer's mind into a beautiful template: "Ayşe, 34 years old, good with technology, loves practical solutions." If none of these sentences are based on research, the persona is just a decoration and legitimizes wrong decisions.
A good persona bases every claim on research. Behind the sentence "Users are stuck in the installation" is the data "5 out of 8 participants were not able to complete the initial installation". When working with artificial intelligence, the rule does not change: give the model the research summary as a source and ask it to create a persona only from this source, not to add features that have no basis.
Caution: AI is very capable of “filling in the gaps”; It may spontaneously make up details such as age, income, and habits that you do not provide. If these details do not come from the data, the persona is not reliable.
Eliminating stereotypes and bias
Personas are where AI bias most visibly leaks out. The model moves stereotypes from the training data into profiles: “older user is afraid of technology,” “young user is impatient,” behaviors attributed to specific professions or gender. These generalizations are both wrong and ethically objectionable.
The purification method has three steps:
- Mark: Scan each generalization sentence in the output ("everyone", "usually", "people this age").
- Test: Does this sentence have an equivalent in the research data?
- Correct or remove: If there is a basis for it, attach it to the evidence, otherwise delete it. Don't keep the cliché by "softening it"; If there is no proof, he leaves.
Persona expression
problem
Evidence based correction
"Older users are afraid of technology"
Age based stereotype
"5 out of 6 participants expected additional assurance text with the new feature"
"Young users are impatient"
Generational generalization
"Participants abandoned the flow beyond 3 steps (regardless of age)"
"Women want simplicity"
gender assumption
"Simple flow was preferred in all segments"
"His favorite color is blue"
Ornamental information (does not affect decision)
(Subtract — does not affect design decision)
Anatomy of the user journey map
A journey map typically consists of lines:
- Phase: Major periods such as awareness, evaluation, use and return.
- Action: What the user does at each stage.
- Thought: What goes through your mind.
- Emotion: What does one feel; It is usually drawn with a curve.
- Pain point: Where does one get stuck, get angry, give up.
- Opportunity: Where design can intervene.
Artificial intelligence quickly fills in this skeleton. But the “emotion” line is particularly risky: the model reasonably predicts what the user is feeling, but that prediction is guesswork unless supported by real data. Mark unsubstantiated emotions on the map as assumptions; Get “verified” status when confirmed by research citation.
three mini cases
Case 1 — Quick draft, real fix. One team produced 3 persona sketches with AI from 12 interview summaries; The job took 25 minutes. Then he checked 9 characteristics of each persona against the research data; 4 features had no basis and were removed. The remaining personas were both fast and reliable.
Case 2 — Leaked cliché. “Retired users do not trust digital banking,” the AI persona wrote. The team looked at the data: 5 out of 6 retired participants used regular digital banking. The sentence was completely incorrect and was removed; replaced with an evidence-based observation such as "additional assurance text pending on new features." Lesson: age-based stereotypes are disproved by data.
Case 3 — The sentiment curve is exaggerated. The model wrote that the user was "furious" at checkout. In the actual recordings, users were just "hesitant." The designer fixed the sentiment; because "anger" and "hesitation" require different solutions (one is trust, the other is clarity). Lesson: the emotion line is calibrated with evidence.
How many personas, how much detail?
A common trap that new designers fall into is producing too many and overly detailed personas. AI magnifies this trap: as many personas as you want, with as much detail as you want, arrive in seconds. However, what is valuable is little but solid. For most products, 2-3 primary personas are sufficient; each must represent a real user segment and be able to differentiate design decisions. If two personas behave almost the same, it is probably one persona. On the detail side, the criterion is this: every information added to the persona is valuable if it affects a design decision; If it does not impress (for example, the user's favorite color), it is ornamental and should be removed. When requesting a persona from artificial intelligence, saying "only add features that affect the design decision" is the most practical way to get rid of this trappings.
Tip: Test a persona feature: "Will my design change if I change this?" If the answer is no, that feature is unnecessary in the persona.
Copiable prompts
Your role: UX research synthesizer.Source: Anonymized research summary below.Task: Create up to 2 persona sketches from this source.Rule: Link each feature to a finding in the source and write the basis in parentheses.DO NOT ADD any features (age, income, habits) that have no counterpart in the source.Leave unknown fields as "no data".Source: <<summary>>
Check this persona outline for bias. Mark each sentence that contains a generalization, cliché, or unsubstantiated assumption, write why it is questionable, and indicate whether it is supported in the source data. Persona: <<text>> Source data: <<summary>>
Generate a draft user journey map from the persona and research brief below.Lines: Stage | Action | Thought | Emotion | Pain point | Opportunity.Add a citation to each "Emotion" and "Painpoint" cell if there is a source for it; otherwise, mark the cell with "[ASSUMPTION]".Persona: <<x>> Summary: <<y>>
List the cells marked [ASSUMPTION] on this journey map. For each, write what additional research question would confirm this assumption. Map: <<table>>
Weak prompt / Strong prompt
Weak: "Create an e-commerce user persona."
The result: a profile that is completely made up, irrelevant to data, and full of clichés.
Strong: "Extract no more than 2 personas from this research summary; tie each feature to a finding; do not add any details that are not in the source; leave the unknown as 'no data'."
The result: a defensible persona based on evidence, with gaps honestly pointed out.
Difference: strong prompt brings source obligation + prohibition of fabrication + marking the unknown.
Common mistakes
- Creating personas without data. Telling AI to “make a persona” gives false legitimacy to your assumptions.
- Keeping the cliche by softening it. Saying "a little cautious" doesn't correct the stereotype; If there is no evidence, it should be removed.
- Mistaking the emotion line for reality. The model predicts emotion; Emotion without evidence is assumption.
- Not updating the persona at all. The persona should also be updated as new research becomes available; frozen persona is misleading.
- Creating too many personas. 5-6 persona paralyzes the team; Keep a little, but hold on to what's firm.
In summary
Persona and journey map are powerful tools that translate research into the team's common language; but their power comes entirely from the data on which they are based. AI produces a blueprint for these tools in minutes, whereas the real work is connecting each claim to the source, flagging clichés and testing them against data, and isolating unsubstantiated sentiment/pain points as assumptions. A persona that honestly leaves the gaps as "no data" is much more valuable than a "realistic" persona full of made-up details.
Application task
- Generate 2 persona drafts with the first prompt from a research brief (real or fictional).
- Check the personas for bias with the second prompt; Test each marked sentence with data.
- Remove or make “no data” any flimsy features.
- With the third prompt, draft a journey map from a persona.
- List the [ASSUMPTION] cells in the map and write a verification question for each.
checklist
- [ ] I based the persona on the research summary, I did not create it from imagination.
- [ ] I did not add any features to the persona that were not in the source.
- [ ] I marked the clichés and generalizations and tested them with the data.
- [ ] I confirmed the feelings and pain points with evidence, and marked the unsupported ones.
- [ ] I kept the number of personas manageable.
- [ ] I prepared verification questions for the assumptions.