--- name: persona-development description: > Builds personas in which every attribute is either traceable to evidence or explicitly labelled as interpretation, with invented biography, fabricated quotes, stock photographs and decorative detail prohibited. Use for "build personas", "create user personas", "turn the segments into personas", "bring the audience to life", "we need personas for the team", "are these personas real", "how many personas should we have", "our personas are out of date", "make the research feel human". category: 09 Segmentation and Audience Understanding ref: 09.02 tier: 1 inherits: [K2, K3, K4, K5] --- # Persona Development ## 1. One-line description Turns segmentation output or qualitative evidence into personas that make an audience usable in day-to-day decisions, under a strict rule that every attribute is either traceable to a named source with a base or visibly labelled as the researcher's interpretation, and that a thin evidence base produces a shorter persona rather than a richer one. ## 2. What this skill is used for **The research problem it solves.** A persona is a compression device. It exists so that a product manager choosing between two features, or a copywriter choosing between two headlines, can hold an audience in mind without re-reading a 90-page report. When it works it is one of the most useful artefacts research produces. It fails in a specific and well-documented way: the format invites colour. A persona with a name, a face, a morning routine and a favourite coffee feels more real than one without, so the colour gets added, and almost none of it was measured. What arrives on the wall is then a mixture of evidence and invention with no visible seam, and downstream users have no way to tell which is which. Six months later the team is designing for a fictional person's fictional commute. This is not a presentational quibble: it is fabrication under K4 §2.2, and it is the single likeliest thing to be produced when a persona template is handed to an AI system, because the template has slots and the model fills them fluently. This skill supplies the evidence annotation that makes the seam visible, the list of elements that earn their place, the prohibition on the ones that do not, and the honest response when the evidence will not fill the template. **Where it sits.** Synthesis and activation. It takes validated segments from segmentation work, or themes and needs from qualitative work, and hands a usable audience representation to design, product, communications and reporting. **Typical use cases.** - Converting a validated segmentation into personas a business can work with day to day. - Building personas from depth interviews or ethnography where no quantitative segmentation exists. - Auditing existing personas that stakeholders distrust, or that contain claims nobody can source. - Reducing an unwieldy set of personas to the number that are genuinely distinguishable. - Refreshing personas whose underlying evidence is several years old. - Producing a persona set for a design or service team who will use it in daily decisions rather than in a deck. **Who uses it.** Qualitative and mixed-method researchers producing persona deliverables; insight managers asked for "something the team can use"; UX and service designers who inherit personas and need to know which parts are evidence; brand and communications planners; and anyone reviewing AI-generated personas, where invented biography arrives by default and is indistinguishable from research at the point of reading. ## 3. When to use it - A validated segmentation exists and the segments need to become usable in daily decisions. - Rich qualitative evidence exists about distinct types of user, and the team needs a shared reference for them. - Multiple teams are making decisions about the same audience and describing that audience differently. - Existing personas are in circulation and their evidence base is unknown or disputed. - A design or service team keeps designing for themselves, and needs a specific, evidenced alternative in the room. - Personas exist but nobody uses them, and the reason may be that they contain nothing decision-relevant. - You need to reduce the number of personas to the point where they are genuinely distinguishable. - An AI system or an agency has produced personas, and someone has to check what in them is real. ## 4. When NOT to use it - **There is no evidence base.** This is the hard stop. Personas cannot be developed from a category, a brief, a stakeholder workshop or general knowledge of a market. What that produces is a fictional character, and because the format is identical to an evidenced persona, nobody downstream can tell. If the request is for personas and no research exists, say so and offer either an explicitly labelled set of hypotheses for testing, or a research design, per **01.04 Research Method Selection**. **A hypothesis persona is a legitimate artefact provided it is labelled as one on every page.** An unlabelled one is fabrication. - **The segmentation the personas would sit on has not been validated.** A persona inherits every weakness of the solution beneath it and then hides it behind a name and a face. If the segmentation failed stability testing, or the segments differ only on the variables used to build them, do not personify them. Go back to **09.01 Audience Segmentation**. Personifying an unstable solution is the most efficient way to make an artefact of the algorithm permanent. - **The audience does not divide into distinguishable types.** Where the evidence describes one population varying by degree, the honest output is a description of that population, its range, and the dimensions on which it varies. Producing four personas from a homogeneous audience manufactures the differences that the segmentation work refused to invent. - **The question is really what people do, not who they are.** Behaviour, repertoire, frequency and occasion are **09.03 Behavioural Profiling**, and they are frequently more useful than a persona for the decision at hand. A persona is a person-level construct, so it is the wrong container for behaviour that varies by occasion within the same person. - **The question is really what people need.** Needs, motivations, barriers and jobs to be done are **09.05**. A persona is a good vehicle for carrying a need to a team; it is a poor instrument for discovering one, and building personas first tends to freeze a needs framing before it has been tested. - **The purpose is to make a deck feel human.** A persona is a decision aid. Where the request is for warmth rather than for use, the honest answers are a real quote with its participant identifier per **07.04**, a photograph clearly credited as stock and not presented as a participant, or a short evidenced audience description. Colour added to satisfy an aesthetic requirement is invention, and per K4 §9 the response is to offer the strongest honest alternative rather than to comply. - **Participants could be identified.** In small samples, niche B2B audiences and sensitive topics, the combination of role, employer type, condition and location that makes a persona vivid can also make an individual recognisable. That is **13.05 Research Ethics and Consent Design** before it is a synthesis question, and it is a real risk in specialist categories where the population is small. - **The evidence is more than a couple of years old in a category that has moved.** Personas decay silently, because the artefact carries no date on the wall. Where the underlying study predates a material change in the category, the technology or the population, the honest output is a statement that the persona set needs re-evidencing, not a refreshed layout. ## 5. Required inputs **Required.** Without these the skill cannot run. If absent, ask. If the first is absent, stop. - **A real evidence base about real people.** Either a validated segmentation with its profile data and bases, or a qualitative corpus with participant identifiers, or ideally both. **If none exists, do not proceed.** Say what is missing, and offer a labelled hypothesis set or a research design instead. - **Source references for every attribute you intend to include**, per K2 §4: the question and base for a quantitative claim, the participant identifier and prevalence for a qualitative one. If a claim cannot carry a reference, it cannot enter the persona. - **The decisions the personas will inform, and who makes them.** This determines which elements earn their place. A persona for a service redesign team needs different content from one for a media planner, and a persona built without a decision fills its space with biography. - **The validation status of the underlying segmentation**, where one exists: stability results, external profile, and typing tool accuracy per **09.01**. This sets the confidence the persona can carry. **Optional, and what each one adds.** - **Behavioural or transactional data linked to the segment or the participants:** replaces claimed behaviour with observed behaviour, which is the strongest content a persona can carry and the content most often faked. - **Verbatim material with participant identifiers:** allows real quotes, which are the only quotes permitted. Without it, the persona carries no quotes at all rather than paraphrases dressed as speech. - **Journey, service or usage-context evidence:** supplies the context element honestly, so that "where and when this person encounters the service" is observed rather than imagined. - **Prevalence data across the whole population:** allows each persona to carry its size, which is what stops a vivid minority persona dominating a roadmap. - **The existing persona set:** allows continuity, and allows the specific unsourced claims in circulation to be identified and retired by name. - **Accessibility, assistive technology or language-need data:** where relevant, prevents a persona set that silently represents only the majority user. ## 6. Questions to ask before starting 1. **What decisions will these personas be used to make, by whom, and how often?** Determines which elements earn their place and which are decoration. *Default if unanswered:* build on goal, job to be done, context, barrier, decision criteria, current behaviour and unmet need, and omit everything else, flagging that relevance could not be assessed per K5 §2.1. 2. **What is the evidence base, and what is its status?** Determines the confidence each persona carries and whether it is a persona or a labelled hypothesis. *Default:* state the base and status on the face of every persona. 3. **Is the underlying segmentation validated, and how accurate is its typing tool?** Determines whether a persona can claim to represent a real group, and how confidently anyone can say which persona a given customer is. *Default:* if validation status is unknown, label the personas as unvalidated and say so on the artefact. 4. **How many groups are genuinely distinguishable in the evidence?** Determines the number of personas, which is a property of the evidence rather than of the layout. *Default:* build only the personas that differ on something decision-relevant, and report how many were requested and how many the evidence supported. 5. **Will these personas be used outside the team that commissioned them?** Determines how self-contained the evidence annotation must be, because a persona travels without its report. *Default:* assume it travels, and annotate accordingly. 6. **Is there any risk of identifying a participant?** Determines whether contextual detail must be generalised or removed. *Default:* generalise employer, location and role detail in any sample where the population is small. 7. **When will these be re-evidenced?** Determines the currency statement. *Default:* put a "evidence collected" date and a re-validation trigger on the face of every persona. ## 7. Step-by-step methodology **The position this method takes.** A persona is a communication device built on evidence. Both halves are binding. If it is not a communication device it is a data table, and nobody will use it. If it is not built on evidence it is a character, and using it is worse than using nothing, because it carries the authority of research. The discipline that keeps both halves true is a single rule applied without exception: **every attribute in a persona is either traceable to a source or visibly labelled as interpretation.** When the evidence runs out, the persona gets shorter. It never gets richer. **1. Establish the base and its status before writing anything.** Record what each persona will rest on: which segment, at what size, with what typing accuracy; or which participants, how many, recruited how. Write the confidence per K3 §3 at this point, not at the end, because it constrains what may be written. A persona built on a validated segmentation with linked behavioural data can carry declarative statements. One built on eight depth interviews carries a described pattern with its prevalence, and nothing more. **A persona built on a segmentation that failed stability testing is not built.** *Correct result:* a one-line provenance statement per persona that will appear on the face of the finished artefact. **2. Decide the element set from the decision, not from the template.** Seven elements earn their place in almost every case, because each one changes a decision: the **goal** the person is pursuing in this context; the **job to be done**, stated as the progress they are trying to make; the **context** in which they encounter the category, including constraints of time, money, place and capability; the **barrier** that stops or slows them; the **decision criteria** they actually apply when choosing; their **current behaviour**, ideally observed; and their **unmet need**. Everything else must justify itself against a named decision. Favourite coffee, car, weekend hobby, brand of phone, a morning routine and a personality descriptor are decoration unless the category makes them load-bearing, and the test is specific: **would a different value on this attribute change any decision the persona is used for?** In a coffee retail study the coffee preference is evidence; in a banking study it is filler. *Correct result:* a written element list with a decision justification for anything beyond the seven. **3. Extract each element from evidence, one at a time, with its reference attached.** Work element by element across the whole persona set rather than persona by persona, because writing one persona at a time invites the writer to complete a character. For each element and each persona, find the evidence, record the source in K2 §4 format, and record the strength: how many participants, what base, whether observed or claimed. Where an element has no evidence for a given persona, write "not established" and move on. That phrase appearing three times in a persona is information: it says the evidence base is thinner than the format implies. *Correct result:* an evidence grid with elements as rows, personas as columns, and a source reference or an explicit gap in every cell. **4. Separate evidence from interpretation, visibly, per K2 §3.** Some of the most useful content in a persona is interpretive: why the barrier exists, what the behaviour suggests about priorities, what the person is likely to do next. Interpretation is permitted and often necessary. It is not permitted to look like observation. Mark it, either with an explicit label or with the signal language in K2 §3.2, and give it a confidence level. A persona in which the reader cannot tell the measured from the inferred has failed, however accurate the inferences turn out to be. *Correct result:* every interpretive statement carrying a signal word or a tag, and a confidence level attached to each. **5. Apply the fabrication prohibitions without negotiation.** These are specific, and each one is common. **No invented biography:** no age unless measured, no job title unless measured, no family composition, no home circumstances, no daily routine, no life history. **No invented names attached to real material:** a real quote from participant P07 does not become a quote from "Sarah". Naming a persona is acceptable as a label ("Persona B: The Deadline-Driven Applicant"), but the moment a name is attached to verbatim material, a real person's words have been reassigned to a fictional one, which is misattribution under K4 §2.3. **No fabricated verbatim:** if no participant said something that captures the point, the persona does not have a quote for that point. A composite sentence in quotation marks is a fabricated quote regardless of how faithfully it summarises. **No stock photography presented as a participant:** an image may appear only if it is clearly labelled as an illustration and could not be read as a photograph of a research participant. **No invented numbers:** a persona that says "spends around 40 minutes a week" when nothing measured that has produced a statistic from nothing, and it will be quoted. There is an honest alternative to each of these, and offering it is the correct response under K4 §9 rather than refusing flat. Where a theme has no single quote that carries it, offer the closest real quote with its identifier, or a short summary explicitly attributed to the analyst rather than to a participant, in ordinary prose without quotation marks. Where a face is wanted, offer an abstract mark or an illustration that could not be read as a photograph of a participant. Where a name is wanted, offer a label describing the defining difference. Where a routine is wanted, offer the observed context evidence that actually exists, which is usually more specific and more persuasive than an invented one. In practice teams accept these readily, because the real material is better than what they asked for. *Correct result:* a persona set in which every sentence could be defended by pointing at source material. **6. Attach the evidence annotation to the face of the artefact, not to an appendix.** Each claim carries its source and base in compact form: `[Q14, segment C, n=214]`, `[9 of 14 participants]`, `[observed, transaction data, n=1,610]`. This is the step that is always cut for design reasons and always regretted, because a persona travels without its report and arrives on a wall where nobody remembers what was measured. Where space is genuinely constrained, use a two-layer artefact: a clean face and a companion evidence sheet with the same claim numbering, and make the numbering visible on the face. *Correct result:* a reader who has never seen the study can tell, for any statement on the persona, where it came from and how many people it rests on. **7. Decide the number of personas by distinguishability, not by layout.** Two tests. **The substitution test:** take a claim from persona A and put it in persona B. If it still reads as true, those personas are not distinguishable on that element. Run it across all seven elements; personas that survive substitution on fewer than two or three elements should be merged. **The decision test:** would the team make a different choice for A than for B? If not, the distinction costs attention and buys nothing. Three to five personas is the usual working range, not because of a rule but because beyond that, users stop distinguishing them and default to the one they remember. Report the number requested, the number the evidence supported, and the merges you made. *Correct result:* a set in which every persona survives both tests, plus a record of what was merged and why. **8. Size each persona and put the size on the artefact.** A persona without a size invites the team to treat all personas as equally important, and the vivid minority persona reliably wins that competition. Where a validated segmentation exists, use its size in population and in value. Where the base is qualitative, do not invent a percentage: state the prevalence honestly as a count of participants, and say explicitly that qualitative evidence does not support a population estimate, per K4 §3.3. *Correct result:* a size statement on every persona, in the strongest currency the evidence supports, with its limitation stated where it is a count rather than a proportion. **9. Include the counter-evidence and the variation within the persona.** Every persona compresses a group whose members differ, and the compression is where the damage happens: a team designs for the persona and misses the third of the segment who behave differently. Add two lines to each persona: **what varies within this group** on the elements that matter, and **what evidence cut against this characterisation**. This is a small addition that prevents the largest downstream error, and it is never present in a fabricated persona, which makes it a useful audit signal too. *Correct result:* every persona carrying its own internal range and any contradicting evidence. **10. Write what the persona does not tell you.** Name the questions this persona cannot answer: elements not established, decisions it should not be used for, and populations it does not cover. Teams use personas beyond their evidence because nothing on the artefact says where the evidence stops. *Correct result:* a short "not established" block on every persona, which is also the brief for the next study. **11. Date it, and set the re-validation trigger.** Put the fieldwork date on the face of the persona, not the design date. Name the trigger that should force a refresh: a category shift, a product change, a channel change, or a fixed interval. Personas decay silently because the artefact looks the same on the wall in year four as it did in year one. *Correct result:* a date and a trigger visible on every persona. **12. Mark the judgement points and route them, per K5.** Which persona the business should prioritise is a materiality judgement (§2.1). Whether the characterisation reads correctly in each market and language is a cultural one (§2.2). Whether a vivid persona will be used responsibly by teams who did not commission it is a professional judgement about the artefact, and it belongs to the named researcher whose work it is (§2.8). *Correct result:* markers placed at the point of the judgement, each naming what changes depending on the answer. ## 8. Analytical framework The chain for every statement in a persona: Element → Evidence → Source and base → Level (observed or interpreted) → Confidence → Placement on the artefact **Element.** One of the seven that earn their place, or one justified against a named decision. **Evidence.** The specific material behind it, not the general impression. **Source and base.** In K2 §4 format, attached to the claim, on the face of the artefact. **Level.** Observed or interpreted, marked so a reader can tell, per K2 §3.1. **Confidence.** Per K3, set by the base and by whether the behaviour was observed or claimed. **Placement.** On the persona, on the companion evidence sheet, or nowhere. There is no fourth option, and "nowhere" is used often. **The earn-its-place test.** Every candidate element is assessed on one question: *would a different value on this attribute change a decision the persona is used for?* | Element | Usually earns its place because | Usually decoration because | |---|---|---| | Goal | Determines what the product or service is being judged against | | | Job to be done | Stable across solutions, so it survives a product change | | | Context and constraints | Determines what is feasible for this person | | | Barrier | Names the thing to remove, which is usually the intervention | | | Decision criteria | Determines the message, the comparison and the proposition | | | Current behaviour | The baseline any change is measured against | | | Unmet need | The opportunity, and the reason the persona exists | | | Name and face | | Adds recall, adds no decision value, and invites invention | | Age, family, home life | | Rarely changes a decision, almost never measured, always assumed | | Daily routine | | Pure narrative unless the category is time-structured | | Favourite brands, hobbies, car | | Decoration unless the category makes them load-bearing | | Personality descriptor | | Not measured by most studies, and reads as characterisation | **The two confidence classes.** A persona built on a **validated segmentation** carries population claims with sizes and tested differences, and can use declarative language per K3 §4.1 where the base supports it. A persona built on **qualitative work** carries a described pattern with participant counts, uses the moderate or low-confidence registers in K3 §4.2 and §4.3, and does not carry percentages. A persona set built from both should say, per element, which stream the claim came from, because the two do not carry the same weight and a mixed artefact silently promotes the weaker one. ## 9. Output format **A. Persona face** (one page or one card per persona) ``` PERSONA LABEL A description of the defining difference, not a first name SIZE Population share and value share, or participant count with an explicit note that qualitative evidence carries no population estimate BASED ON Segment and validation status, or participants and how recruited EVIDENCE DATE Fieldwork date, and the re-validation trigger GOAL [source, base] JOB TO BE DONE [source, base] CONTEXT [source, base] CURRENT BEHAVIOUR [source, base; observed or claimed, stated] DECISION CRITERIA [source, base] BARRIER [source, base] UNMET NEED [source, base] INTERPRETATION Clearly separated, with confidence per K3 VARIATION WITHIN What differs among people in this group COUNTER-EVIDENCE What cut against this characterisation, or "none found" NOT ESTABLISHED Elements with no evidence, and decisions this should not inform QUOTES Real verbatim only, with participant identifiers. Omitted entirely where no suitable real quote exists ``` **B. Evidence sheet**, one per persona, with the same claim numbering as the face, giving the full source reference and base for each statement. **C. Distinguishability record.** The substitution test and the decision test, per element, with any merges and the reason for each. **D. Persona set summary table.** | Persona | Size (population) | Size (value) | Defining difference | Base and status | Confidence | Key decision it informs | |---|---|---|---|---|---|---| **E. What the set does not cover.** Populations, occasions and needs not represented, and any group deliberately excluded. **When the evidence is thin.** The persona gets shorter. A persona with four evidenced elements and three marked "not established" is an honest artefact and a useful one; the same persona with those three gaps filled by plausible invention is a fabrication that will outlive the study, per K4 §1. Specifically: no quotes where there are no real quotes, no age where age was not measured, no photograph that could be read as a participant, no percentage on a qualitative base, and no persona at all where the evidence supports a description of one population rather than several. ## 10. Quality checks Run before anything is presented. These sit on top of K4 §8. 1. Can every statement on every persona be pointed at in source material, or is it visibly labelled as interpretation? 2. Is every quote verbatim, with a participant identifier, and is it the quote of a real participant rather than a composite? 3. Has any quote been reattributed from a participant identifier to a persona's name? 4. Does any persona contain an age, occupation, family situation, routine or possession that was not measured? 5. Does any image appear that a reader could mistake for a photograph of a research participant? 6. Does every element carry its source and base on the face of the artefact, not only in an appendix? 7. Is the observed and claimed distinction stated for behaviour, per **09.03**? 8. Does every persona carry its size, and where the base is qualitative, is the absence of a population estimate stated? 9. Has the substitution test been run across all elements, and are the merges recorded? 10. Does every persona name the variation within the group and any counter-evidence? 11. Does every persona carry a "not established" block? 12. Is the fieldwork date on the face of the artefact, with a re-validation trigger? 13. Is the validation status of the underlying segmentation stated, including the typing tool's accuracy? 14. Would a reader who has never seen the report be able to tell which parts of this persona are measurement? 15. Could any participant be identified from the combination of details given? 16. Is any persona present because the layout had a slot rather than because the evidence supported it? ## 11. Common failure modes | Failure | How to recognise it | How to prevent it | |---|---|---| | **Invented biography** (the signature failure) | Age, job, family, commute and hobbies appear, and none of them is in the questionnaire or the transcripts | Element list justified against decisions; the earn-its-place test; every attribute carries a source | | **The fabricated verbatim** | A quotation that captures the theme perfectly and matches no transcript line | Real quotes only, with identifiers. Where none exists, the persona has no quote there, per K4 §2.3 | | **Reattributed quote** | P07's words appear as spoken by "Sarah" | Persona labels are labels. Quotes keep participant identifiers | | **Stock photograph as participant** | A face on the card, uncredited | Either no image, or an image explicitly labelled illustrative | | **Persona on an unvalidated segmentation** | A vivid character sitting on a solution that failed split-half testing | Validation status stated on the face; do not personify a failed solution | | **Evidence-free persona from a workshop** | Personas built in a stakeholder session and presented as research output | Label as hypotheses on every page, or do not produce them | | **The vivid minority** | A small, memorable persona dominating a roadmap | Size on every persona, in population and value | | **Compression damage** | A team designs for the persona and misses the third of the group who differ | The "variation within" line, and the counter-evidence line | | **Too many personas** | Seven personas, three of which the team cannot tell apart | Substitution test and decision test; merge and record | | **Silent decay** | Personas on the wall in year four with no visible date | Fieldwork date and re-validation trigger on the artefact | | **Decoration mistaken for richness** | The persona is a page long and none of it changes a decision | Seven elements plus justified extras; length is not a quality measure | | **AI: template completion** | Every field filled, every persona equally detailed, no gaps anywhere | A uniformly complete persona set is a warning sign. Gaps are normal and are reported as "not established" | | **AI: plausible inner life** | Confident statements about what the person feels, fears or values, with nothing measuring it | Inferred inner states are interpretation, marked and confidence-rated, per K4 §3.4 | ## 12. AI guardrails Skill-specific only. K4 applies in full and is not repeated here. Evidence labelling follows **K2 §3**; confidence language follows **K3 §4**. 1. **Never write an attribute that was not measured.** No age, occupation, household, routine, possession, hobby or personality trait unless a variable or a transcript supports it, and the source is attached. 2. **Never write a quote that no participant said**, and never present a composite or a paraphrase in quotation marks. Where no suitable real quote exists, the persona carries none. 3. **Never attach a persona's name to a real participant's words.** The participant identifier stays with the quote. 4. **Never include an image that could be read as a photograph of a participant.** 5. **Never fill an empty element to complete a template.** Write "not established". A uniformly complete persona set is evidence of invention, not of thoroughness. 6. **Never build a persona on a segmentation whose validation status is unknown** without saying so on the face of the artefact. 7. **Never state a percentage on a qualitative base**, and never convert participant counts into population proportions, per K4 §3.3. 8. **Never present an inferred motivation, emotion or value as observed.** Mark it as interpretation with a confidence level. 9. **Never produce more personas than the evidence distinguishes**, whatever number was requested. Report the number requested and the number supported. 10. **Never omit the size, the date or the "not established" block** for design reasons. If they do not fit, the layout is wrong, not the requirement. 11. **Never carry a persona forward from a previous study into a new one** without re-evidencing it. A persona quoted three years on with no date has become folklore, per K2 §7. ## 13. Best-practice principles - **A persona is a compression of evidence, not an illustration of a hunch.** Its whole value comes from the fact that a reader can trust it without checking; that trust is only warranted if checking is possible. - **When the evidence is thin, the honest output is a shorter persona.** The instinct to compensate for weak evidence with richer detail is precisely backwards, and it is the mechanism by which persona sets become fiction. - **The seam must be visible.** Interpretation belongs in personas. Interpretation that looks like observation does not. - **Personas do not make research human; people do.** Warmth comes from a real quote with a real identifier, from an observed behaviour that surprises, from a barrier stated in the person's own words. It does not come from a name and a face. - **Behaviour beats attitude, and observed beats claimed.** A persona whose behaviour section comes from transaction data is worth several whose behaviour section comes from a frequency question, per **09.03**. - **Name the defining difference, not the personality.** "The Deadline-Driven Applicant" tells a team what distinguishes this group. "Ambitious Amy" invites them to invent the rest. - **Size stops the vivid persona from winning.** The most memorable persona will be over-served unless the artefact says how many people it represents. - **Say what varies inside the group.** A persona is an average with a face, and the people it hides are the ones the design will fail. - **A persona should be able to lose an argument.** If it contains nothing specific enough to be contradicted by the next study, it contains nothing. - **Fewer personas get used.** Beyond four or five, teams stop distinguishing and default to the one they remember, so the extra ones cost attention and change nothing. - **Date everything, because personas decay silently.** The artefact looks identical in year four; the market does not. - **The test of a persona is whether it changed a decision.** Ask six months later. A persona nobody used was either wrong, undistinguished, or full of things that could not be acted on. ## 14. Worked example *Fictional scenario, used to demonstrate method. The organisation, participants, figures and findings below are invented.* **INPUT.** A national debt advice charity wants personas for its service design team, who are redesigning the first contact journey. Evidence available: 22 depth interviews with people who contacted the service in the past six months (P01 to P22), an intake dataset of 4,100 cases with presenting issue, channel and referral source, and a needs analysis from **09.05**. No validated segmentation exists. The brief asks for "five personas, with names and photos, so the team can connect with them". **PROCESS.** *Step 1, provenance.* No segmentation means no population sizes and no percentages. The confidence class is qualitative: described patterns with participant counts. This is written down first, and it immediately rules out three-quarters of what the brief asked for. *Step 2, elements.* The decision is the first contact journey, so the elements that change it are: the trigger that prompted contact, the job to be done at that moment, the constraints on time and privacy, the barrier to making contact, and current behaviour before contact. Favourite brands, family composition and daily routine change nothing in this decision and are excluded, with the reason recorded. *Steps 3 and 4, the evidence grid.* Built element by element across all candidate groups. Two patterns hold strongly: contact triggered by a specific enforcement event (14 of 22) versus contact triggered by a gradual accumulation with no single event (6 of 22, with 2 unclear). The intake data corroborates the split independently, since referral source and channel differ sharply between enforcement-triggered and self-referred cases (n=4,100). That corroboration on data not used to build the grouping raises confidence from low to moderate. *Step 5, the judgement call.* The brief asked for names and photographs. The team genuinely will connect better with them, and refusing outright is unhelpful per K4 §9. The resolution: persona labels describe the defining difference rather than a first name; no photographs of people are used, and an abstract illustration is used instead, so nothing can be read as a participant; and the emotional weight the team wanted is carried by real, identified quotes from the transcripts, which turn out to be far stronger than anything a name would have added. One planned persona had no suitable real quote, so it carries none, and the gap is visible. *Step 7, the number.* Five were requested. The substitution test collapses two of the candidate groups immediately: their goals, barriers and decision criteria are interchangeable, differing only in the amount of debt, which the service does not treat differently at first contact. A fourth candidate rests on 2 participants and is not distinguishable from the enforcement-triggered group. Three personas survive, and the merges are recorded with the substituted claims shown. *Steps 8 to 10.* No population sizes, because the qualitative base cannot support them. Instead each persona carries its participant count and the corroborating intake proportions where the intake data measures the same thing, labelled as intake-recorded rather than as a population estimate. Each carries a "variation within" line: within the enforcement-triggered group, willingness to involve family varies substantially and the journey must not assume it. Each carries a "not established" block, and one of those blocks (whether digital-first contact would be preferred by people with limited privacy at home) becomes the brief for the next study. **OUTPUT.** Three personas, each with provenance, participant counts, elements sourced line by line, interpretation separated and confidence-rated, real quotes with identifiers, internal variation, counter-evidence, a "not established" block, a fieldwork date and a re-validation trigger. A distinguishability record showing the two merges and the dropped candidate. A **researcher review recommended** marker on whether the enforcement-triggered persona should lead the redesign, since that is a materiality judgement about the charity's mission and capacity, not an analytical one, per K5 §2.1. ## 15. Advanced usage **Personas as a live artefact rather than a poster.** Where the underlying evidence is refreshed regularly, keep the personas in a form that shows what changed and when: which elements have been re-evidenced, which are inherited from an older wave, and which have shifted. This solves decay far better than a periodic rebuild, and it makes the age of each claim visible at the level of the claim rather than the document. **Anti-personas and edge cases.** A short, evidenced statement of who the audience is *not*, and of the users at the extremes of capability, access or constraint, prevents the mainstream persona set from quietly defining the design boundary. Where accessibility or language need evidence exists, an edge-case profile is frequently more valuable to a design team than a fourth mainstream persona. **Personas for B2B and multi-stakeholder decisions.** A B2B purchase involves several roles with different jobs, and personifying only the end user misses the person who blocks the purchase. Build one persona per role in the decision unit, and add a line to each on their influence over the decision and what they need in order to say yes. Identification risk is materially higher here, per Section 4. **Combining with journey work.** A persona plus a journey (**06.05**) is more useful than either alone, because the barrier element becomes a location rather than an attribute: not "struggles with the application", but "abandons at the document upload step". Where the two are combined, keep the evidence annotation on both, or the journey inherits the persona's authority without its sourcing. **Auditing an inherited persona set.** Take each statement and try to source it. In most inherited sets, somewhere between a third and two-thirds of the statements have no traceable source. Produce a marked-up version showing sourced, unsourced and contradicted claims, rather than a rebuild: the marked-up version is what persuades the organisation that the rebuild is needed, and it identifies by name the claims currently circulating as fact. ## 16. Skill chain **Recommended previous skills:** - **09.01 Audience Segmentation.** Hands over validated segments with sizes, stability results, external profiles and typing tool accuracy, which set what a persona may claim and how confidently. - **09.03 Behavioural Profiling.** Hands over observed rather than claimed behaviour, and the observed-versus-claimed distinction the persona must carry. - **09.05 Needs, Motivation and Jobs-to-be-Done Analysis.** Hands over the goal, job, barrier and unmet need elements with the evidence behind each. - **07.01 Thematic Analysis** and **07.04 Quote and Evidence Extraction.** Hand over themes with prevalence and counter-evidence, and verified verbatim with participant identifiers, which are the only quotes a persona may use. **Recommended next skills:** - **11.05 Research Presentation Development** and **12.02 Research Report Design.** Take the persona set and present it without stripping the evidence annotation, which is the point at which annotation is usually lost. - **14.01 Insight Activation and Socialisation.** Takes personas into the teams who will use them, which is where decay and misuse begin. - **08.03 Implication Development.** Takes persona-level needs and works out what they mean for the organisation. - **06.05 Customer Experience and Journey Analysis.** Combines personas with journeys so barriers become locations rather than traits. **Runs well alongside:** - **13.03 AI Output Verification**, run against any persona set, with source-tracing of every attribute as its focus. This is the single highest-yield audit in the category. - **13.05 Research Ethics and Consent Design**, wherever the sample is small enough or specialised enough that a vivid persona could identify a participant. - **K2 §4**, for the source-reference format that every persona statement carries onto the face of the artefact. --- A Yazi Supplied Skill and resource.