By Phillip Gales
Picture the following scenario: your product team has three buyer personas pinned to the wall. They have names. They have stock photographs. One of them is called "Marketing Mary" and enjoys yoga. Nobody can remember when they were last updated, and everyone suspects they are wrong. The personas were created two years ago in a workshop facilitated by an external consultant. The consultant interviewed six customers, synthesised the findings into three archetypes, and delivered a slide deck that was enthusiastically received, pinned to the wall, and never validated against a single data point since.
This is not an unusual situation. It is, in fact, the norm. A 2023 Forrester study found that fewer than 30% of B2B companies regularly update their buyer personas, and a separate survey by the Content Marketing Institute revealed that while 77% of marketers claim to use personas, only 36% consider them "very useful." The gap between those two figures is the gap between having personas and having good ones.
The problem is not that personas are a bad idea. They are, in principle, an excellent one. A well-constructed persona aligns product, marketing, sales, and customer success around a shared understanding of who they serve. It informs messaging, feature prioritisation, content strategy, pricing, and channel selection. The problem is that most personas are fictional composites. They are invented in conference rooms by people who already think they know the answer, and the "research" backing them consists of a handful of anecdotal interviews with customers who were selected because they were available, not because they were representative.
Ditto offers something qualitatively different. The platform maintains a population of over 300,000 synthetic people, grounded in census data, each with demographic profiles, psychographic attributes, behavioural patterns, and articulated preferences. When you recruit a research group on Ditto, you are not creating fictional composites. You are sampling from a population that mirrors the statistical reality of your target market. The personas are not invented. They are, in a meaningful sense, the respondents themselves.
This article describes how to use Claude Code and Ditto to develop buyer personas from scratch, and, more valuably, how to validate existing personas against real synthetic data. For a companion piece on segmentation methodology, see How to Segment Customers with Claude Code and Ditto. For the broader voice-of-customer framework that underpins this approach, see How to Run Voice of Customer Research with Claude Code and Ditto.
Why Most Buyer Personas Fail
The failure modes of traditional persona development are well documented, yet remarkably persistent.
Small sample bias. The typical persona workshop draws on six to twelve customer interviews. This is not a sample size; it is an anecdote collection. With six interviews, you cannot distinguish between a universal pattern and a coincidence. You cannot identify segments, because you lack the statistical power to observe divergence. You end up with a composite that smooths out the very differences you should be paying attention to.
Selection bias. The customers who agree to participate in persona interviews are not randomly selected. They are the customers your account managers have good relationships with, the ones who respond to emails, the ones who are engaged enough to give you thirty minutes. They are, by definition, your happiest and most engaged users. The frustrated customers, the ones who considered your product and chose a competitor, the ones who churned after two months, are not in the room. Your personas are built on the testimony of people who like you, which is a deeply unreliable foundation for understanding your market.
Confirmation bias. Workshop participants arrive with hypotheses. The product manager believes the primary buyer is a VP of Marketing at a mid-market SaaS company. The sales lead believes it is a CMO at an enterprise. The consultant is hired to resolve this disagreement, conducts the interviews, and produces personas that confirm whichever hypothesis the most senior person in the room already held. This is not malice. It is the ordinary operation of motivated reasoning in a qualitative research context.
Decay. Markets change. Customer needs evolve. New competitors enter. Pricing expectations shift. A persona document created in 2023 describes a market that no longer exists. But because the document is pinned to the wall and referenced in every strategy deck, it continues to inform decisions as though the world has remained static. The persona becomes a monument to a moment that has passed.
Lack of behavioural grounding. The most common persona templates include demographic information (age, title, company size) and "goals and challenges" that read like motivational posters. "Wants to drive ROI." "Struggles with limited budget." These are truisms, not insights. They describe every buyer in every market. A useful persona tells you how someone makes decisions, where they seek information, what language they use to describe their problems, and what would cause them to switch from their current solution. Most persona documents contain none of this.
Ditto does not eliminate all of these failure modes. But it eliminates the two most damaging: small sample bias and selection bias. A Ditto research group of ten synthetic personas, each representing a statistically grounded archetype within your target demographic, provides richer qualitative data than a dozen hand-picked customer interviews. And because the personas are synthetic, you can recruit the hard-to-reach segments: the churned customers, the competitor users, the prospects who never converted. The people your sales team cannot get on the phone are precisely the people Ditto can model.
The Seven-Question Persona Study
The study design below follows a deliberate sequence. Each question moves deeper into the persona's world, from context to behaviour to motivation to aspiration. The order matters because it mirrors the way people actually think about their own decisions: situation first, then action, then reasoning, then ambition.
Q1: Day-in-the-Life
"Walk me through a typical workday. What are the first three things you do, and where does [problem domain] fit into your routine?"
This is the foundation question. It establishes context before asking for opinions. The responses reveal not what people say they care about, but where the problem actually lives in their daily experience. A persona who encounters the problem at 8am while triaging emails is fundamentally different from one who encounters it at 3pm during a quarterly review. The timing, frequency, and emotional context of the problem shape everything downstream: messaging timing, channel selection, urgency framing, and feature prioritisation.
The day-in-the-life question also reveals adjacencies. When a respondent describes their morning routine, they mention the tools they use, the colleagues they interact with, the meetings they attend. These adjacencies are gold for positioning. If your target buyer spends the first hour of every day in Slack and HubSpot, and your product requires them to open a separate dashboard, you have an adoption barrier that no amount of feature superiority will overcome.
Q2: Information Sources
"When you need to learn about a new tool or approach in your field, where do you go first? Who do you trust?"
This question maps the information ecosystem of your persona. It tells you where to distribute content, which influencers to partner with, and which channels to invest in. But it does more than that. The distinction between "where do you go?" and "who do you trust?" separates information discovery from information validation. A buyer might discover your product through a LinkedIn ad but validate it through a peer recommendation. If your marketing only addresses the discovery channel, you are winning the first click and losing the decision.
The trust dimension is particularly valuable. Responses that cite specific publications, podcasts, analysts, or peer communities reveal the credibility architecture of your market. A persona who trusts Gartner operates in a fundamentally different decision framework from one who trusts a subreddit. Your sales enablement materials, your proof points, and your credibility signals need to match the trust architecture of each persona.
Q3: Goals and Barriers
"What are you personally measured on this quarter, and what is the single biggest obstacle preventing you from hitting that target?"
Note the specificity. Not "what are your goals?" but "what are you measured on this quarter?" Not "what challenges do you face?" but "what is the single biggest obstacle?" The narrower framing forces concrete responses rather than aspirational platitudes.
The measurement question reveals the persona's incentive structure. A marketing director measured on pipeline generation has different purchasing criteria from one measured on brand awareness, even if both hold the same title at the same company size. Title-based personas miss this entirely. Incentive-based personas capture it.
The barrier question, framed as a singular obstacle, forces prioritisation. Everyone has twenty problems. The one they name first is the one that keeps them up at night, and it is the one your messaging should address. If multiple respondents name the same barrier, you have found a genuine pain point. If each respondent names a different barrier, you may be looking at multiple segments rather than one persona.
Q4: Channel Preferences
"Think about the last business tool you purchased or recommended. How did you first hear about it, and what was the journey from awareness to decision?"
This reconstructs the buyer journey through retrospective narration rather than hypothetical projection. Asking "how would you evaluate a new tool?" invites aspirational answers. Asking "how did you actually evaluate the last one?" invites honest recollection. The difference is substantial.
The responses reveal the buying process: how many touchpoints, how much time, who else was involved, what nearly killed the deal, and what sealed it. This feeds directly into your go-to-market strategy, your content funnel design, and your sales process. If the typical journey from awareness to decision takes four months and involves three stakeholders, your nurture sequence and sales cycle need to accommodate that reality.
Q5: Decision Criteria
"When choosing between two competing solutions, what are the three factors that matter most to you? And which factor, if missing, would be an absolute dealbreaker?"
The "three factors" framing prevents the common failure of listing every conceivable criterion. Three forces ranking. The dealbreaker question separates table-stakes features from differentiators. If every persona lists "ease of use" as a dealbreaker, that is not a competitive advantage; it is an entry requirement. Your differentiation must come from the factors that vary across respondents.
Cross-referencing Q5 with Q3 (goals and barriers) reveals alignment or misalignment between what people say they want and what they actually need. A persona who says they are measured on pipeline generation (Q3) but lists "reporting depth" as their top decision criterion (Q5) is telling you something important: they need to demonstrate ROI to justify the purchase. Your product does not just need to generate pipeline; it needs to prove it does so, visibly and repeatedly.
Q6: Adoption Behaviour
"How do you typically adopt new tools? Do you trial extensively before committing, rely on peer recommendations, or follow a structured evaluation process? What has caused you to abandon a tool after initially choosing it?"
This question identifies whether your persona is an early adopter, a pragmatist, or a conservative buyer, categories that map directly to Geoffrey Moore's technology adoption lifecycle. The adoption behaviour determines your launch strategy, your onboarding design, and your retention approach.
The abandonment question is equally important. The reasons people leave tell you more about your product requirements than the reasons they arrive. If abandonment is driven by poor onboarding, you have a time-to-value problem. If it is driven by missing integrations, you have a workflow problem. If it is driven by lack of internal adoption, you have a change management problem. Each failure mode demands a different response, and traditional persona documents rarely capture any of them.
Q7: Aspirational Jobs-to-Be-Done
"If you could wave a magic wand and have one capability you do not currently have, what would it be? How would it change the way you work?"
The final question lifts above the current problem and into the aspirational space. Clayton Christensen's jobs-to-be-done framework distinguishes between functional jobs (what the customer needs to accomplish), emotional jobs (how they want to feel), and social jobs (how they want to be perceived). The magic wand framing invites all three.
The responses often reveal opportunities that are invisible in feature requests. A respondent who says "I wish I could prove to my CEO that our marketing actually works" is not asking for a dashboard. They are asking for organisational credibility. A respondent who says "I wish I could test ideas without asking engineering for help" is not asking for a no-code tool. They are asking for autonomy. These aspirational jobs are the foundation of your positioning narrative: not what your product does, but what it makes possible.
Setting Up the Study in Claude Code
The workflow follows a consistent pattern: create a research group that matches your target audience, create a study with the seven questions above, run the study, and extract the insights. Claude Code orchestrates the entire process through Ditto's API.
Step 1: Define the Research Group
The research group is the most consequential decision in the entire process. If you recruit the wrong people, the data is worthless regardless of how good your questions are. Ditto's filters allow you to target by country, age range, gender, and for US respondents, by state.
For a B2B SaaS persona study targeting mid-market marketing directors in the United States, you would instruct Claude Code to create a research group of ten personas with filters for country (USA), age range (30 to 50), and a description that specifies professional context: marketing directors or VPs at companies with 100 to 1,000 employees, responsible for demand generation and brand strategy.
For a consumer product study, you would adjust the filters to match your target demographic. A DTC skincare brand targeting women aged 25 to 40 in the UK would specify country (UK), age range (25 to 40), gender (female), and a description covering skincare purchasing behaviour, income bracket, and lifestyle indicators.
The research group description is critical. It is how Ditto selects personas from its population of 300,000+ synthetic people. Be specific. "Marketing professionals" is too broad. "Senior marketing professionals at mid-market B2B SaaS companies who manage a team of 3-8 people and are responsible for demand generation pipeline targets" is the level of specificity that produces useful personas.
Step 2: Create and Run the Study
Once the research group is recruited, Claude Code creates a study with a title and objective, then asks the seven questions sequentially. Each question is submitted to all ten personas simultaneously, and Ditto returns qualitative responses from each persona. The entire process, from group creation to completed study with AI-generated insights, takes approximately forty-five minutes.
The study objective should frame the purpose without leading the respondents. "Understand the daily workflow, information habits, purchasing criteria, and aspirations of mid-market marketing leaders" is neutral and comprehensive. "Validate that mid-market marketing leaders need our product" is leading and will bias the AI-generated insights.
Step 3: Extract and Analyse
When the study completes, Ditto generates AI-synthesised insights that identify patterns across all ten personas. Claude Code extracts these insights and organises them into the deliverable formats described in the next section. The raw responses are equally valuable: the specific language personas use to describe their problems, the analogies they draw, the competitors they mention by name. This language feeds directly into messaging, content strategy, and sales scripts.
The Deliverables
A persona study produces three primary deliverables. Each serves a different audience within the organisation, and each requires a different analytical lens.
1. Data-Backed Persona Documents
The persona document is the most familiar output, but the Ditto version differs from the workshop version in one critical respect: every claim is traceable to a response. When the persona document says "this buyer consults peer communities before making a purchase decision," that claim is grounded in specific Q2 responses from specific synthetic personas. It is not an inference drawn from a single anecdotal interview.
A strong persona document includes: a demographic and professional profile (drawn from the research group filters), a day-in-the-life narrative (synthesised from Q1 responses), an information ecosystem map (from Q2), a goals-and-barriers statement with specific metrics (from Q3), a buyer journey reconstruction (from Q4), a decision criteria ranking with dealbreakers identified (from Q5), an adoption profile with churn risk factors (from Q6), and an aspirational jobs-to-be-done statement (from Q7).
The document should be written in narrative form, not as a bulleted specification. Narrative forces coherence. If the persona's goals (Q3), decision criteria (Q5), and aspirations (Q7) do not form a coherent story, you have either identified a segment split or asked a question poorly. Either way, the incoherence is informative.
2. Persona Segmentation
If you run the study with a single research group, Ditto's AI insights will identify clusters within the ten respondents. Three respondents might be motivated primarily by efficiency, four by credibility, and three by autonomy. These clusters are embryonic segments.
For more robust segmentation, run the same seven questions across two or three research groups with different demographic filters. Compare the responses. Where the groups diverge, you have found a genuine segment boundary. Where they converge, you have found a universal truth. The customer segmentation article describes this multi-group comparison methodology in detail.
The segmentation deliverable is a matrix that maps each persona cluster against the seven dimensions of the study. Which cluster has the highest urgency (Q3)? Which cluster has the longest buying cycle (Q4)? Which cluster is most price-sensitive (Q5)? Which is most likely to churn (Q6)? This matrix becomes the foundation for segment prioritisation: which persona should you build for first, message to first, and sell to first.
3. Persona Validation Report (Assumed vs. Actual)
This is the deliverable that justifies the entire exercise, and the one that most persona projects never produce.
Take your existing persona documents, the ones pinned to the wall, the ones nobody has validated in two years. Extract the specific claims they make. "Marketing Mary is 35 to 45, values ease of use above all else, discovers new tools through industry conferences, and makes purchasing decisions collaboratively with her team." Each of these claims is a testable hypothesis.
Now compare those claims against the Ditto study responses. Did the synthetic personas in that age range actually prioritise ease of use (Q5)? Did they cite conferences as an information source (Q2)? Did their buyer journey descriptions indicate collaborative decision-making (Q4)?
The validation report presents this comparison in a three-column format: the assumed attribute, the actual finding, and the gap. Where assumptions match reality, the persona is validated. Where they diverge, the persona needs updating. And where the study reveals dimensions the existing persona never addressed, like adoption behaviour (Q6) or aspirational jobs (Q7), you have identified blind spots.
The validation report is the single most politically useful document a product marketer can produce. It does not argue that the existing personas are wrong. It presents data that allows everyone in the room to see, simultaneously, where the assumptions hold and where they do not. The data does the persuading. You merely present it.
The Validation Angle: Testing Assumptions Against Synthetic Data
The validation workflow deserves its own discussion because it is where Ditto's value proposition is most distinctive. Any research platform can help you build personas from scratch. Very few can help you validate the ones you already have.
The process is straightforward. First, document your current persona assumptions in a structured format. For each persona, list: the demographic profile, the stated goals, the assumed pain points, the expected information sources, the hypothesised decision criteria, and the predicted adoption behaviour. Be precise. "Values efficiency" is not testable. "Prioritises integration with existing tools over feature depth" is testable.
Second, design the Ditto research group to match the demographic profile of the persona you are testing. If "Marketing Mary" is a 35-to-45-year-old marketing director at a mid-market SaaS company, create a research group with those exact filters.
Third, run the seven-question study and compare every response against every assumption.
The results typically fall into three categories. Category one: confirmed assumptions. These are the claims your personas make that the synthetic data supports. They are reassuring but not particularly actionable, you were already right. Category two: refuted assumptions. These are the claims the data contradicts. They are uncomfortable but extremely valuable, because every refuted assumption represents a decision you have been making on false premises. Category three: blind spots. These are dimensions the study reveals that your existing personas do not address at all. Adoption behaviour, churn triggers, aspirational jobs: if your personas never covered these, you have been navigating with an incomplete map.
In practice, most teams find that 40 to 60 percent of their persona assumptions are confirmed, 15 to 25 percent are refuted, and 20 to 30 percent represent blind spots. The refuted assumptions are where the immediate value lies. The blind spots are where the strategic value lies.
Keeping Personas Alive: The Quarterly Refresh
A persona is not a document. It is a living model of your market, and it requires maintenance.
The traditional approach to persona maintenance is to re-run the workshop every twelve to eighteen months, which means the personas are outdated for most of their lifespan. The Ditto approach is to run a shortened validation study quarterly. Three questions instead of seven: Q3 (goals and barriers), Q5 (decision criteria), and Q7 (aspirational jobs). These three questions capture the dimensions most likely to shift over time. Goals change with market conditions. Decision criteria shift as competitors enter and exit. Aspirations evolve as the category matures.
A quarterly study takes twenty minutes to design and forty-five minutes to run. The comparison between this quarter's responses and last quarter's responses reveals drift. If the primary barrier shifted from "lack of budget" to "lack of internal expertise," your messaging, content strategy, and sales enablement all need to adapt. If a new competitor started appearing in Q5 responses that was absent three months ago, your competitive intelligence needs updating.
Claude Code can automate this entirely. Instruct it to create a new research group each quarter with the same filters, run the three-question study, and produce a comparison report against the previous quarter's findings. The quarterly cadence transforms personas from static documents into a longitudinal dataset. Over four quarters, you have a trend line. Over eight quarters, you have a market narrative.
Common Mistakes and How to Avoid Them
Recruiting too broadly. A research group described as "professionals aged 25 to 55" will produce responses so varied that no coherent persona emerges. The whole point of the exercise is to identify patterns within a defined audience. If the audience is too broad, the patterns dissolve into noise. Be specific in your group description. It is better to run three narrow studies than one broad one.
Leading the objective. The study objective shapes the AI-generated insights. "Understand how mid-market marketers evaluate tools" is neutral. "Understand why mid-market marketers need AI-powered research tools" is leading. The second framing will bias every insight toward confirming your product's value proposition, which defeats the purpose of validation.
Ignoring the language. The most underutilised output of a persona study is the specific language respondents use. When a persona says "I need to prove my team's value to the C-suite," that exact phrasing belongs in your messaging. When they say "I am drowning in tools that each do one thing," that metaphor belongs in your positioning. The quantitative patterns matter. The qualitative language matters more.
Treating the persona document as the end product. The document is a communication tool. The real product is the shared understanding it creates. If you produce a beautiful persona document and nobody changes their behaviour, you have produced expensive wall art. The validation report, with its assumed-versus-actual format, is specifically designed to provoke action. Use it.
Skipping the blind spot analysis. Most teams focus on whether their assumptions were right or wrong and ignore the dimensions they never considered. The blind spots are where the competitive advantage lives. If your competitor's personas cover goals, barriers, and decision criteria, and yours also cover adoption behaviour, churn triggers, and aspirational jobs, you are operating with a richer model of the same market. That richness compounds over every decision.
Where This Fits in the PMM Stack
Persona development is not a standalone activity. It is the connective tissue of the entire product marketing function.
Messaging testing requires personas to evaluate message-market fit. Pricing research requires personas to segment willingness-to-pay by buyer type. Competitive intelligence requires personas to understand which competitors each segment considers. Content marketing requires personas to determine topics, channels, and formats. Sales enablement requires personas to build talk tracks and objection-handling guides. Go-to-market strategy requires personas to prioritise segments and select channels.
Every other function in the PMM stack either consumes personas or produces data that should update them. This is why the quarterly refresh matters. A persona that was accurate six months ago may be directing resources toward a segment that has shifted, a channel that has declined, or a message that no longer resonates.
Ditto and Claude Code make this feedback loop practical rather than aspirational. A full persona study takes under an hour. A quarterly refresh takes forty-five minutes. The cost of keeping personas current is negligible compared to the cost of operating on outdated assumptions.
The Claude Code and Ditto for Product Marketing Series
This is part of a series exploring how AI agents handle the core disciplines of product marketing. Each article covers one function of the PMM stack, explains the methodology, and links to a companion Claude Code guide you can run yourself.
Part 1: How to Develop Product Positioning (guide)
Part 2: How to Build Competitive Battlecards (guide)
Part 3: How to Validate Pricing Strategy (guide)
Part 4: How to Test Product Messaging (guide)
Part 5: How to Run Voice of Customer Research (guide)
Part 6: How to Segment Customers (guide)
Part 7: How to Validate GTM Strategy (guide)
Part 8: How to Build a Content Marketing Engine (guide)
Part 9: How to Build Sales Enablement Materials (guide)
Part 10: How to Research a Product Launch (guide)
Part 12: How to Develop and Validate Buyer Personas (guide) -- this article