Custom AI Interviewer Persona: Brand Voice & Tone for AI-Moderated Research

The Bottom Line

A custom AI interviewer persona is the configurable voice, tone, formality, and personality of the AI that moderates your customer interviews. When you're running research on behalf of a brand — Stripe, Notion, your own startup — the AI shouldn't sound like a generic chatbot. It should sound like a thoughtful interviewer from your team. That means matching the formality of your brand voice, using vocabulary your customers already hear from you, and projecting a personality that puts the right kind of participant at ease.

In Koji, you shape the AI interviewer's persona through three levers: the brief-level instructions (high-level tone, formality, must-mentions and must-avoids), the company context documents you upload (which the AI reads to understand your product and audience), and the branding settings on the interview landing page (logo, colours, intake messaging that primes the participant before the AI says a word). Together, these turn the default friendly-curious-researcher persona into a voice that feels native to your brand.

This guide explains the personality dimensions worth thinking about, which combinations work for which audiences, and the exact configuration knobs available in Koji.

Why Persona Matters for AI Interviews

It's tempting to dismiss persona as cosmetic. It isn't. Three concrete effects:

1. Completion rate. When the AI's tone matches the brand the participant signed up with, they stay engaged longer. A formal AI interviewer on a casual consumer app feels off; a chatty AI on an enterprise compliance survey feels unserious. Either way, drop-off rises.
2. Answer candour. Participants are 2x more candid with a moderator who feels like one of the team than with a generic third-party voice. Persona signals trust before the first answer.
3. Brand health. Every customer interview is also a brand interaction. A clumsy AI moderator damages the relationship; a thoughtful one strengthens it.

For B2B research where the participants are your customers (not random panel respondents), persona is especially load-bearing. They're comparing the AI moderator to the human CSM or AE they've already met. The persona either lives up to that bar or undermines it.

The Four Persona Dimensions Worth Configuring

A good persona is more than "friendly vs formal". Here are the four dimensions that actually shape how an AI interviewer feels:

1. Formality

On a spectrum from very casual to very formal. A SaaS startup interviewing free-trial users wants conversational ("Got it — what happened after you tried the import?"). A regulated industry like healthcare or finance wants neutral-to-formal ("Understood. Could you describe what occurred after the import attempt?"). Pick the level your brand uses in customer-facing emails and match it.

2. Pace

How quickly does the AI move through questions? Brisk personas confirm the answer and move on — good for power users with limited time. Patient personas linger on each answer, asking 2–3 follow-ups before progressing — good for empathy interviews or hard-to-reach customer segments. In Koji, pace is controlled by the probing depth setting (1 to 3 follow-ups per question) and the interview mode (structured = brisk, exploratory = patient, hybrid = balanced).

3. Warmth

How much affect does the AI project? Cool personas are professional and neutral; warm personas acknowledge feelings ("That sounds frustrating") and use lightly conversational connectives ("Totally — and after that…"). Warmth is helpful for stay interviews, exit interviews, and any emotionally loaded topic. It's less appropriate for B2B procurement research where buyers expect efficiency.

4. Domain Voice

Does the AI use industry vocabulary correctly? If you're researching developers, "shipped" and "PR" should land naturally. If you're researching CFOs, "burn rate" and "MRR" should be used precisely. Domain voice is mostly shaped by the context documents you upload to the study — your product one-pager, glossary, or sample interview transcripts.

Three Persona Archetypes (and When to Use Each)

Most teams converge on one of three archetypes:

The Curious Colleague. Casual formality, brisk pace, warm tone, light domain voice. Works for: consumer apps, prosumer SaaS, early-stage startup customer discovery, exit interviews where you want the participant to feel heard. Default starting point.

The Professional Researcher. Neutral formality, patient pace, cool tone, precise domain voice. Works for: B2B enterprise research, healthcare or finance regulated industries, market sizing studies. Signals that the research is serious without being cold.

The Subject-Matter Peer. Casual-to-neutral formality, brisk pace, cool-to-warm tone, heavy domain voice. Works for: developer research, designer research, technical buyer interviews — any audience where being seen as a peer matters more than being seen as a researcher. The AI uses the right jargon without explaining it.

Most Koji users start with the Curious Colleague archetype, then adjust one or two dimensions as they see real transcripts.

How to Configure Persona in Koji

There are three places where persona is shaped:

1. Brief-Level Tone Instructions

When you're generating the brief with the AI consultant (or editing it manually), you can add tone-and-style instructions directly:

"Tone: casual but precise — match Stripe's developer-facing voice. Use 'shipped', 'rolled out', 'flag' naturally. Avoid corporate-speak ('synergy', 'leverage'). Acknowledge frustration when participants describe friction."

These instructions are loaded into the AI interviewer's system prompt and apply across every conversation in the study. The more specific, the better. Generic instructions ("be friendly") produce generic personas. Specific instructions ("warm but brisk, use 'ship', avoid 'utilise'") produce distinctive personas.

2. Company Context Documents

Upload your product one-pager, brand voice guide, sample transcripts, or competitor analysis as context documents on the study. The AI references these to:

• Use your product's terminology correctly (feature names, plan tiers, key metrics).
• Match your brand's linguistic register.
• Understand which customer concerns are already known vs. genuinely new.
• Avoid pitching the product (research, not sales) while still sounding like part of the team.

In Koji, every study has a context-documents slot. The AI consultant reads these when drafting the brief; the AI interviewer references them mid-conversation when relevant.

3. Branding & Intake Settings

The interview landing page is the participant's first impression. Configure:

• Logo and brand colours — so the participant knows immediately whose research this is.
• Custom intake messaging — a short welcome explaining the goal and what to expect. This primes the participant for the AI's tone.
• Consent and intro language — if your brand has specific data-handling language, include it here.
• Voice mode default — pick whether voice is the default or text is, depending on which feels more on-brand.

With brand consistency from landing page → intake → AI moderator, participants experience one coherent voice rather than a generic AI dropped into your domain.

Persona Across the Six Structured Question Types

Koji's AI interviewer asks six native question types — open_ended, scale, single_choice, multiple_choice, ranking, and yes_no. Persona shapes how each is phrased:

• Open-ended in a Curious Colleague voice: "Walk me through the last time you ran into that — what happened?"
• Scale (1–5) in a Professional Researcher voice: "On a scale of 1 to 5, where 5 is excellent — how would you rate the experience overall?"
• Ranking in a Subject-Matter Peer voice: "If you had to keep one tool and drop the rest — which stays, which goes first?"
• Yes/no in a Curious Colleague voice: "Did you end up shipping that, or did it get parked?"

The structured value extracted is identical regardless of phrasing — but the conversational version is what shapes the participant's experience and (downstream) their willingness to give honest answers.

Common Mistakes

Trying to make the AI sound human. Don't. Participants know it's an AI; pretending otherwise erodes trust. Studies show participants are more candid with AI moderators on sensitive topics (compensation, dissatisfaction with employers, dropped subscriptions) precisely because they know it's not a human. Lean into thoughtful AI, not fake human.

Over-engineering the persona. Three lines of tone instructions usually beat thirty. Brief, specific, and concrete > a 500-word brand voice manifesto. The AI is good at applying a few constraints; less good at distilling a long abstract document into voice.

Mismatching persona to topic. A breezy Curious Colleague voice in a stay-interview about why people are considering leaving the company feels tone-deaf. Match persona to emotional register of the topic, not just to your brand.

Forgetting branding on the landing page. The AI's persona doesn't exist in a vacuum. If the participant arrives at an unbranded white page, the AI's personality has no context. Always brand the landing page.

Setting and forgetting. Read 5–10 real transcripts after launching a study and adjust the persona based on what you see. Real participant interactions reveal where the AI's tone is misfiring far better than dry-run testing.

A Worked Example: Configuring Persona for a Developer Research Study

Let's walk through configuring a persona for a study interviewing developers about their CI/CD tool experience.

Step 1: Pick an archetype. Subject-Matter Peer — developers want to talk to someone who speaks their language, not a market researcher.

Step 2: Add tone instructions to the brief:

"Tone: casual peer voice. Use 'shipped', 'rolled out', 'flag', 'main' naturally. Don't explain technical concepts unless the participant asks. Acknowledge engineering frustrations ('yeah, flaky tests are the worst') but don't pile on. Keep questions short — engineers prefer crisp."

Step 3: Upload context documents. Add your product one-pager, your engineering blog's last three posts (for voice), and a glossary of CI/CD terminology you want used correctly.

Step 4: Configure interview mode. Set to hybrid — cover required questions about current toolchain, then free-roam on what they wish existed. Probing depth: 2 follow-ups (engineers volunteer detail without prompting).

Step 5: Brand the landing page. Use your engineering team's logo variant and a 1-sentence intro like: "Quick chat with our AI interviewer about your CI/CD setup. ~8 min, no sales pitch."

Step 6: Run 3 test interviews and read the transcripts. Adjust as needed. If the AI is using too much jargon, dial it back. If it's being too formal, add an explicit "lean casual" instruction.

This 30-minute setup loop produces a persona that feels like your engineering team interviewing customers — not a generic AI chatbot.

Frequently Asked Questions

Can I configure the AI interviewer's voice (literally — the voice it speaks in) for voice interviews? Yes. Koji's voice mode supports configurable voice profiles — pick the gender, accent, and pace that match your brand. For most B2B research, a neutral, mid-pitch voice with a measured pace works best. For consumer research, you can lean warmer or more energetic.

Does persona affect interview quality scores? Persona shapes engagement (completion rates, answer depth) more than it shapes the 1–5 quality score itself. Quality scoring measures whether questions were covered, whether answers were on-topic, and whether structured answers parsed cleanly — these are persona-agnostic. But because better personas drive longer, more candid answers, they indirectly raise quality scores.

Can I run different personas in different studies under the same account? Yes. Persona is study-level, not account-level. You can run a Curious Colleague persona on your free-trial onboarding study and a Professional Researcher persona on your enterprise CFO interviews simultaneously.

What about multilingual studies — does persona translate across languages? Tone instructions transfer, but specific vocabulary doesn't (industry terms differ across languages). For multilingual studies, upload language-specific context documents or note in the brief that the AI should use the local industry vocabulary in each language.

Can I clone a persona across studies? Yes — duplicate a study and the brief (including tone instructions and context documents) copies over. Easiest way to maintain a consistent voice across multiple research projects.

Is there a default persona if I don't configure anything? Yes — Koji's default is "thoughtful, curious, professional researcher" with neutral formality, balanced pace, mild warmth, and no specific domain voice. It's a sensible starting point but you'll get better engagement by tuning at least the formality and domain voice dimensions.

Related Resources

• Customizing Your Study — study-level configuration including persona
• Customizing Branding — logo, colours, and landing page configuration
• Uploading Context Documents — how to feed the AI your product context
• Working with the AI Consultant — shape the brief (including tone instructions) at draft time
• AI Interviewer Tuning Guide — deeper tuning of conversational behaviour
• Structured Questions Guide — the 6 question types your persona shapes