AI-Moderated Interviews Sample

Larkspring makes insulated steel water bottles. Its own tracking shows that many owners stop carrying their bottle within a year. The team wants to know why bottles end up in the cupboard, and whether a new lid that comes apart for cleaning addresses the problem. We ran 180 one-on-one interviews, moderated by AI from a guide our senior researcher wrote, in five days of fieldwork.

Senior researcher

Designs, reviews, reports

Writes the guide and the probe rules
Pilots with 12 people and revises
Reads transcripts every day of fieldwork
Audits probes, removes failed interviews
Checks the theme coding and writes the readout

AI moderator

Runs each interview

Asks the guide questions in order
Follows up when an answer is vague or opens something new, up to two probes per question
Shows the lid concept in section 4
Keeps neutral wording and never suggests an answer
Flags interviews that look rushed or off topic

Participant

A real, screened person

Owns an insulated bottle
Chooses text or voice
Takes part from home, at any hour
About 20 minutes
Paid an incentive, as in any study

Who took part Measured

Six completed interviews were removed in review: two gave one-word answers throughout, three contradicted their screener, one was a duplicate.

Sample, after review Measured

	Daily carriers	Lapsed	Total
Larkspring owners	50	40	90
Other brand owners	46	44	90
Total	96	84	180

Lapsed: owns a bottle but carries it less than once a week. Ages 18 to 65, all regions.

Mode chosen

Text 112, voice 68

Text 62%Voice 38%

Median length: text 24 minutes, voice 17 minutes.

Timeline, working days

Days 1 to 3design

Guide written and approvedDraft, review with Larkspring, final.

Day 4pilot

12 pilot interviewsResearcher reads every one. Three changes made.

Days 5 to 9field

186 interviews completedDaily transcript review and probe audit.

Days 10 to 12report

Coding and readoutThemes checked by a second researcher.

What the pilot changed

Moved the cleaning question laterIn the pilot, asking about cleaning early made it the answer to everything. Now cleaning is never mentioned by the moderator before section 4, so every count of it on the Themes tab is unprompted.

Order

Added a probe rule for leaksPilot answers said "it leaks" and stopped. The moderator now asks where the bottle was and what was in it.

Probe

Cut the concept from 80 words to 40Voice participants could not hold the longer version in mind. Text and voice now see the same short card.

Stimulus

The guide is the researcher's main tool. It sets the questions and, just as important, the rules the AI moderator follows when it decides whether to probe. Click a section to open it. Target length: 19 minutes.

1Warm-upYesterday, in their own words2 min

Questions

Tell me a little about your day yesterday. Where were you, and what did you drink?

Rules for the moderator

Do not mention bottles, brands or water.
If the answer is under 15 words, ask for one more detail.

2The bottles they ownWhat gets picked, and why4 min

Questions

How many reusable bottles are in your home right now? Tell me about each one.
Which one do you reach for most, and why that one?

Rules for the moderator

If they name a feature, ask for a recent moment when it mattered.
Ask "why" once at most. Do not ladder further.
Do not name any brand they have not named.

3When it stays homeThe core of the study6 min

Questions

Think about the last time you left the house without your bottle. What happened?
Has there been a bottle you stopped using altogether? Tell me the story.

Rules for the moderator

Never raise cleaning, smell or leaks first. Let them come up on their own.
If "leak" comes up: ask where the bottle was and what was in it.
If "smell" or "gross" comes up: ask which part, and how they clean it.
If they say "I forgot": ask what was different about that day.

4A new lid ideaConcept reaction5 min

Questions

Concept card shown on screen or read aloudA lid that comes apart into three pieces with no hidden seal, so every part can go in the top rack of the dishwasher. Fits Larkspring 20 oz and 32 oz bottles.

What is your first reaction?
What, if anything, would make you doubt it?
Would it change how often you carry a bottle? Why or why not?

Rules for the moderator

Ask for doubts even when the reaction is positive.
Do not defend or explain the concept beyond the card.
If asked about cost or where to buy: "I don't have that information. What would you expect?"

5Wrap-upOne change, anything missed2 min

Questions

If you could change one thing about the bottle you use most, what would it be?
Is there anything we should have asked and did not?

Rules for the moderator

No probes on question 2. Thank them and close.

Rules for every question

Two follow-ups at most per question, then move on.
Skip the probe when the answer already has an example and a reason.
Neutral prompts only: "tell me more," "what happened next," "what makes you say that."
Never introduce a brand, feature or problem first.
If someone is upset or wants to stop, end politely and flag for the researcher.
If someone asks for a person, pause and notify the researcher.

Follow-up probes used Measured

2,900 probes across 180 interviews, about 16 each.

Probe audit by the researcher Measured

300 probes reviewed

91%

On target 273Not needed 21Leading 6

Answers that followed the six leading probes were left out of the theme counts.

Every transcript was coded into themes by a researcher. Counts show how many people raised each theme on their own, before the moderator mentioned cleaning or leaks. With 180 interviews, the counts can be compared across groups.

Recommendation

Launch the Clean Lid,
and lead with the seal

Cleaning the lid is the most common reason lapsed owners stop carrying a bottle: it is the main reason for 36% of them, ahead of any other. 62% of lapsed owners reacted well to the Clean Lid. Their first doubt is whether a lid that comes apart still seals, because one leak is enough to retire a bottle. Show the seal before claiming easy cleaning, then size the opportunity with a survey.

Why bottles get left behind, unprompted Measured

Share of each group who raised it at any point. People could raise more than one.

Main reason, lapsed owners Measured

One main reason coded per person, n=84. Adds to 100%.

Themes and quotes Illustrative quotes

The lid is where the trouble lives 86 of 180 · 48%

Cleaning complaints were almost always about the lid and its seal, rarely the bottle. People described tools: butter knives, toothpicks, brushes bought just for the lid.

I bought a special little brush just for the lid. That's when I knew it was a problem.0062 · lapsed · text

It smells like a damp towel. You notice it when you take a sip.0147 · lapsed · voice

One leak ends it 71 of 180 · 39%

A single leak in a bag was enough to retire a bottle for good. People rarely blamed themselves, even when the lid had been on crooked.

It leaked once on my laptop. Once. That bottle is a pencil cup now.0031 · lapsed · voice

I carry it upright in the side pocket now, like it's a baby.0118 · daily · text

Full is heavy 43 of 180 · 24%

The large size is what people buy and what they leave at home. Several now fill a smaller bottle twice.

32 ounces is great at my desk and awful on a walk.0089 · daily · text

Routines break, bottles stay home 31 of 180 · 17%

A change in routine, such as a new job or working from home, quietly ended the habit.

I carried it every day to the office. Once I started working from home, it just stayed on the counter.0173 · lapsed · voice

Reaction to the Clean Lid Measured

PositiveMixedNegative

Coded from the first-reaction answer and its probes. 98 positive, 52 mixed, 30 negative.

Doubts raised about the Clean Lid Measured

All 180, asked of everyone including positive reactions. People could raise more than one.

Next steps

Leak test the three-piece lidbefore any claim
Survey, 400 bottle ownerssize the lapsed group
Test two claims: seal first or cleaning firstsame survey
8 in-home interviews, human-moderatedwatch people clean a lid

AI moderation and human moderation are good at different jobs. Many projects use both: AI-moderated interviews to map themes across many people, then a few human-moderated sessions to go deeper with the most interesting ones. Every participant in either method is a real, screened person.

Side by side

	AI-moderated	Human-moderated
Number of interviews	Dozens to hundreds	Usually 8 to 30
Time in field	A few days	One to three weeks
Scheduling	Any hour, the participant's choice	Set appointments
Consistency	Same guide and probe rules every time	Varies with the moderator and the day
Follow-up	Rule-based probes, up to a set limit	Free to chase anything that matters
Reading the person	Words only, typed or spoken	Tone, face, hesitation, body language
Physical products	Images, video, or a product mailed ahead	Hands-on, in a facility or at home
Sensitive topics	Some people say more to a screen	Better when care and judgment are needed
Counting themes	Enough people to count and compare groups	Too few to count
Client watching live	Reads transcripts as they arrive	Watches and steers from the back room

Which fits your project?

Examples

Why do people cancel a subscription? 150 recent cancellers, two regions.AI-moderated

Watching people cook from a new recipe card in their own kitchen.Human-moderated

Reactions to three ad scripts, compared across five age groups.AI-moderated

Talking with caregivers about a hard health decision.Human-moderated

Map the reasons people quit a habit, then go deep with ten of them.Both

What stays human in every AI-moderated study

The guide. A senior researcher writes the questions, the order and the probe rules.
The pilot. Every pilot transcript is read before full launch.
Quality review. Interviews that fail checks are removed by a person.
The probe audit. A sample of probes is read for leading wording.
The readout. Themes are checked and the recommendation is written by a researcher.