# How to synthesize customer interview notes into clear themes on Mac for product managers

> Turn customer interview notes into clear themes on your Mac. Costs, causes, a week routine, Apple Notes tips, AI limits and where Lazur helps.

URL: https://www.lazur.app/for/product-managers/problems/synthesizing-customer-interview-notes-into-themes
Updated: 2026-11-08

To synthesize customer interview notes into clear themes, follow three steps. Put every note in one place. Tag each observation with a short code. Then group the codes into a few themes you can back with quotes. On a Mac, Apple Notes handles the tags and filters. The harder part is discipline. Use the same codes every time, and check that each theme is more than your favorite idea.

## Quick answer

- Tag each observation with a short code written in the customer's words before you group anything.
- Count how many participants said each thing. A theme from one loud call is a hunch.
- Apple Notes handles tags and Smart Folders on Mac. It does not cluster themes for you.
- AI can draft a first pass, but a small study found many of its codes did not match human codes.
- Write each theme as one sentence in the customer's language, with quotes you can trace back.

## What turning interview notes into themes costs you

- **Hours you do not get back.** The lyssna survey found that 60.3% of respondents named time-consuming manual work as their top research frustration. Synthesis is the part of research people do by hand after the calls end. Every hour spent re-reading notes is an hour not spent on the next round of calls or on the decision itself.
- **Decisions made on thin evidence.** If the themes come from the loudest call, the roadmap follows that call. You may build a feature for one account and miss the job most customers share. The cost shows up later, when a shipped feature does not move the numbers you expected.
- **Days of waiting before anyone decides.** The same survey put 79% of respondents at a day or more for research synthesis. Decisions wait while that work happens. Teams often plan the next sprint without the findings, so the findings arrive after the choice is already made.
- **Paid time, measured roughly.** The BLS median for market research analysts was $78,760 in 2025. That occupation is a proxy, not a product manager. Divide the figure by a 2,080-hour year and you get about $37.87 an hour. Use that as a rough scale for planning, not as a rate card.
- **Tool spend that adds up.** Dedicated research platforms add their own cost. Dovetail's pricing page lists a Bronze plan at $4,800 a year, plus usage credits at 15 cents each. Silver and Gold are quote-only. A tool like that only pays off when your team runs research every week.
- **Quotes that disappear.** When codes live only in memory, quotes go missing. The strongest sentence from a call can vanish before the write-up. The team then argues from summaries instead of the customer's own words. Attach each quote to a code on the day you capture it.

## Why it happens

1. **Notes are written for the note-taker.** Shorthand makes sense to the person who wrote it, on the day they wrote it. A note like 'billing confusing, wants fix' loses its context within weeks. Without the participant, role and date, nobody can judge how much weight the note deserves.
2. **No shared codebook.** Two people can read the same note and code it differently. 'Pricing' may mean cost to one person and value to another. Without a written list of codes, the group reads the same notes and reaches two different themes, then spends the meeting debating which one is right.
3. **Themes get picked too early.** The first calls set a story, and later calls get read through it. Notes that fit the story get more attention, and notes that do not fit get filed as noise. This is a common human pattern, and it is the main reason to write counts before names.
4. **Observations mixed with solutions.** Customers describe problems. Product managers often write their own ideas in the same line, such as 'add a filter.' Mixed notes cannot be sorted into themes until someone separates the evidence from the guess. That separation takes time most people do not schedule.
5. **Notes spread across too many places.** Quotes end up in Notion pages, Slack threads, Google Docs, sticky notes and voice memos. When nothing sits in one place, nobody can count how often a problem came up. Counting is what turns a pile of notes into a theme.
6. **Synthesis waits until the end.** Teams often wait until the whole round finishes before they read anything back. The first problem then appears as a surprise, and the team has already scheduled the remaining calls with the same gaps. A short pass after each call catches drift early.

## How to fix it

- **Write a one-page codebook first.** List each code as a short tag with one line of meaning and one example quote. Keep the list in a pinned Apple Note. Change it only at a planned checkpoint, not in the middle of a batch, so the early and late notes stay comparable.
- **Put the participant on every line.** Add the participant ID, role, company type and date to the first line of each note. Those few fields let you check whether a theme holds across roles, and they let you count people instead of mentions.
- **Code observations, not solutions.** Move feature ideas to a separate list. If a note says 'they want a dark mode,' the code should describe what the customer was trying to do, such as reading reports at night. The idea can wait on its own list until the theme is clear.
- **Count before you name a theme.** Build a Google Sheet with codes down the side and participants across the top. Mark a cell each time a participant raised a code. Sort by the number of participants, not by how strongly you remember one call. The guide on [counting unique values in Google Sheets](https://www.lazur.app/how-to/count-unique-values-in-google-sheets) covers the formulas.
- **Keep a disagreement note for each theme.** For every theme, record one note that does not fit and say why. Sometimes the note shows a real edge of the theme. Sometimes it shows a second theme hiding inside the first. Both results make the write-up stronger.
- **Use Smart Folders as live buckets.** Create one Smart Folder per theme tag from File > New Smart Folder in Notes. Each folder updates as you tag new notes. Remember that a Smart Folder holds references to notes, so keep the original notes in their normal folder.
- **Let AI suggest matches, not decide themes.** Use an AI tool to suggest which notes might share a code. Check every match against the raw note before you accept it. Keep the codebook fixed before you ask, because the study with a fixed codebook agreed far more with human coders than the study that invented codes.
- **Run a mid-round check with a colleague.** After the first few calls, read your codes with one colleague. Merge codes that overlap, and split codes that hide two problems. This check costs a short meeting and catches drift before the last call is in.

## What bad synthesis costs you in time

Start with the cost, because it decides how much routine you need. In a 2025 vendor survey run by lyssna, 300 professionals were asked what frustrates them about research. Of those, 60.3% named time-consuming manual work as their top frustration. Treat that as a signal, not a census. The survey is one data point, but the time numbers below point the same way.

The time question is harder to shrug off. The same survey asked how long research synthesis takes. The answers fell into four buckets. 21% said less than a day. 35.0% said one to two days. 30.3% said three to five days. 13.7% said more than five days. Add the three longer buckets and you get 79%. That is where the claim that synthesis takes at least a day comes from. The survey gives no average in hours, so do not quote one.

- **60.3%** of 300 professionals in a 2025 lyssna survey named time-consuming manual work as their top research frustration ([source](https://www.lyssna.com/reports/research-synthesis/))
- **79%** said research synthesis takes a day or more. This adds the three longer buckets: 35.0%, 30.3% and 13.7%. ([source](https://www.lyssna.com/reports/research-synthesis/))
- **31%** of GPT-4o's 137 inductive codes matched human codes in one small JMIR study with four students. ([source](https://preprints.jmir.org/preprint/85572))
- **$78,760** is the 2025 BLS median pay for market research analysts, a proxy occupation, not product managers. ([source](https://www.bls.gov/ooh/business-and-financial/home.htm))

**Time spent on research synthesis, share of respondents**

- Under a day: 21 %
- 1 to 2 days: 35 %
- 3 to 5 days: 30.3 %
- Over 5 days: 13.7 %

Source: https://www.lyssna.com/reports/research-synthesis/

Put the number on your calendar, not in a slide. Look at your last round of calls. Count the afternoons you spent re-reading notes. Then count the decisions that waited on those afternoons. For many teams, the waiting costs more than the reading.

![Product manager working on a Mac laptop with a notebook for customer interview notes](https://images.lazur.app/content-images/problem/synthesizing-customer-interview-notes-into-themes/synthesize-customer-interview-notes-into-clear-themes-3-9a5467.jpg)

Money is harder to pin down. The BLS Occupational Outlook Handbook lists a 2025 median pay of $78,760 for market research analysts. That is not a product manager's job title, so treat it as a proxy. Divide it by a 2,080-hour work year and you get about $37.87 an hour. That division is my arithmetic, not a published number.

Use the hourly figure to ask a simple question. What does a day of synthesis cost your team in paid time, and what could that day have built instead? You do not need an exact answer. You need a rough scale you can say out loud in planning.

## Why notes do not turn into themes on their own

Notes are written for the person who took them. A line like 'billing confusing, wants fix' makes sense that afternoon. A few weeks later it can mean several things. Nobody can tell which customer said it, or whether it came up once or in most calls.

Themes also get picked too early. The first calls set a story. Later calls get read through that story, so the notes that fit it get more attention. Mixing observations with feature ideas makes it worse. A customer describes a jam, and the note says 'add a filter.' The jam is the evidence. The filter is a guess.

## A routine to synthesize customer interview notes in one week

Run this routine across the week. Each day has one job, so the notes never pile up. You can do it with one Notes folder and one Google Sheet. Each step builds on the one before it.

1. Capture the same day. Put the participant ID, role and date on the first line of each note.
2. Write observations. Read each note once and write one line per observation. Keep the exact customer words in quotes.
3. Code the observations. Give each one a short tag. Name the problem, such as manual-reporting, not your fix.
4. Count by participant. In a Google Sheet, list codes down the side and participants across the top. Mark a cell when that person said it. The [unique-value formulas](https://www.lazur.app/how-to/count-unique-values-in-google-sheets) help when the list grows.
5. Group codes into themes. Cluster codes that describe the same job or blocker. Name the theme for the job.
6. Test the theme. Find the interview that disagrees, and write down why it does not fit.
7. Write the sentence. Phrase the theme in the customer's words, with the best quotes attached.

Notice what the routine skips. It does not ask you to pick the top themes on day one. It also does not treat your favorite feature as evidence. Counts come before names, and names come before the write-up.

## A worked example of coding one batch of notes

Say you talked to customers about a reporting feature. Several notes mention copying numbers into a spreadsheet every Monday. Another says they export to CSV because the dashboard cannot filter by team. A later note says the team already pays for another tool for the same job.

Do not name a theme yet. Write the codes first, one per observation. The table shows how the raw note, the code and the eventual theme sit together. Keep the raw note beside every code, so you can always go back to the words.

| Raw note | Code | Theme |
| --- | --- | --- |
| Copies numbers into a sheet every Monday | manual-reporting | Getting reports out of the dashboard |
| Exports to CSV because the dashboard cannot filter by team | export-workaround | Getting reports out of the dashboard |
| Team already pays for another tool for this job | tool-overlap | Paying twice for reporting |

_Codes come first. Themes come last, after counts._

The theme is the job, not the fix. 'Getting reports out of the dashboard' is a theme. 'Add a CSV button' is a solution, and it belongs on its own list. When a theme names a fix, the team starts arguing about the fix instead of the evidence.

![Two people reviewing interview notes together at a desk with a laptop open](https://images.lazur.app/content-images/problem/synthesizing-customer-interview-notes-into-themes/synthesize-customer-interview-notes-into-clear-themes-1-afd41b.jpg)

## Tagging and filtering interview notes in Apple Notes on Mac

Apple Notes is already on your Mac, so it is the cheapest place to start. Type # followed by a word to make a tag. Tags must be one word, but hyphens are allowed inside them. That means export-workaround works, and export workaround does not. Apple's help page says this directly.

To see every note with a code, open the sidebar and select a tag. Select more than one tag and the Tags pop-up menu lets you choose All Selected or Any Selected. Use All Selected when you want notes that carry two codes at once, such as manual-reporting and tool-overlap. Use Any Selected when you want everything that touches a broad area.

## Smart Folders as live theme buckets

A Smart Folder collects notes by rule. Open File > New Smart Folder, name it, and pick rules by Tags, Date Created, Date Edited or Checklists. You can require any or all conditions. A rule for the tag manual-reporting gives you a bucket that fills itself as you add notes.

Smart Folders have limits, and you should know them before you rely on one. Apple says a Smart Folder cannot be locked, turned into a subfolder, or shared, though it can reference shared notes. A Smart Folder holds references to notes in other folders, not the notes themselves. Treat it as a view, not a filing cabinet.

Notes does not cluster themes. I could not find a clustering feature in the Notes help pages I checked. Use Smart Folders as buckets you name yourself, not as a clustering engine.

> **Pin your codebook:** Keep your codes in a pinned note. Give each code one line: the tag, what it means, and one example quote. Read it before every batch of notes.

## Using AI on interview notes without trusting it blindly

AI can help with the sorting, but the evidence is mixed. In a JMIR Preprints study, a one-hour focus group with four students was coded by three human coders. GPT-4o applied the human codebook with a mean percent agreement of 96%. The mean Cohen's kappa was 0.71.

Now the other half of the result. When GPT-4o created its own codes, the study produced 137 initial codes. Only 31% matched human codes. Another 26% were reasonable alternatives, and 42% were not reasonable. Fixed codes are easier for a model. Themes you discover from scratch are harder.

A second study coded 21 interviews with men who survived firearm violence. It describes manual thematic analysis as time-consuming and labor-intensive. It reports low relevance in the LLM-identified codes. It also notes that guardrails caused substantial narrative erasure. The topics differ from yours, so do not generalize too far. The pattern is still worth noticing. A model can drop the parts of a story that do not fit its rules.

A practical rule follows from this. Let AI suggest which notes might share a code. Then check each match against the raw note. Keep your codebook fixed before you ask, because the study that agreed most worked from a fixed codebook.

> **Check the data rules first:** Before you paste a transcript into an AI tool, check your employer's rules on customer personal data. Remove names, emails and account details. If your company bars it, skip the AI step and code by hand.

## Research tools and what they cost

Dedicated research platforms exist for this job. Dovetail's pricing page lists a Bronze plan at $4,800 a year for the platform. Usage credits cost 15 cents each, and the plan covers up to 5 users, billed annually. Silver and Gold have no listed prices, and the page sends buyers to sales.

I did not find a free plan on that page. Pricing pages change, so check current plans before you build a budget around one. I also could not confirm Mac-native apps for Dovetail or Miro. Check their download pages before you assume one exists.

For most product managers, a Notes folder and a Google Sheet cover the first few rounds. Move to a paid tool when you run many rounds, work with several researchers, or need a shared quote library. Then compare the cost against the afternoons you saved. A [Notion database](https://www.lazur.app/how-to/create-a-database-in-notion) can hold the themes if your team already lives in [Notion](https://www.lazur.app/for/product-managers/notion).

## Mistakes that waste the most time

- Coding ideas instead of observations. 'We should add export' tells you nothing about what the customer did.
- Naming themes before counting. A loud call is not a pattern, and the roadmap will treat it like one.
- Dropping the participant ID. Without it, you cannot tell whether one person said the same thing many times.
- Skipping the note that disagrees. Write it down and explain why it does not fit. That note often shows where the theme ends.
- Using vague tags like feedback or misc. A tag that fits every note tells you nothing about the customer.
- Waiting until the round ends. Problems you could have caught after the first call show up as surprises at the end.

## Edge cases: small rounds, mixed roles and sensitive topics

A small round makes counts fragile. A theme backed by just a couple of people may be real, or it may reflect who you happened to call. Say the count in the write-up, and name the limit. A theme from one account is still useful, but label it as a hypothesis, not a finding.

Mixed roles need separate tags. A finance buyer and an end user can describe the same problem in different words. Add the role to each note, then check whether a theme holds for both groups or only one. A theme that holds only for buyers is a different product problem.

Sensitive interviews need more care. Customers may talk about health, money or legal trouble. Store notes only where your company allows, and strip identifying details before you share a quote. If you work under a regulated employer, such as a hospital, law firm or accounting practice, follow its rules on client or patient data before you use any tool. Your employer's rules come first.

## Check your themes before you share them

Before a theme goes into a roadmap doc or a Slack post, run a few checks. Can you point to notes that carry the code? Can you name the participant behind each quote? Can you say what evidence would prove the theme wrong?

Then write the theme as one sentence in the customer's words. 'People copy numbers into a sheet every week because the dashboard cannot filter by team' is a theme a colleague can test. 'Reporting is a pain' is not. Vague themes survive meetings because nobody can disagree with them.

![Sticky notes arranged on a wall for grouping research findings into clusters](https://images.lazur.app/content-images/problem/synthesizing-customer-interview-notes-into-themes/synthesize-customer-interview-notes-into-clear-themes-2-940d5f.jpg)

Once a theme is ready, move it out of Notes. Put each theme in a Notion database or a [Linear project](https://www.lazur.app/for/product-managers/linear), so it can link to real work. Keep the counts in a Google Sheet, so the numbers update as new calls come in. Keep one source for each quote, not several copies.

## What to do next

Pick one round of calls you already have. Start with the oldest notes, so you learn the routine before a deadline is on you. Make one tag system, one Smart Folder per theme, and one Sheet with your counts. Then write the first theme sentence and test it against the note that disagrees.

## With Lazur

Lazur is a voice assistant for the Mac that lives in the notch. Hold ⌥ ⌘ and ask; it sees your screen, answers out loud, points at buttons and walks you through apps. Hold fn to dictate. Free for 3 days, then $18/month: https://www.lazur.app/download

- “Where do I make a Smart Folder in Notes?”: Lazur circles File > New Smart Folder on screen and labels it. You click it yourself. (Points at it)
- “Where is File > New Smart Folder in Notes?”: Lazur circles the menu item on screen and labels it. You click it yourself.
- “Walk me through making a Smart Folder for one tag.”: Lazur guides you one click at a time and waits for each click before the next step.
- “Explain what a Smart Folder can and cannot do.”: Lazur answers out loud about what is on your screen. It does not change your notes.
- “Hold fn and dictate a quote with its participant ID.”: Dictation types clean text at the cursor in any app, so adding quotes takes less typing.
- “Draft one theme sentence from these quotes.”: Lazur drafts text you can edit and type into your doc. Check every word against the raw notes before you share it.

## A plan you can start this week

- **Monday: Capture the same day.** Put each call's notes in one Notes folder before you do anything else. Add the participant ID, role and date on the first line. A note with no ID is hard to trust later.
- **Tuesday: Tag each observation.** Read each note once. Write one-line observations, tag them with short codes, and keep the codes in your pinned codebook note. Stop tagging when the same codes start to repeat.
- **Wednesday: Count by participant.** Build a Google Sheet with codes down the side and participants across the top. Mark a cell each time a participant raised a code. Sort by participant count, not by how strongly you remember a call.
- **Thursday: Test the themes.** Group codes into themes named for the job. Then find the note that disagrees with each theme and write down why. A theme that cannot survive its disagreeing note needs to shrink or split.
- **Friday: Write and share.** Write each theme as one sentence in the customer's words, with the best quotes and the count. Post it in a Slack channel or a Notion page, and label small samples as hypotheses.

## FAQ

### How long does it take to synthesize customer interview notes into clear themes?

The lyssna survey did not report an average in hours. It found that 79% of respondents spent at least a day on research synthesis, by adding the three longer time buckets. Your time depends on how many calls you ran and how clean the notes are.

### Can Apple Notes group interview notes into themes on Mac?

Apple Notes can tag notes and filter them with Smart Folders. Based on the help pages I checked, it does not cluster notes into themes by similarity. Use tags as codes and Smart Folders as named buckets.

### How do I make a Smart Folder in Notes on Mac?

Open File > New Smart Folder. Name it, then set rules by Tags, Date Created, Date Edited or Checklists. Choose whether any or all conditions must match. Apple says a Smart Folder cannot be locked, turned into a subfolder or shared.

### What is the difference between a code and a theme?

A code is a short tag on one observation, such as export-workaround. A theme is a pattern across many codes, stated as a job or a blocker. A theme should point back to counted participants and to quotes you can trace.

### Is AI good enough to find themes in interview notes?

The evidence is mixed. In one small JMIR study, GPT-4o agreed with a fixed codebook at 96% mean agreement. Its own inductive codes matched human codes only 31% of the time. Use AI to suggest matches, then check each one against the raw note.

### How many interviews make a theme trustworthy?

The sources I checked do not give a safe number. Count participants, not mentions. Say the count in your write-up, and label themes from a small round as hypotheses.

### Do I need Dovetail or another research tool?

No. A Notes folder and a Google Sheet can carry a first round. Dovetail's pricing page lists a Bronze plan at $4,800 a year for the platform, and it sends buyers to sales for higher tiers.

### Can I paste customer interview notes into an AI tool?

Check your employer's rules on customer personal data first. Remove names, emails and account details before any paste. If your employer is regulated, follow its rules on client or patient data, even if the tool seems harmless.

### Does Lazur edit my notes or set up Smart Folders for me?

No. Lazur guides, points, explains and drafts text. It does not click buttons or edit your notes. You tag and click. Lazur can show you where a command is and walk you through it one step at a time.

## Sources

1. [lyssna Research Synthesis Report 2025](https://www.lyssna.com/reports/research-synthesis/), lyssna (vendor survey), checked 2026-11-08
2. [BLS Occupational Outlook Handbook: Business and Financial Occupations](https://www.bls.gov/ooh/business-and-financial/home.htm), U.S. Bureau of Labor Statistics, checked 2026-11-08
3. [Use Smart Folders in Notes on Mac](https://support.apple.com/guide/notes/use-smart-folders-apd58edc7964/mac), Apple Support, checked 2026-11-08
4. [Use tags in Notes on Mac](https://support.apple.com/en-ca/guide/notes/apdc88ed7f1d/mac), Apple Support, checked 2026-11-08
5. [Dovetail Pricing](https://dovetail.com/pricing/), Dovetail, checked 2026-11-08
6. [JMIR Preprints #85572: Evaluating LLM Assisted Qualitative Analysis in Medical Education Research](https://preprints.jmir.org/preprint/85572), JMIR Preprints, checked 2026-11-08
7. [Can LLMs Understand the Impact of Trauma? Costs and Benefits of LLMs Coding the Interviews of Firearm Violence Survivors](https://arxiv.org/abs/2604.16132), arXiv, checked 2026-11-08