How the Test Was Run

The recording is "Focus Group Session 5 Demonstration," a public training video published by the University of Minnesota MCH program on YouTube. It shows a moderator and five participants discussing what makes a healthy family, which makes it a realistic stand-in for consumer and community research. The test used the first 10 minutes (0:00 to 10:03), which include introductions, ground rules, an icebreaker and open discussion with overlapping speech.

All seven received the same file on October 27, 2026, on a free plan or trial at default settings. Each was timed from upload to finished transcript, and the results were scored against a clean verbatim human transcript, checked in a second full pass by another transcriptionist.

Word accuracy is 100% minus the word error rate. Because the reference is clean verbatim, filler words (um, uh), discourse markers (like, you know, okay, so) and immediate repeats were stripped from every transcript before scoring, so no tool lost points for capturing real speech, and participant surnames were left out. Speaker accuracy is the share of correctly transcribed words assigned to the right person once each tool's labels (Speaker 1, Speaker A) were matched to the real speakers. The test also counted how many of the six spoken first names each tool got right (Dick the moderator, Kara, Ghazaleh, Laurel, Hannah and Noah, with surname spellings not counted) and listed every error that changed what someone actually said.

Test Results at a Glance

*Amberscript was scored only on the portion it transcribed.

Test recording: "Focus Group Session 5 Demonstration" by the University of Minnesota MCH program, published on YouTube. Short excerpts and screenshots are used here for review and comparison only, and the full transcript is not published.

What the Results Mean for IDIs and Focus Groups

Speaker labels decide whether a transcript is usable

A focus group transcript without speaker labels is a wall of text, and coding it means going back to the recording to work out who said what, which removes most of the time AI was supposed to save. In an IDI the same gap makes it hard to separate the moderator's questions from the respondent's answers, and dyads and triads sit somewhere in between. Five of the seven tools separated all six speakers, and Instant Draft, Rev, HappyScribe and Amberscript attributed almost every word to the right person. Otter labelled the entire session "Unknown Speaker," and TurboScribe returned unlabelled text because speaker recognition is an optional setting that is switched off by default. TurboScribe users can turn it on before uploading, but a researcher who uploads with the default settings gets a transcript that cannot be coded by participant.

Names are where most tools slip

Names matter in market research because moderators address participants by name, and analysts use those moments to anchor quotes. Instant Draft, HappyScribe and TurboScribe got five of six names right. Rev, Sonix, Amberscript and Otter all turned "Kara" into "Carl," and Rev, despite its high word score, also left one participant's name out of the transcript entirely. No tool got "Ghazaleh" right, which is common with AI transcription when names come from languages the speech models hear less often, and it is a good reason to check names before a transcript goes into a report.

Word accuracy depends on what counts as an error

TurboScribe (96.7%) and Rev (96.5%) scored slightly above Instant Draft (95.1%) on raw word accuracy. Most of the gap comes from words people actually said that the clean verbatim reference left out, such as the moderator's restart "what is a health, what does a healthy family do" and the repeated greeting "Good afternoon. Good afternoon." Instant Draft keeps these, while Rev and TurboScribe smooth them over. A tidier transcript reads more easily, but researchers who study hesitation and self-correction often want the false starts kept, and the guide to full vs clean verbatim transcription explains when each format fits.

Meaning-changing errors matter more than the percentage

A transcript at 96% accuracy can still put the wrong words in a participant's mouth, and in market research those words often become the verbatims in a client report, so the table below lists every substitution that changed what someone said, excluding names and spelling variants.

The "exercise" line was spoken during crosstalk and laughter, and the second-pass reviewer's reading was used. Leaving that one aside, Instant Draft and HappyScribe produced no meaning-changing errors in the section they both covered.

Free plans don't always transcribe the whole file

Two tools stopped short on their free plans. HappyScribe's free AI allowance is a 10-minute trial, so the last 16 seconds were cut off, and Amberscript transcribed only the first 6 minutes and 34 seconds. That's fine for evaluating a tool, but worth knowing before you rely on a free tier for real fieldwork.

Interview volume decides which plan you need

An IDI study often runs 20 to 30 interviews of 45 to 60 minutes each, which adds up to 15 to 30 hours of recordings in a few weeks, and per-file limits matter as much as total minutes. Otter's Pro plan caps each conversation at 90 minutes and allows only 10 file imports a month, so a 25-interview study would not fit in one month. TurboScribe's free plan stops at 30 minutes per file. Instant Draft's Pro+ plan covers 20 hours a month and Researcher+ covers 50 hours shared across five seats, enough for a full IDI study plus the focus groups that often run alongside it. The interview transcription page covers human options for interviews that need a final verbatim record.

Field recordings are harder than this test

The test recording is a clear, in-room session. TDIs recorded over phone lines, and shop-alongs or in-home ethnographies recorded on a phone with store or household noise behind them, usually bring accuracy down for every tool. Speaker labels and name handling matter even more on those recordings, and they are the sessions where checking quotes against the recording, or ordering human market research transcription, is worth the extra step.

The Seven Tools, Reviewed for IDIs and Focus Groups

1. Qualtranscribe Instant Draft: best overall for IDIs and focus groups

Instant Draft was the fastest tool in the test and the only one to go past the transcript without being asked. It labelled all six speakers with 99.8% accuracy, made one meaning-changing error, and as soon as the transcript was ready, Smart Insights produced a summary with key findings and quotes linked to their timestamps, so an analyst can click a finding and hear the moment it came from.

Market research teams also have to answer to clients about where recordings go, and the Insights Association's Code of Standards says personal data should not be used in AI training data sets without informed consent. Qualtranscribe never trains AI on customer recordings on any plan, keeps EU data in Frankfurt and US data in Northern Virginia, redacts personal information with its Anonymize feature (with human de-identification available too) and signs agency NDAs and DPAs on request. Smart Insights adds 17 research prompts, including Braun and Clarke thematic analysis, and when a session is headed for a client deliverable, the same account can order human focus group transcription in 25 languages.

Pricing

• Free: 75 minutes a month, 5 Smart Insights runs, no credit card

• Pro: $18 a month ($12.60 billed yearly), 600 minutes, HIPAA and GDPR, Word, SRT and NVivo export

• Pro+: $35 a month ($24.50 yearly), 1,200 minutes, unlimited Smart Insights

• Researcher+: $75 a month ($52.50 yearly), 3,000 minutes shared across five seats

2. Rev: high word accuracy, weaker on names and speed

Rev produced a clean, readable transcript with perfect speaker attribution and 96.5% word accuracy, the second-best raw score. It was also the slowest tool at 2 minutes 40 seconds, and it wrote "our discretion" for "our discussion," the kind of slip that is awkward in a client report.

Rev now positions its platform mainly around legal and investigative work, and it says it does not allow customer data to be used to train external large language models. The Qualtranscribe vs Rev comparison goes further.

Pricing:

• Free: 45 AI minutes a month, English only

• Essentials: $25.49 per seat a month, 5,000 minutes, English and Spanish

• Pro: $47.99 per seat a month, 10,000 minutes, 37+ languages

• Unlimited: custom pricing, includes HIPAA

3. HappyScribe: accurate speaker labels and names, limited free trial

HappyScribe matched Instant Draft on names and attributed every word to the right speaker, and at 25 seconds it was the second-fastest tool. Its 92.8% word score was the lowest of the seven, partly because the free plan's 10-minute limit cut off the last 16 seconds.

HappyScribe is GDPR compliant and SOC 2 Type II certified with EU hosting, but its own help center states that it is not HIPAA compliant. See the Qualtranscribe vs HappyScribe comparison.

Pricing (billed annually, with a yearly pool of minutes):

• Basic: $8.50 a month, 1,440 minutes a year

• Pro: $19 a month, 7,200 minutes a year, three seats

• Business: $59 a month, 72,000 minutes a year, five seats

4. Sonix: solid speaker separation, more substitutions

Sonix separated all six speakers with 97.5% accuracy in about a minute, but it changed "the cooking together thing" to "the working together thing," which alters what the participant was describing.

AI summaries, chapters and sentiment analysis are built in, but Sonix has no human transcription option, and HIPAA with a signed BAA is available only on Enterprise. The Qualtranscribe vs Sonix comparison covers the rest.

Pricing:

• Pay as you go: $10 an hour

• Core: $25 a month, 5 hours

• Advanced: $50 a month, 20 hours

• Pro: $80 a month, 40 hours

5. Amberscript: strong speaker labels, but stopped early on the free plan

On the part it covered, Amberscript attributed 99.8% of words to the right speaker, but the free transcript stopped at 6 minutes 34 seconds, so its scores reflect only two-thirds of the clip.

Amberscript is ISO 27001 and ISO 9001 certified, GDPR compliant and stores files in Europe, and it also sells human-made transcripts.

Pricing (machine-made, yearly billing saves 17%):

• Starter: €19 a month, 5 hours

• Pro: €29 a month, 10 hours

• Power: €49 a month, 25 hours

6. TurboScribe: top raw word score, no speaker labels by default

TurboScribe had the highest raw word accuracy in the test at 96.7%, but at default settings it returned no speaker labels, because speaker recognition is an optional checkbox that has to be ticked before upload. It also wrote "night out" for "when it's nice out."

TurboScribe says it does not train AI models on customer files or transcripts. Its security pages do not list HIPAA, and it has no research analysis features.

Pricing:

• Free: 3 transcripts a day, up to 30 minutes each, lower priority

• Unlimited: $10 a month billed yearly or $20 month to month, uploads up to 10 hours

7. Otter.ai: built for meetings, not focus group recordings

Otter scored 93.8% on words but labelled every line "Unknown Speaker," so the transcript could not be coded by participant, and it made the most meaning-changing errors in the test, including "a lot of extra data" for "a lot of exercise."

Otter is designed for live meetings on Zoom, Teams and Google Meet. It transcribes six languages, offers HIPAA only as an Enterprise add-on, and its privacy policy lists training its AI on de-identified recordings and transcriptions, which is worth checking against any client confidentiality agreement.

Pricing:

• Basic: free, 300 minutes a month, 30 minutes per conversation, 3 file imports in total

• Pro: $16.99 a month ($8.33 billed yearly), 1,200 minutes, 90 minutes per conversation, 10 imports a month

• Business: $30 a month ($19.99 billed yearly), 4 hours per conversation

Which Tool Fits Which Research Team

For market research agencies running IDIs and focus groups for clients, Instant Draft on Researcher+ gives five seats on one account, reliable speaker labels, NDAs and DPAs on request, and Smart Insights to move from transcript to debrief and topline faster.

In-house insights teams with health or regulated clients will find that Instant Draft's Pro plan is the lowest-priced plan in this test that publishes HIPAA and GDPR coverage. Rev, Sonix and Otter reserve HIPAA for enterprise or custom plans, and HappyScribe does not offer it.

Multi-country studies can run through Instant Draft's 99+ AI languages with human transcription in 25 languages on the same account. HappyScribe is a reasonable alternative for EU-only studies with no health data.

IDI-heavy studies, including B2B and stakeholder interviews, need clean moderator and respondent labels and enough monthly minutes for 20 or more interviews. Instant Draft's Pro+ and Researcher+ plans cover that volume, and Smart Insights pulls key quotes and themes from each interview as soon as its transcript is ready, which gives a head start on analyzing interview transcripts.

Solo researchers on a budget transcribing one-to-one interviews will find TurboScribe's flat yearly price hard to beat, provided speaker recognition is switched on. Instant Draft's Free plan is the better place to start if the work includes group sessions or needs analysis.

Whatever tool you choose, check the names and any quote headed for a client report against the recording. A Quirk's article on AI for insight teams gives the same advice, describing transcription as a low-risk place to start and recommending regular review of AI output. Before signing up, ESOMAR's 20 Questions to Help Buyers of AI-Based Services covers what to ask a vendor about data ownership and human oversight.

Sessions where every word will be quoted are where human transcription still earns its place, and the guide to the best human transcription services for focus groups covers that side of the decision.

Try Instant Draft free with 75 minutes and no credit card.

Frequently Asked Questions

What is the best AI transcription tool for IDIs and focus groups? In a test of seven tools on a six-person focus group, Instant Draft gave the strongest all-round result: fastest transcript, 99.8% speaker accuracy, five of six names correct, the fewest meaning-changing errors and an automatic research summary.

Which AI transcription tool is most accurate? On raw word accuracy, TurboScribe (96.7%) and Rev (96.5%) scored highest, with Instant Draft at 95.1%. Most of Instant Draft's gap came from keeping false starts and repeated phrases that people actually said. On speaker labels, Instant Draft, Rev, HappyScribe and Amberscript led, and Instant Draft tied for the most names correct.

Do focus group test results apply to in-depth interviews? Yes. A focus group is the harder case, with six voices, crosstalk and more names, so a tool that separates six speakers accurately will separate a moderator and one respondent. For IDIs, also check per-file length limits and whether monthly minutes cover the full study.

How many AI transcription minutes does an IDI study need? A study of 25 one-hour interviews needs about 1,500 minutes. Instant Draft's Pro+ plan includes 1,200 minutes a month and Researcher+ includes 3,000, while Otter's Pro plan limits uploads to 10 files a month.

Does AI transcription work for TDIs and shop-alongs? Yes, but expect lower accuracy than in a quiet interview room. Phone lines compress voices and shop-along recordings pick up store noise, so check names and verbatims against the recording before they go into a report.

Can AI transcription identify speakers in a focus group? Five of the seven tools separated all six speakers. Otter labelled everyone "Unknown Speaker," and TurboScribe only separates speakers if the option is switched on before upload.

Is AI transcription accurate enough for client reports? AI drafts are accurate enough for analysis and internal coding, but every tool in the test got at least one name wrong. Check names and any quote going into a client report against the recording, or order human transcription for sessions that will be quoted extensively.

Which AI transcription tools are HIPAA compliant? Qualtranscribe's Instant Draft is HIPAA compliant from the Pro plan at $18 a month. Rev includes HIPAA on its custom Unlimited plan, Sonix and Otter on Enterprise, and HappyScribe is not HIPAA compliant.

Does AI transcription work for multilingual market research? Instant Draft transcribes 99+ languages and offers human transcription with translation in 25 languages. Otter supports six languages, while HappyScribe covers 150+ and TurboScribe 98+ for AI transcription.