Enhanced Multilingual Support: Non-English Interviews, Delivered in English

We have supported interviews in dozens of languages for as long as we have existed. An expert in Tokyo could be interviewed in Japanese, an expert in São Paulo in Portuguese, and the AI interviewer handled both.

That was never quite the whole job.

The interview would happen in Japanese, and then a Japanese transcript would land in the workspace of an analyst who does not read Japanese. The summary came back in Japanese. The quotes came back in Japanese. Everything worked exactly as specified, and the research team still needed someone bilingual before any of it became usable — which is the constraint they had been trying to avoid.

Conducting the interview and delivering it are two different problems. We had solved the first one. This release solves the second.

What changes

A non-English interview is now delivered as an English transcript by default. Summary, key quotes and quality scoring come back in English too, whatever language was actually spoken.

The original is never discarded. It is retained in full and sits one click away from the translated view — in the app, and through the API for teams that pull transcripts into their own systems. Timestamps are identical across both, so moving between them costs you nothing against the recording.

For a research team, the practical effect is that a Japanese interview stops being a separate workflow. It arrives in the same shape as every other interview, readable by the same people, with the original available when somebody needs to check a word.

You can write the interview in English

The second change is less obvious and, for most teams, the one that removes real friction.

Questions, the call subject and the compliance statement can be written in English on a Japanese interview, and the interviewer will ask them in Japanese. Write them in the target language instead and they are used exactly as written, never re-worded.

This means the person designing the research no longer has to be the person who speaks the language. A research lead who has never written a word of Japanese can scope a Japanese study, and the interviewer conducts it properly.

Two things are deliberately left alone: company names and expert names. A translated company name refers to a company that does not exist, and an expert hearing their own employer rendered into another language would not recognise it.

The expert's side

The interface the expert sees can also be in their language — buttons, status messages, microphone and connection prompts — set by a parameter on the join link, or chosen by the expert on the page.

This is deliberately separate from the language the interviewer speaks. The two are not always the same. An expert may be entirely comfortable being interviewed in Japanese while preferring an English interface, or the reverse, and neither should be inferred from the other. It costs nothing to let them choose.

Language coverage has also grown, and the interviewer now works across more than thirty languages.

What this does not solve

Translations are machine-generated. They preserve who said what, in what order, and at what point in the recording — but they are not reviewed by a human translator, and we would rather say so plainly than let anyone discover it in front of a client.

For most of what research teams do with a transcript — scanning for coverage, pulling themes, deciding which calls to read closely — that is the right trade. For direct quotation it is not. If an expert's words are going in front of a client with quotation marks around them, quote from the original. That is what the original is there for, and why it is retained rather than treated as a disposable intermediate.

This is also why the translated view says what it is on screen, and why exported files carry that provenance inside the file. A transcript outlives the interface that explained it. Once a file has been forwarded twice and pasted into a deck, nothing about it says whether those were the expert's words or a machine's rendering of them — unless the file says so.

Why this matters more than the language count

Language coverage is easy to claim and easy to match. What determines whether a team actually runs a study in a market is something narrower: whether the output lands in a form they can use on Monday morning without booking a translator.

A study that would have been scoped to North America and Western Europe because that is where the reading capacity sits can now be scoped to where the supply chain actually runs. The interviews were always possible. The difference is that the findings now arrive ready to read.

For existing customers

Nothing about your English interviews changes — same transcript shape, same exports, same API responses, byte for byte. Everything described here applies only to interviews conducted in another language, and only where a translation exists.

If you have been running non-English interviews and handling the translation on your side, that step is no longer yours to carry.