Can you use AI for transcription and translation, or do you have to hire a team?

The question we get more than any other. Here is the honest, language-by-language answer, in tables you can actually scan.

You have research to run in Swahili, or isiZulu, or Hausa, or across a dozen markets at once, and you want to know one thing: can AI transcribe and translate it, or do you still need to hire a room full of people to do it by hand?

The language table

Two separate ratings, because they are two separate jobs. Transcription is turning a voice note into text in the same language. Translation is turning that text into English. Transcription is the harder job and where nearly all the difficulty lives. Translation runs about a tier ahead almost everywhere.

Grouped by region, with the languages most relevant to emerging-market fieldwork first. "Approach" is what we would actually reach for.

One thing about these ratings: they move. Investment in African and emerging-market languages has accelerated sharply, so a language sitting in "Hard" today can be "Workable" within a year. The floor is rising fast, and faster for the languages with the most speakers.

Which model for which job

"Use AI" is not one decision. The engines differ enormously, and picking the wrong one is the most common reason a multilingual study underperforms. A serious operation routes each language to its best model rather than forcing everything through one provider. This is the difference behind the "recommended approach" column above.

No single model wins everywhere. For translation the same holds: Google Translate has the widest coverage, Meta's open NLLB is strong on low-resource languages, and modern LLMs (Claude, GPT, Gemini) now match or beat dedicated engines on the top languages and handle slang and mixed text better, while weakening on the minority tail.

Code-switching is the thing that breaks everything

Here is the failure mode that catches people out, because it shows up in no benchmark. Real people do not speak one clean language at a time. A respondent in Nairobi slides between Swahili, English, and Sheng in one thought. In Johannesburg it is English threaded through isiZulu. In Lagos, English, Yoruba, and Pidgin at once. In India, Hinglish and Tanglish.

Almost every speech model is trained on the assumption of one language per utterance, so code-switching is exactly where they quietly fall apart. And in emerging markets this is not an edge case. For a huge share of urban respondents, mixing is simply how they talk.

The silver lining: large language models handle code-switched text far better than speech engines handle code-switched audio, because they trained on the messy way people actually write. So the winning pattern is a decent transcript first, then an LLM to clean, interpret, and translate the mix, rather than expecting the speech engine to nail it in one pass. Even so, this is the single biggest reason to keep a human in the loop on the harder tiers. Code-switching is where the machine is most confidently wrong.

The recording environment matters more than the model

Benchmark accuracy is measured on clean studio audio. Your data is a voice note recorded on a cheap phone, on a taxi, in a market, with wind and a TV in the background, then squeezed through WhatsApp compression.

In the real world, expect error rates roughly 1.5 to 2 times worse than any benchmark number, purely from the audio conditions. That is not the model failing, it is physics. Two things follow. Denoising before transcription genuinely moves the numbers, which is why models with built-in cleanup have a quiet advantage on field audio. And a little guidance to respondents (find a quiet spot, hold the phone close) improves data quality more cheaply than any model upgrade. The environment is the variable people ignore and then blame the AI for.

So, AI or a team?

Put it together and the answer is not "AI" or "humans." It is a ratio that shifts with the job.

The old model, a large human team transcribing and translating everything by hand, is now the wrong answer for almost every study. It is slow, it does not scale, and it is expensive where it does not need to be. But pure AI with no human anywhere is equally wrong for the hard cases, because it fails silently and you find out when a client questions a finding you cannot defend.

The right answer is AI-first, with humans placed exactly where the technology is weak. Route each language to its best model, clean the audio, let AI carry the volume, and spend your human hours on the specific failure points: the code-switching, the low-resource languages, the nuance that carries the insight. That is how you get research in people's own words, at scale, without either burning budget or quietly shipping bad data.

"We can only do this in English" is no longer a real constraint. Which languages, in which mode, with how much human review, is now a strategy decision. And it is one worth making on purpose.