What happened
Which? Travel asked fluent speakers of Hindi, Italian and Polish to test four free translation apps: Google Translate, Apple Translate, Microsoft Translator and DeepL. The online article is dated 17 July 2025 and first appeared in the January edition of Which? Travel magazine. The testers used each app for a spoken conversation about finding a bus stop and buying a ticket. They also checked the apps' English translations of restaurant menus. Which? found that DeepL was more accurate than the other apps in spoken translation. Its Polish and Italian speakers said that DeepL translated colloquial language more naturally, which made it easier to understand. DeepL was hit and miss with the menus, has no conversation mode and offers only 33 languages. The article does not call the apps AI or give error counts. DeepL says that its translator uses artificial neural networks trained on many millions of translated texts.
Sources
-
Assessment report, by authors with no tie to the system.
Which? (Rory Boland, Editor, Which? Travel). The best translation apps in 2025 that'll cost you nothing. 17 July 2025, first published in the January edition of Which? Travel magazine. Fluent Hindi, Italian and Polish speakers tested four free apps. DeepL review.
Very accurate, it turns out. Our foreign-language speakers found substance to the claim. It wasn't just more accurate than other apps in spoken language translation, but our Polish and Italian speakers both commented on how it translated colloquial language more naturally, making it easier to understand.
-
Other, by the makers.
DeepL SE (DeepL Team). How does DeepL work? Blog post, last updated 31 October 2021. The maker's description of its translation technology.
Like most modern machine translation systems, DeepL Translator uses artificial neural networks to translate text. We train these networks on many millions of translated texts.
Details
AI system
- Workflow
-
Fluent speakers tested the apps on a spoken conversation about finding a bus stop and buying a ticket. The speakers also checked the apps' English translations of restaurant menus.
- Relation to AI
-
Which? Tested the app directly, so the translations that its speakers judged were the app's own output. Its testers compared four apps with one another and did not compare them with a human interpreter. The article does not call the apps AI or show that they learn from data. DeepL states that its translator uses artificial neural networks trained on many millions of translated texts.
Follow-up
- Conditions for replication
-
DeepL needs an internet connection because it cannot be used offline. The language must be among the 33, mostly European, languages that DeepL offers.
Entry details
- Sector
-
Consumer technology (travel translation).
- Purpose
-
DeepL SE built the DeepL app to translate written and spoken language.
31 more questions not yet checked
model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). DeepL translates speech most accurately in a Which? app test (AIBD-1056). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1056.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)