How we use AI

Where AI helps, where it never touches the work, and what we promise about your files.

The short version

AI makes some of our work faster and cheaper, but nothing goes out unchecked. Every job that leaves Mangrol has been read, watched or heard by a person who knows the language and the subject.

You choose how your job is made, and the quote says which way it will be done.

Two ways to work

Human-led. Our people do every step. A native linguist, writer, artist or editor does the work, and a second person reviews it. AI is used only for spell checks and file clean-up.

AI-assisted. AI does the first draft, the first voice or the first cut. Our team then edits every line, checks every name and number, and signs it off. It is faster and costs less.

Where AI helps, service by service

  • Translation: a machine draft that a native linguist edits line by line. Never for legal, medical or literary work unless you ask.
  • Subtitles: machine transcription and timing, corrected by a linguist.
  • Voice: AI voices for volume and quick updates; voice clones only with the person’s written consent.
  • Film: concept frames, backgrounds, some shots, music beds and versions in other languages. The script, direction and edit stay with people.
  • Reels: finding the best moments in long videos, and first-pass captions.

Where AI is never used

  • Story Doctor. Your script is read by Sanjay alone.
  • Rare languages and scripts such as Kutchi, Garo, Mizo, Meitei and Modi script. AI is not good enough in them.
  • Certified, legal and medical translation, unless you ask for it and accept the risk in writing.

Your files

  • We do not put confidential files into public AI tools.
  • When we use AI, we use tools that do not keep your files or train on them.
  • Tell us at the start if your material is confidential, and we will keep the job Human-led.
  • We sign an NDA if you ask.

Tools change

AI tools improve every few months, so we test new ones often and change what we use. This page is updated when we do. Last reviewed: November 2026.

Try it yourself. Four machine translations, one mistake each. Can you spot them?