CART captioning and AI captions both put spoken words on a screen in real time. The difference is who writes them. With CART, a trained human captioner types every word as it is spoken. With AI captions, software listens and writes. That single difference changes the accuracy you can expect, how far ahead you book, how the service scales across rooms, and whether the captions can be read in another language.
What CART captioning is
CART stands for Communication Access Realtime Translation. Despite the word "translation", it is not about changing languages. It means turning speech into text, live, so that people who are deaf or hard of hearing can follow what is said.
A CART captioner works on a stenotype, a special keyboard on which several keys are pressed together to write whole syllables or words in one stroke. That is how a person can keep pace with natural speech. The captioner can sit in the room or work remotely from an audio feed. The text appears on a screen, a laptop or a personal device.
In France, the same service is known as vélotypie, or live subtitling by stenotype. The principle does not change: a professional writes what is said, as it is said.
CART is booked in advance. Captioners are specialists, and for a technical event they usually ask for the agenda, the slides and the speakers' names ahead of time, so they can prepare their personal dictionary.
What AI captions are
AI captions rely on automatic speech recognition. The software receives the room's audio and writes the text a moment later. Current systems also add punctuation, detect when the speaker changes and, in some tools, translate the text into other languages on the fly.
Nobody has to be booked. The service starts when the session starts and can run in as many rooms as the programme needs.
The differences that matter in practice
Judgment. A human captioner understands context. They can tell two similar-sounding words apart from the meaning of the sentence, mark a change of speaker, signal laughter or applause, and correct themselves on the fly. AI has improved a lot, but it still makes mistakes a human would not, especially with names, strong accents, people talking over each other and poor audio.
Preparation. Both get better with it. A CART captioner builds entries for the event's terms in their dictionary. AI tools increasingly accept a glossary of product names and jargon before the session.
Language. CART works in the language being spoken. Producing text in another language is a different job, translation or interpretation. AI captions can be translated automatically, so one talk can be read in many languages at once. Glot, for instance, shows each attendee subtitles in their own language, chosen from more than 60.
Availability and scale. CART depends on qualified people being free, which can be hard on busy dates or for less common languages. A conference with parallel tracks needs a captioner in every room, and long days may need a relay. AI captions are available on demand and follow the number of rooms.
Where the text appears. Both can feed a screen in the room. AI tools designed for events often add personal access: attendees scan a QR code and read the captions on their own phone, which helps people sitting far from the screen.
Cost structure. CART is paid for professional time, so the cost follows the number of hours and rooms. AI captions are sold as a service, usually per room and by the day or hour. Rates vary between providers, so compare real quotes for your programme rather than averages.
When CART is the right choice
Choose CART when every word counts and an error would have consequences. If a deaf participant has asked for CART as an accommodation, their request is the best guide: they know what works for them. Legal or medical settings, formal meetings where the record matters, and highly technical talks with dense vocabulary are also strong cases for a human captioner.
CART is also the safer option when the audio is difficult. Overlapping voices in a workshop, a speaker who refuses the microphone, a noisy room: a trained person can often follow what software misses.
When AI captions are the right choice
AI captions make sense when you need coverage more than perfection: many sessions, several rooms, last-minute changes, or an international audience. They are also a practical way to caption content that would otherwise get no captions at all, such as breakout sessions, internal meetings or a small seminar with no budget for a specialist.
For audiences from several countries, translation is often the deciding point. From the same talk, one AI service can give a deaf French speaker French captions and a visitor from Madrid Spanish ones. Glot also writes a summary at the end of each session, which helps anyone who wants to go back over what was said.
The quality of the result depends heavily on the sound. Take the feed from the mixing desk or a good microphone, and test it before the first session.
Using both
The two are not mutually exclusive. Some organisers book CART for the main stage or for a participant who asked for it, and use AI captions in the parallel rooms and for translation. That way the most demanding need gets a human, and nobody else goes without captions.
Whatever you choose, ask at registration whether attendees need captions and in what form. The answers will tell you more than any comparison table. Our article on event accessibility and live captioning covers what European rules now expect from organisers, and this first-hand account shows what a conference feels like from the audience when captions are missing.
Questions to ask before deciding
Who needs the captions, and has anyone requested a specific service? In which languages? How many rooms run at the same time, and for how long? How much notice do you have? What is the audio setup in each room? And where will the audio and text be processed, under which data protection rules? With these answers in hand, the choice between a human captioner, AI captions or a mix of the two usually becomes clear.
Live captions in every room, in each attendee's language.
See how Glot works