Where We Stand on AI

Last updated: August 7, 2026

An actor asked us whether importing a script into Subtext meant feeding it to a model. It is a fair question and it deserves a straight answer, so here is the whole position — what the AI does, what it will never be allowed to do, and who it is for.

1. What the AI actually does

Subtext uses AI for exactly two jobs, and neither of them is the acting.

• Reading the script. When you import sides, the text is analyzed to find the characters, the scenes, and where your lines fall. That is a parsing job. It turns a PDF into something you can rehearse against.
• Being the other person in the room at 11pm. The scene-partner voices are synthetic, and they exist because a reader is not always available at the hour you actually work. That is the entire use case: a cue, on demand, so you can get off book.

Your scripts are not training data. Our AI analysis runs on Anthropic's commercial API, whose terms state plainly that Anthropic may not train models on customer content. That is contractual and it is the default — not a setting we had to find. On the voice side we have turned model training off for our account, so neither the lines we send for speech nor the microphone audio used to detect your cue is used to improve anyone's models. The full processor-by-processor breakdown is in our privacy policy.

2. An AI voice will never appear on your tape

There is a category of product now that will read your scene with a synthetic voice and call the result a self-tape. We think that is a bad deal for actors, and we will not build it.

Not on principle alone — on the merits. Acting is reacting, and you cannot react to something that is not there. A synthetic reader gives you correct words on a metronome: no breath, no timing that shifts because of what you just did, nothing to play against. The tape that comes back is technically complete and dramatically dead, and everyone watching it can tell. Casting directors are not confused about what they are seeing. Sending one costs you the room.

It is also a quiet transfer of work away from people. Every synthetic reader on a tape is a reading a human did not get paid for — usually another actor, often a friend, frequently someone earlier in their career than you.

So the AI voice in Subtext is a rehearsal tool with a hard stop. When it is time to put something on tape, we built Send to Reader instead: you send a link, a real person reads with you from wherever they are, and their actual voice — with their actual timing — is what lands on the tape. That feature exists precisely because we refused the easier version.

3. The whole point is getting a human more work

Every feature here is measured against one question: does this get a working actor closer to being hired?

Being off book by Tuesday. Having someone to run it with at midnight. Turning a tape around before the deadline without begging a favor. Walking into the room having already made choices instead of still learning the words. That is the job the tool is doing.

We are not interested in generating performances. There is no version of Subtext that reads the scene for you, submits for you, or stands in for you.

Subtext exists to get you, a human, more work. If a feature ever fails that test, it does not ship.

Questions

If you want more detail on how any of this works — or you think we have got something wrong — write to hello@getsubtext.co. The actor who asked the original question is the reason this page exists.