Subtext Exercise Generator

What the prompt is for

Subtext needs a want and a reason to route around it. Staring at a blank page produces either on-the-nose speeches or a mood with no spine. A dealt prompt is a spine you didn’t have to invent before coffee. The wants are hidden from the dialogue, not from you. You should know that one person is trying to leave with the keys. They should talk about the kettle, the train, the form, unless the constraint says otherwise.

The obstacle is why plain speech would cost them. A child in the next room, a recording, a promise made to someone else, pride. If you ignore the obstacle, you will write a clean argument and learn nothing. The constraint is a game rule: no one may say sorry, or the last line has to be about an object. Constraints are how you find a tactic your habits wouldn’t have offered.

The people are jobs and bonds, not names, so you aren’t tempted to reach for a stock couple you already write. Cast them as you like. Change the gender, the ages, the decade. Do not change the incompatibility. If both wants can be satisfied by the same action, you dealt yourself a cooperation scene and should draw again or sharpen one want until it hurts.

How to write the ten minutes

Set a timer. Ten minutes. Write in the medium you actually work in — prose quotes, screenplay cues, stage names — because subtext’s size changes with the room. Do not stop to repair punctuation. When the timer ends, underline the one line that most clearly risks saying the hidden want, and the one object that carries it instead. If you have only the line and no object, the scene is a confession. Confessions are allowed. They are not practice at subtext.

Then mark the streak on the workbook if you want a reason to come back tomorrow. The streak lives in this browser and nowhere else. It is not an account. It will not survive a cleared cache, which is a feature: the Lab is not building a profile of your practice. The reps are the point. The dots are a nudge.

If a prompt feels offensive or lazy — two people reduced to a job in a way that sneers — throw it out. The dataset is a deck, not a canon. Draw another seed. A generator that can’t be refused is a bad collaborator, and this one is only a deck of situations with the serial numbers showing.

What it refuses to do

It will not write the scene for you. That is the whole difference between a prompt and a model that returns a polished, average argument in which both people are strangely reasonable. If you want that failure mode examined, the AI essays on this site are blunt about it. Here, the blank page stays blank until you put a tactic on it.

It does not know your novel’s continuity. A prompt about a funeral may be useless in a book set on a ship. Take the structure — incompatible wants, a cost, a constraint — and swap the furniture. The seed is a starting constraint, not a scene you owe anyone. Share the seed with a writing partner if you want the same furniture. They still won’t see your pages, because you never typed them here.

It is not therapy and not a dare to dramatise real people who haven’t agreed to be in your book. If the dealt situation is close to a private wound, you may write it and you may also close the tab. Memoir dialogue has its own ethics page. A random prompt is a poor excuse to put a real person’s words in quotation marks they never said.

A teaching note

In a class, deal one seed to the room. The scenes will not match, and that is the lesson: situation is not voice. Read two aloud and ask which tactics showed up. You will hear probe, deflect, soothe, accuse, trade. Name the tactics. Students who can name a tactic can revise. Students who can only say “make it more natural” cannot.

Pair the generator with the annotated subtext scene when someone says they don’t understand what they were supposed to avoid. The annotation shows text, subtext, and context on a finished page. The generator supplies the unfinished one. Together they are a workshop hour. Apart, each is only a hint.

A worked ten minutes

Draw a prompt and write the hidden wants on paper before you write a line, then turn the paper over. If you leave the wants visible, you will write a scene that explains them, which is the failure the exercise exists to prevent. Set a timer for ten minutes. Two people, the constraint on the card, no greeting, no apology at the end. When the timer goes, stop even if the last line is mid-thought. Unfinished is useful. A tidy ending is often the synopsis sneaking back in.

Turn the paper over and mark any line that states a want in the want’s own words. Cut those first. If the scene collapses, the context was never on the page — the reader, who is you, was supplying it from the card. Put one checkable fact back into the action, not into a speech. A key, a roster, a glass, a badge. Run the ten minutes again with the same seed if you want the same pressure, or draw again if the place was fighting your book. The seed is shareable. The lines you wrote are not in the link. That split is deliberate. A workshop can share a prompt without anyone handing over unfinished pages.

If you use a model to ‘finish’ the exercise, you have skipped the exercise. A model will reconcile the two wants, because reconciliation looks like competence. Your job is to keep the wants incompatible for as long as the scene can bear, and then to let a cost land. The AI guide on this site is about using a machine as a diagnostic after you have a draft, not as the drafter of the drill. Do the drill yourself. The generator will still be here tomorrow, which is the point of a local deck.