Chat Picker

AI Chats in Screenwriting: Character Building and Plot Ideation Skills

A practical afternoon test compares ChatGPT, Claude, and Gemini on character depth, plot logic, dialogue, consistency, and workflow fit using documented vendor limits rather than outside scores.

Sources checked 2 Oct 2026

ChatGPT, Claude, and Gemini can be compared for screenwriting by testing character building and plot ideation, but their vendor pages do not establish which writes better scripts. Those pages document features, input limits, usage allowances, data settings, and accuracy warnings that shape the test. Chat Picker has not tested, scored, or ranked the assistants for screenwriting; unsourced statistics and test results from the earlier version have been removed.

Know these limits before you start

Check inputs before using unpublished material. As read on October 1, 2026, OpenAI’s file uploads FAQ gives each file a 512 MB hard limit and caps text or documents at 2M tokens. Anthropic’s upload guide gives 500 MB per chat file and 30 MB per project file, with possible extra token limits; Google’s Gemini file-upload guide gives up to 10 files per prompt and 100 MB per non-video file.

For long work, Anthropic’s pricing page, read on the same date, lists context of up to 1M on every Claude plan, varying by model. On that date, OpenAI’s pricing page lists windows by plan and model, while Google’s US plans page gives no app token count.

Accuracy also needs checking. OpenAI’s accuracy note says ChatGPT can be incorrect or misleading and may sound confident when wrong. Anthropic’s incorrect-response guidance says not to rely on Claude as the only source of truth. Google’s Gemini guidelines describe outputs as probabilistic and include artistic context among their considerations.

Before uploading a script, check data settings: OpenAI’s pricing page says an opt-out is available on Free, Go, Plus, and Pro; Anthropic’s pricing page says Free, Pro, and Max have an opt-out, while Team is not trained on by default; Google’s US plans page does not state a training setting. Anthropic’s Usage Policy requires AI involvement to be disclosed at the beginning of a session when model output is presented directly to consumers.

The test plan

Use the same brief, plan level, feature state, and time budget for each assistant. The packet below is fictional. Save each prompt before starting, record the plan and feature labels shown, and log the first response before asking for a revision.

  1. Character depth. Give each assistant this prompt: “Create a character bible from this brief: Mara Vale is a ferry dispatcher who hears approaching engines through a damaged radio. She wants to stop her brother from sailing, but each rescue reveals that he has sabotaged the same route. Give every character a want, fear, secret, contradiction, and scene behavior. Use the same fields throughout and mark assumptions.” Look for motives that drive action and contradictions that create scenes. Record format use, repeated traits, added constraints, and contradictions.

  2. Plot logic and pacing. Give this prompt: “Create an act-based plot outline from this brief: Mara Vale is a ferry dispatcher who hears approaching engines through a damaged radio. She wants to stop her brother from sailing, but each rescue reveals that he has sabotaged the same route. For each beat, give the trigger, choice, consequence, and new question. Make the brother’s motive progress, and end on a decision that changes Mara’s relationship with him. Do not write dialogue.” Look for cause and effect, repeated turns, and reliance on coincidence. Record missing links, motive resets, and unmotivated changes.

  3. Dialogue and subtext. Give this prompt: “Write a ferry control-room scene from this input: Mara wants to keep her brother from sailing; he wants her to admit she heard the same incorrect engine signal months earlier. Neither may state the full truth. Let each line pursue a tactic, allow interruptions, and reveal subtext through action rather than explanation.” Look for power shifts and implied accusations. Record exposition lines, overly polished speech, and changes in the scene objective.

  4. Long-form consistency. Prepare a short treatment with the character bible, outline, and scene, then upload them through the same method for each assistant. Give this prompt: “Read the attached files. List every contradiction in names, motives, object locations, and timeline. Quote each conflicting passage, identify its source, and propose the smallest repair. Do not silently rewrite.” Look for traceability to the supplied text. Record the first conflict, missed conflicts, unrelated changes, and whether each repair remains traceable.

  5. Workflow handoff. Prepare one packet containing the locked premise, current outline, latest scene, and revision request. Give this prompt: “Create a handoff checklist. Separate locked facts, proposed changes, unresolved questions, and rejected ideas. For each proposed change, state which earlier passage it affects. End with the next scene’s dramatic objective and immediate obstacle.” Look for a revision trail rather than another broad rewrite. Record manual transfers, lost constraints, unsupported changes, and passages the assistant cannot locate.

How to read your results

Before reading, write your non-negotiables: locked story facts, requested structure, no unrequested rewrite, and traceable revisions. Mark each response “meets constraints,” “needs revision,” or “unusable.” Treat an invention as a revision option unless it breaks locked facts; treat contradictions, broken instructions, or invented sources as failures for that task.

Separate fit from value. Fit depends on your requirements and preferred voice. Value depends on the plan price, input ceilings, usage allowance, data setting, and workflow features you will actually use. If prompt wording changes an output, record that change rather than treating it as proof of a permanent capability.

Where the plans differ

In Chat Picker’s Claude vs ChatGPT vs Gemini matrix, check the rows Free plan, Low-cost tier, Main paid plan, Heavy-use plans, Context window in the app, How usage limits are described, Training on your chats, and Ads. Confirmed figures carry a vendor source and read date; unconfirmed cells say “Not verified” and stay blank. Confirm the vendor page before subscribing because plans and regional prices change.

Sources

For current prices and limits, see the dated comparison pages.

Compare assistants