ChatGPT vs C
ChatGPT vs Claude vs Gemini: AI Assistants Compared in Creative Writing
Compare documented plan, limit, and data-setting differences, then run a practical creative-writing test across ChatGPT, Claude, and Gemini.
Sources checked 2 Oct 2026
For creative writing, the vendors’ own pages document some practical boundaries, including plan access, context, usage limits, data settings, and policy limits, but they do not rank creative-writing quality. Chat Picker has not tested ChatGPT, Claude, or Gemini for this use. This rewrite removes older statistics and test results that lacked sources; what follows is a test plan you can run in an afternoon.
Know these limits before you start
The vendor figures below were as read on October 1, 2026. OpenAI’s ChatGPT pricing page lists app context windows of 27K for Instant on Free, 54K for Go and Plus, and 128K for Pro; its reasoning row lists 256K for Go and Plus and 400K for Pro. Anthropic’s Claude pricing page lists context of up to 1M on every Claude plan, varying by model. Google’s US Google AI plans page gives no app token context figure.
Neither OpenAI nor Anthropic publishes exact message counts on its pricing page. Google describes compute-based limits that refresh every 5 hours, up to a weekly limit, on its Google AI plans page.
For chat data, OpenAI says an opt-out is available on Free, Go, Plus, and Pro. Anthropic says Free, Pro, and Max have an opt-out, while Team is not trained on by default. Google’s plans page states no training setting.
For publication workflows, OpenAI’s Usage Policies prohibit using someone’s photorealistic image or voice without consent in ways that could confuse authenticity. Anthropic’s Usage Policy says consumer-facing chatbots must disclose AI involvement at the start of each session and calls for relevant human expertise in elevated-risk uses. Google’s Gemini guidelines consider artistic context but warn that Gemini may sometimes violate guidelines, reflect limited viewpoints, or overgeneralize. Keep test material fictional and check factual details independently before publication.
The test plan
Use one common packet in a separate conversation for every run. Paste this premise: “Leena, a night-shift archivist, finds a film reel showing a possible future in which several people disappear, including her sister. The final reel is blank. Leena’s editor wants every copy destroyed before dawn but refuses to say who made the reel.”
Start a log with these fields: assistant, plan, prompt, tools enabled, constraints met, quoted issue, revision needed, and notes on structure, voice, and originality. Record errors and limit messages instead of inferring their cause.
-
Narrative coherence. Input: the common packet. Instruction: “Write a sustained short story that begins when Leena finds the blank reel, explains why it is blank without resolving Leena’s decision, and keeps every cause and effect consistent. End before the decision.” Look for: chronology, motivation, setup and payoff, and contradictions. Log: a short structure summary and exact lines supporting each problem.
-
Prompt adherence. Input: the same packet. Instruction: “Write for an adult literary magazine. Name only Leena and her editor. Begin in the archive at night, use no flashback, and end on a concrete image that does not reveal Leena’s choice. Then list every constraint you followed.” Look for: actual compliance, not the assistant’s claim. Log: mark each constraint met or missed and quote the relevant passage.
-
Creative originality: breadth and depth. Input: the same packet. Instruction: “Offer alternative opening passages, each using a different narrative strategy. Avoid the premise’s wording, and do not repeat the same image, conflict, or structure across the alternatives.” Look for: distinct approaches, developed depth, specificity, and clichés. Log: label each strategy in your own words and quote repeated language.
-
Dialogue and voice consistency. Input: the archive confrontation. Instruction: “Write the scene between Leena and her editor. Leena uses short, exact questions and notices physical detail. The editor uses long, reassuring sentences and avoids naming the future. Keep the dialogue intelligible without speaker tags, and do not let either speaker state information only they could know.” Look for: voices converging, unmotivated exposition, or knowledge that appears from nowhere. Log: quote the line where voice or knowledge breaks.
-
Poetry and lyrical writing. Input: the blank film reel. Instruction: “Write a lyric poem that moves from observed details to fear, avoids the abstractions fate, stars, and forever, and ends on an image rather than a conclusion.” Look for: image specificity, transitions, repeated abstractions, and familiar phrasing. Log: preserve lines that work and lines that weaken the progression.
-
Marketing copy and persuasive writing. Input: a fictional event brief. Instruction: “Write launch copy for a fictional public event called The Last Reel, an after-hours archive night with short readings, projected urban photographs, and a question-and-answer session. Address new visitors, keep the language plain, include a clear invitation, and invent no attendance figures, awards, testimonials, or guarantees.” Look for: audience fit, specificity, hierarchy, unsupported claims, and a clear call to action. Log: quote phrases that persuade, distract, or need revision.
Give every output the same revision request: “Preserve the premise and voice, fix continuity errors, remove unsupported claims, and replace generic phrasing with concrete detail. Explain each change in a brief list.” Record whether the revision addressed the logged issue without erasing a strength.
How to read your results
Start with constraints, not impressions. Discard any run that breached your data choice, hit a plan limit, failed a nonnegotiable instruction, or ran into a documented policy issue. Then compare quoted evidence for continuity, voice, originality, factual care, and audience fit. Do not treat length, polish, or confidence as proof of quality.
Use the revision log to see which weaknesses were repairable without losing strengths. Choose “keep,” “revise,” or “discard” for your workflow, not as a universal ranking, and confirm current plan terms before subscribing.
Where the plans differ
Open the three-way comparison matrix and focus on the rows for Free plan, Low-cost tier, Main paid plan, Heavy-use plans, Team plan if applicable, Context window in the app, How usage limits are described, Training on your chats, and Ads. Each matrix figure carries its vendor source and read date.
For individual use, OpenAI’s pricing page organizes access around Free, Go, Plus, and Pro tiers; Anthropic’s page lists Free, Pro, and Max; Google’s page lists Free, Google AI Plus, Google AI Pro, and Ultra. Check the context and usage rows before working with long drafts, and review the training row before using sensitive source material. OpenAI says Go may include ads, while the Anthropic and Google pricing pages make no ad statement that could be verified. Chat Picker’s method page leaves unconfirmed cells blank rather than filling them with estimates.