Chat Picker

AI Chat Tools: Building Your Own Ranking by Features and Price

Vendor pages document different prices, usage rules, data controls, export terms, and support, so an afternoon test can help you rank ChatGPT, Claude, and Gemini by your own requirements.

Sources checked 3 Oct 2026

Build your ranking by giving ChatGPT, Claude, and Gemini the same tasks, logging what you can verify, and applying your own must-haves and spending ceiling. Vendor pages can settle listed prices, access, usage rules, data controls, exports, and support, but not which assistant fits your work; Chat Picker has not tested them. An earlier version used unsourced statistics and test results; those claims have been removed.

Know these limits before you start

As read on October 1, 2026, OpenAI’s Plus Help page listed Plus at $20 a month, Anthropic’s Pro Help page listed Pro at $20 a month in the US, and Google’s US AI plans listed AI Pro at $19.99 a month. The OpenAI pricing page, also read October 1, 2026, showed Go at A$13 a month on its Australian page and said it may include ads; Chat Picker did not verify a US-dollar figure under its sourcing method. Recheck before paying because prices and regions differ.

Usage labels are not directly comparable. OpenAI’s pricing page gives no exact message counts, and its Plus Help page says caps can vary with system conditions. Anthropic’s Pro Help page says use varies with message, file, and conversation length and the model or feature; its session limit resets every five hours, with a weekly limit. Google’s Gemini Apps limits Help page says limits depend on prompt complexity, models or features, and chat length, refreshing every five hours until the weekly cap. These vendor pages were read October 1, 2026.

Data controls need the same treatment. OpenAI’s data controls page, read October 1, 2026, says personal Free, Go, Plus, and Pro users can turn off model improvement; temporary chats are not used to improve models but may be retained for safety for up to 30 days. Anthropic’s privacy page says chats and coding sessions are used to improve Claude when the user allows it, while Incognito chats are not. The Gemini plans page and Gemini Apps Privacy Hub do not state a plan-level training switch, so mark it unverified.

OpenAI warns that ChatGPT can be incorrect or misleading; Anthropic gives the same warning for Claude; Google says Gemini Apps sometimes show sources and related content. Your log must separate an answer from evidence you opened and checked.

Check support before paying. OpenAI Support starts with a virtual assistant, and a human may join if it cannot resolve the issue; the OpenAI Developer Community is not for ChatGPT app discussion. Claude support is asynchronous and written, with no phone or live-chat support. Google’s related-sources Help page says Google One members can get support from an expert, while the Gemini Apps Community lists community posts and guides. These pages were read October 1, 2026.

Claude’s export Help page, read October 1, says individual Free, Pro, and Max users can export data, but its link expires after 24 hours. OpenAI’s cancellation Help page, read the same day, says uninstalling does not cancel a plan, cancellation stops future renewals but does not automatically refund prior charges, and store-billed subscriptions must be canceled through the store account used to subscribe. The Gemini plans page does not settle export or cancellation terms.

The test plan

Run each step in a fresh chat with identical inputs. Use one log row for the assistant, plan, task, observed result, checked evidence, limit, cost, and unresolved issue. Use synthetic files, not material you are not allowed to share.

  1. Text baseline. Prepare a one-page memo with five dated facts and two contradictions. Send: “Read the attached memo. Identify the facts and contradictions, and cite the section for every claim.” Record time, omissions or unsupported claims, and errors found when you check the memo.

  2. Files and images. Prepare a two-page PDF and a screenshot with the same five labeled values, changing one value in the screenshot. Send: “Compare both files. List each value, its location, and any mismatch; mark unreadable evidence.” Record upload acceptance, extraction errors, and whether citations identify the evidence.

  3. Current information. Prepare a dated packet with one primary document and one news summary about the same change. Send: “Using only the attached files, state what each source establishes, give both publication dates, and flag conflicts. Do not add facts from memory.” Record missing dates, unsupported additions, and whether each answer traces to the packet.

  4. Workflow controls. Prepare a five-item neighborhood book-drive launch plan. Send: “Create a project called Neighborhood Book Drive, save the launch checklist, and list actions that need a connected tool.” Record whether project, file, memory, temporary-chat, or Incognito controls appear. Mark an absent control “not shown,” not “unavailable.”

  5. Price and operations. Make a needs list covering source links, files, a reusable project, export, and cancellation. Attach each service’s current plan and support pages, plus any export or cancellation pages it provides. Send: “Using only the attached pages, mark each need documented, not listed, or not found; quote the relevant text and do not guess.” Record currency, billing cadence, seat minimum, published limit or vague label, training setting, support channel, export route, and cancellation route.

How to read your results

Turn the log into a decision in this order. Label each non-negotiable need met, not met, or unresolved. Remove options that miss a requirement, exceed your ceiling, or fail a required data, export, or cancellation condition. Compare only claims you independently verified; treat a missing citation as unresolved, not proof of an error. Separate output quality from plan limits. Repeat a decisive task only when a limit, source, or plan may have changed. Select the lowest documented cost among options that meet your requirements, or change the workflow if none does. This ranks the options for your use; it does not show that one assistant is generally better.

Where the plans differ

On the Claude vs ChatGPT vs Gemini matrix, inspect the rows for Free plan, low-cost tier, main paid plan, heavy-use plans, team plan, context window in the app, usage limits, training on chats, and ads. Start with the plan and usage rows that match your test, then check data, export, cancellation, and support pages for anything unresolved. Under Chat Picker’s method, an unconfirmed figure is marked “Not verified” and left blank. Treat that blank as unknown, not as zero or evidence that a feature is unavailable. Use the matrix to set your shortlist and spending check; use your own log to make the final decision.

Sources

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

Compare assistants