Chat Picker

Choosing AI Collaboration Tools: Multi-User Management and Permissions

A practical afternoon test for multi-user collaboration, including Claude’s documented view and edit project permissions, across ChatGPT, Claude, and Gemini.

Sources checked 2 Oct 2026

Choosing among ChatGPT, Claude, and Gemini for multi-user work means comparing documented seats, sharing controls, data settings, usage limits, and prices; the vendors do not present one identical access model. Their reviewed pages do not fully settle audit logs, data residency, certifications, or native model switching. Chat Picker has not tested the assistants for this use, so this is an afternoon test plan, not a ranking.

Know these limits before you start

Start with documented team pricing. OpenAI’s ChatGPT Business overview, as read on October 2, 2026, lists standard seats at $20/month with annual billing or $25 billed monthly, and premium seats at $100 with annual billing or $125 billed monthly, with at least 2 seats. Anthropic’s Claude pricing page, as read on October 1, 2026, lists Claude Team at the same seat figures for 2 to 150 people. Chat Picker left Gemini’s business price marked Not verified; the reviewed Google AI plans page was read on October 1, 2026.

Sharing controls differ. OpenAI’s Projects page, checked October 1, 2026, says shared projects use project-only memory and cannot access a member’s outside context, but it does not define Claude-style view and edit roles. Claude’s Projects page, checked October 1, defines view members as able to read contents, knowledge, and instructions and chat without editing; edit members can change instructions and knowledge and manage members and settings. Gemini’s Connected Apps help, checked October 1, describes app permissions and supported or unsupported actions, not a full multi-user project role matrix.

For data use, OpenAI’s pricing page, checked October 1, offers opt-out on Free, Go, Plus, and Pro. Its Projects page says Business data is not used for model training by default, and shared personal-project data may be used only when the owner and every contributor enable improvement. Anthropic’s pricing page says Free, Pro, and Max have opt-out and Team is not trained on by default. Chat Picker could not verify a Gemini training setting from the linked Google page.

Use synthetic records. OpenAI’s Usage Policy, checked October 1, 2026, bars unauthorized monitoring or aggregation of private or sensitive information and automated high-stakes decisions in sensitive areas without human review. Anthropic’s Usage Policy, checked the same day, calls for qualified review of covered high-risk advice and disclosure of AI involvement in direct consumer sessions. Gemini’s safety guidelines, also checked that day, warn that outputs are probabilistic and may include overgeneralizations. This is not legal advice.

The test plan

Use the same administrator, contributor, and viewer test identities where those roles exist. Prepare a synthetic file titled “Remote Work Access Policy” for a fictional team, with a contractor code, an approval table, and conflicting wording that requires a source citation.

  1. Check permissions. Invite each identity through the account controls. Then tell the assistant: “Read the Remote Work Access Policy and write an onboarding checklist for the fictional contractor. Put the relevant policy heading beside each rule.” Record the role labels offered and whether each identity can view, chat, upload, edit instructions, manage members, export, or delete. A missing button proves only that the role lacked a visible control.

  2. Look for audit evidence. Start a clean task, record the signed-in identity and visible time, then prompt: “Create a change note for the contractor request, name the policy section used, quote the approval-table row, and mark uncertainty.” Inspect the conversation, project activity, and admin views for user, time, prompt, file access, model or tool activity, export, and retention. Record “visible,” “exportable,” or “not found”; the last does not prove that no backend record exists.

  3. Test context continuity. In one conversation, prompt: “Draft a short remote-access checklist from the uploaded policy, then revise it for a contractor whose access expires today.” Use any native model-switch control shown, without reloading the file. If none appears, paste the exchange into the next assistant and ask: “Continue the checklist and identify which policy details are now missing.” Record whether context, citations, file references, and formatting survive.

  4. Ask about governance. Send each vendor: “Identify the data residency region, subprocessors, chat and file retention, audit-record retention, and compliance certifications that apply to our proposed team plan; mark each item not stated if your documentation does not provide it.” Record the answer, document title, link, date read, and plan covered. The reviewed pages do not settle these points, so a UI check cannot fill the gap.

  5. Price the account shape. Before checkout, ask: “Show the current monthly and annual seat price, minimum and maximum seat count, included usage, reset schedule, overage rule, and renewal terms for our proposed team plan without a promotional price.” Record the plan, seat type, billing basis, exact published limit, administrator control, and unanswered fields. Recheck the vendor page on purchase day; use Chat Picker’s figures to locate the current price, then verify it directly.

How to read your results

Keep one log row per account, role, task, and observation: date and time, assistant, plan, role, exact instruction, required control, observed result, evidence location, and open question. Mark each item “supported,” “blocked,” or “unknown.” Separate an observed control from a vendor-documented right; neither proves behavior in another plan or role.

Begin with non-negotiable controls: required role separation, usable audit evidence, an approved data location, and any certification your organization actually requires. Exclude a candidate only when a requirement is explicitly unavailable or incompatible; keep missing documentation as “unknown” until answered in writing. Then compare the current seat cost and the published usage limit that fits your workflow.

Where the plans differ

In the three-way comparison matrix, start with the “Team plan” row, then read “How usage limits are described,” “Training on your chats,” and “Context window in the app.” If non-admin users are in scope, check “Free plan,” “Low-cost tier,” “Main paid plan,” and “Heavy-use plans” rather than assuming team controls carry over. Treat Not verified as missing evidence, not as zero capability, and confirm availability and pricing with the vendor before subscribing.

Sources

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

Compare assistants