# Project Tasks > Do a task yourself, give it to someone, have Groupthink do it, or hand it to ChatGPT, Claude or Dots. Source: https://groupthink.com/docs/project_tasks/ Last updated: 2026-10-18 Every project page has a **Tasks** section. It lists the project's tasks, with subtasks under the task they belong to. Each task has a **Who does it** button that says who is doing it: you, a person, Groupthink, or an assistant like ChatGPT. ## Add a task Type in **Add a task to this project…** and click **Add**. To add a subtask, click the arrow icon on a task row (it reads **Add a subtask** when you hover it) and type in **Add a subtask…**. To change a task's wording, the task it sits under, or its goal, click the pencil (**Edit task**) on its row. ## Choose who does it Click **Who does it** on a task. The menu has these choices: - **Me**: you do it. If a person was doing it, they come off the task. - **Someone else…**: a search of your contacts opens under the task. Pick a person and the button reads **Who does it:** followed by their name. Groupthink does not email or notify them. This only records who is doing it. - **Hand to ChatGPT**, **Hand to Claude**, **Hand to Dots**: the assistant does it and reports back. See [Hand a task to ChatGPT, Claude or Dots](#hand-a-task-to-chatgpt-claude-or-dots). - **Groupthink does it**: Groupthink works on the task for you. See [Have Groupthink do a task](#have-groupthink-do-a-task). The hand-off choices and **Groupthink does it** are greyed out on a task you have already checked off. While someone other than you works on a task, a status chip next to the button says how it stands: **Waiting for you to confirm**, **Groupthink is working on it**, **Waiting on** followed by the assistant's name, **Done**, or **Failed**. ## Hand a task to ChatGPT, Claude or Dots Pick **Hand to ChatGPT**, **Hand to Claude** or **Hand to Dots**. The chip on the task reads **Waiting on ChatGPT** (or Claude, or Dots), and a box under the task says what happens next: - **ChatGPT:** click **Open ChatGPT**. ChatGPT opens in a new tab with the task already typed in. When ChatGPT is connected to Groupthink, it reports back to the task when it is done. - **Claude:** Claude picks the task up from Groupthink. Ask Claude to do the tasks handed to it, and it reports back when it is done. - **Dots:** Dots picks the task up from Groupthink when it is connected, and reports back when it is done. The box also has **Copy the task**, which copies the task as the assistant gets it, so you can paste it anywhere. Click **Done** to close the box. The task text asks the assistant to draft anything for your review, and not to send an email or a message or put anything on your calendar without showing you first. **Getting the result back.** An assistant can report back only when it is connected to your Groupthink account. Set that up in **Settings → Connections**, under **MCP Access**, with **Connect an agent**. It has tabs for **Claude** and **ChatGPT**, among others. See [MCP Server](/docs/mcp_server) for the steps, and [Project tasks](/docs/mcp_server#project-tasks) there for how an assistant finds and reports on its tasks. When the assistant reports back, the task is checked off and the chip reads **Done**. Under the task, a box headed **From** and the assistant's name shows its summary and links to anything it made. **Where a handed-off task shows up.** A task you hand to ChatGPT, Claude or Dots stays on the project, and is also listed on that assistant's own contact page. The Claude and Dots contacts appear the first time you hand either one a task. Claude and Claude Code are two different contacts: tasks you hand to Claude go on Claude's page, and Claude Code keeps what you dictate or paste into Claude Code. A task handed to Claude before Claude had its own contact stays on the Claude Code page. Handing a task off again starts over, and clears any earlier result on it. ## Have Groupthink do a task Pick **Groupthink does it**. Nothing runs yet. A box under the task, headed **Groupthink will work on this task**, says what will happen: > It reads the task and the people on it, then writes you a plan and, if the task is to write to someone, a draft. Nothing is sent. Under that is an estimate of the run. Click **Confirm** to run it, or **Cancel** to drop it. Nothing happens until you click **Confirm**. While it runs, the chip reads **Groupthink is working on it**. When it is done, a box headed **Groupthink's plan** shows: - **A short summary** and **a list of steps**. - **A draft**, if the task is to write to someone. It is labelled **Draft email** or **Draft text**, and marked **Draft, not sent**. Click **Copy draft** to copy it. For an email, **Open in your mail app** opens it in your own mail app, addressed and ready for you to send. **Groupthink never sends anything.** It does not send email or texts. You send the draft yourself. If a run fails, the chip reads **Failed** and the task says why. Click **Try again** to start a new run. ## Confirm or undo what Groupthink suggests A finished run can also suggest a thing or two to save, under **Groupthink suggests**. Nothing is saved until you confirm it, and you can undo each one. A suggestion can be: - **An interaction logged** with a person the task names, such as a call you say you had. - **A cadence** for how often you keep in touch with a person the task names. Click **Confirm** on a card to save it. The card then says what was saved, for example **Logged with** and the person's name. Click **Undo** to take it back: the logged interaction is removed, or the cadence goes back to what it was. An undone card reads **Undone.** The plan's summary only proposes these. It never says something was logged or changed before you confirm it. A few things can stop an undo: - If the interaction was already on your record before the run, undo leaves it alone, so your own entry isn't removed. - If your cadence with that person changed after you confirmed, undo leaves it as it is. - If the task was run again, the old suggestions are gone. Reload the page to see the new ones.