Message a Grok Bot

Working with a Bot feels like messaging a teammate. Keep the request natural, and make the outcome and the decision boundary explicit: what the Bot may change, and where it must stop for you. Official guide: Work with Grok Bot → Message a Bot.

In a conversation you can paste text, links, and images, attach files, reference a saved skill with /, mention a Bot, group, routine, or plugin with @, reply to or react to a message, and send another instruction while work is already in progress. The transcript shows tool activity, computer use, created files, questions, and approval requests alongside normal messages.

Send a clear first ask

State the finish line and the stop points in the same message. Name the systems to open, the artifact to return, and anything the Bot must not send, purchase, delete, or publish without you. A short example:

The PDF is the signed policy. The spreadsheet is this month's transactions. Reconcile the spreadsheet against the policy, cite the relevant policy section for every exception, and return a new spreadsheet plus a short summary. Don't modify the originals.

Paste a link when the Bot can reach the page from its computer or a plugin. If the page is private, sign in through the computer or connect the plugin first. Always check a link's destination before entering credentials or approving an external action.

Attach files and media

Supported inputs include images, audio, video, PDFs, Office documents, CSV, JSON, YAML, source code, HTML, email files, and notebooks. Large, encrypted, damaged, or unusual files may not be readable; export an unusual format as PDF, CSV, plain text, or an image before you attach it.

On desktop you can send up to six attachments per message. Documents, images, and audio are capped at 25 MB each; video at 200 MB. Tell the Bot what each attachment is and how to use it so it does not guess the wrong sheet or ignore a policy page.

Redirect or stop while work runs

A direct message from you takes priority over background work and can redirect the current turn. Send a new instruction mid-run when the goal changed, a source is wrong, or you want a different artifact. Send Stop now when the turn should end immediately. Stopping does not undo actions the Bot already completed, so keep destructive steps behind approval before you rely on Stop.

Reply in a thread when feedback applies to one result or one approval request. Use reactions for lightweight acknowledgement; a reaction alone should not carry a safety-critical decision.

Skills, mentions, and groups

Type / to pull in a saved skill for the current ask. Type @ to hand work to another Bot, a group, a routine, or a plugin by name. In a group of two to six Bots, write normally when they should decide who responds, and use @ when one owner should take the next step. Save @everyone for group-wide updates. Your group messages can include attachments; Bot-to-group handoff messages are text-only, so a Bot should send an image directly to another Bot when that teammate must inspect it.

Pitfalls

• Vague asks with no finish line or approval boundary — the Bot will keep going until you clarify stop points.
• Pasting passwords or one-time codes into ordinary chat — use Take over or the secure secret request instead.
• Expecting Stop now to reverse completed sends, deletes, or purchases — it ends the turn; it does not roll back finished actions.
• Relying on a reaction alone for a consequential approval — reply in the thread with an explicit decision.