Grok Bot Beginner’s Guide: Build Your AI Team

What is Grok Bot?

Grok Bot, from xAI, gives you AI teammates you can hand real work to. Each bot has its own job, works on its own cloud computer, signs in to the apps and websites you use, and can run saved routines on a schedule, even while your laptop is closed.

It runs on Mac, Windows and Linux, with mobile apps for iPhone and Android. It’s included with eligible paid plans (The X Subscription, SuperGrok and some Cursor plans), and it shares a weekly usage limit with those plans. Plans change, so check xAI’s site for current access before you pay.

Start here

Build a small AI team that helps with real work: researching ideas, drafting content, organising enquiries and keeping track of what needs you. Then bring it together in a dashboard you can make your own.

Our recommended starting point: one chief of staff, one specialist bot and one tested task. Add the rest when the first workflow is useful. You do not need to code.

This guide includes the setup interview, reusable bot instructions, 15 specialist ideas, scheduling prompts and five dashboard styles. Replace anything in [square brackets] before sending a prompt. If you do not know an answer, tell the bot to help you work it out.

• Bot: a named AI teammate responsible for a job.
• Skill: the repeatable instructions for doing that job.
• Routine: when that job runs.
• Page: a place to view results, reports or a dashboard.

The prompts below are starting templates. Review the outputs and adapt them to your own tools and work; they do not install a pre-tested integration or guarantee a particular result.

1. Get into the right app

1. Go to x.ai/bot and check your plan includes Grok Bot. It’s included with eligible paid plans, such as The X Subscription, SuperGrok and some Cursor plans, and it uses the same weekly usage as those plans. This guide is free.
2. Install the desktop app (Mac, Windows or Linux) and sign in through your browser with your account. There’s no separate Grok Bot login. Mobile apps for iPhone and Android are handy for checking in, but do your setup on desktop: some steps, like testing routines, only work there.
3. When you see Introducing your primary Bot, choose Create primary Bot, or Choose an existing Bot. Your primary Bot is your chief of staff: it checks in with you and coordinates your other bots.
4. Have three things ready: a short description of your work, the tools you use, and one repeated task you’d like off your plate.

For current installation and account requirements, use the official getting-started instructions. Button names and availability can change.

2. Set up your chief of staff

Your chief of staff is the place you go to decide what matters today, check progress and coordinate specialists. It should have a clear overview, while each specialist owns its own work. This structure is our recommendation, not a requirement to use Grok Bot.

Grok Bot already treats your primary Bot as the coordinator, so this prompt turns it into a proper chief of staff. Bots save what they learn, but say “remember” for anything that matters.

Paste this into your primary bot. Answer its questions in plain English; rough notes are fine.

You are my primary Bot and Chief of Staff. Help me set up a small, useful team of Grok Bots.

I am a [role / business owner / creator]. I help [audience or customers] with [work]. My main goal is [goal]. My tools are [tools, or ask me]. My time zone is [time zone].

Interview me one question at a time. Start by asking which repeated task takes up the most time. Then find out my priorities, recurring work, source documents, preferred outputs, deadlines and what I want to approve myself. Keep questions beginner-friendly and suggest examples when I am unsure.

After the interview, propose a team starting with you and ONE specialist. For each role, give me: the job in one line, inputs, output and where it goes, apps it needs connected, what it must ask me before doing, and a first test task. Reuse existing Bots where suitable. Do not create a whole team before I choose the roles.

Your ongoing role: organise priorities, assign work to specialists, and give me a short brief of completed work, blockers and decisions. When a handoff should be visible, suggest a group chat with the right Bots. Every Bot message counts toward my usage, so do not message other Bots when nothing has changed.

Keep a team register with each Bot's name, label, responsibility, output location, routines and approval rules.

Research and draft independently within the agreed scope. Ask before sending external messages, publishing, spending money, deleting content or changing live systems. Never ask me to paste a password: use Vault, my password manager or a Secure Form. Never say a task is complete without checking the output.

Remember these operating rules for future work. Do not create any routines until we have tested a task once.

Check the result: you should have an agreed first specialist, a useful test task and clear ownership. If the plan is too big, say: “Reduce this to the smallest setup that will save me time this week.”

3. Create your first specialist

Click + at the top of the sidebar (or press Cmd/Ctrl+N). On a team plan, choose Create a Bot. In Create New Bot, give it a name, pick a shape and colour, and click Create. Then click the name at the top of the chat to add a short label, like “Content research”, so you can tell your bots apart. Give each bot one main job: create a new one only when the work has its own goal, tools, approval limits or schedule. Official bot setup instructions.

Paste the template below into the new bot, then add one of the starter briefs in the next section. The template provides the working rules; the brief defines the actual job.

You are my [BOT NAME]. Your main job is [one clear responsibility]. Remember this role.

Context: [my business, audience and goal].
Inputs: [documents, links, apps or files].
Output: [what I want, the format, and where it should go].
Quality standard: [what a good result looks like].
Scope: [what you should and should not cover].
First task: [one small real task we can check].

Before starting, ask me one question at a time to fill important gaps. Tell me which apps you need connected and check you can open each source. If you need to sign in somewhere, use Vault, my password manager or a Secure Form, never a password pasted in chat. If a step needs me (single sign-on, a passkey or a CAPTCHA), ask me to take over the computer for just that step.

Run the first task once. Include links or evidence for factual claims, label assumptions and flag missing information. Keep original source material intact.

When finished, give me: the output, sources used, anything needing review and the next step. Send a short status note to [Chief of Staff name] only when the task is done or blocked.

Do not send messages, publish, spend, delete or change live settings without my approval. Do not create a routine yet. Remember these rules.

3b. Set your safety rules

Before your bots touch real accounts, set what they must always ask you about. Grok Bot’s Auto Review checks actions before a bot takes them, and you can add your own rules in Settings → General, under Bot: Ask first rules always get an approval card, and Ask first wins if a rule clashes. When a card appears, choose Allow once, Always allow or Deny. Emails from routines wait as drafts until you press Send. While you’re in Settings, set Execution on this computer to Never allow unless you need a bot to run things on your own machine. Official approvals guidance.

Help me write my Grok Bot approval rules. Based on my team register and the apps you can use, draft up to 10 short "Ask first" rules, for example: ask before sending any email or Slack message, posting to social media, deleting or moving files, changing a calendar, buying anything, or editing a live website.

Then draft up to 5 narrow "Allow automatically" rules for safe, read-only work, like reading my Google Drive content folder or searching the web.

Keep each rule specific and under 300 characters. Do not apply anything yourself. Give me the final list to paste into Settings, General, Bot, Custom Rules.

4. Choose from 15 bot ideas

Pick one job you already do repeatedly. Give the new bot a label that matches its job, then paste the specialist setup prompt followed by one starter brief. These are suggested roles, not prebuilt integrations. Start with an attached file or selected links when you are not ready to connect an account.

Find five recent developments about [topic] for [audience]. Include the original source, publication date, what changed, why it matters and one practical content angle. Check whether each item is genuinely new. Do not describe something as viral without evidence. Return your best three first.

Turn [transcript / article / notes] into three distinct LinkedIn drafts for [audience]. Match the supplied writing examples. Give each a clear opening, one useful lesson and a natural ending. Do not invent my experiences, results or opinions. Keep everything as drafts.

Review [selected partnership threads] and build a tracker with brand, contact, stage, agreed fee and currency, deliverables, deadline, usage rights, payment status, next action and source link. Mark unconfirmed terms as unknown. Draft the next reply for review; do not negotiate or send it.

Create five Threads drafts from [source] for [audience]. Aim for 400 to 500 characters each, with a clear takeaway and natural language. Count characters. Avoid repeating the same opening or adding unsupported claims. Return drafts only.

Research ten organisations from [list] against [ideal customer criteria]. Show the evidence for each match, a relevant business problem and a personalised draft introduction. Use public business information. Do not collect unnecessary personal details or send messages.

Turn [approved content folder] into a one-week plan for [channels]. List the asset, channel, proposed date, caption draft, approval status and missing items. Identify repurposing opportunities. Do not upload or publish anything.

Analyse [Search Console export] for [website] over [date range]. Find pages with declining clicks, high impressions but low click-through rate, and promising queries. Separate observations from possible causes. Recommend five prioritised actions with evidence. Do not edit the website or promise rankings.

Review [inbox] for [time window]. Group messages into urgent, needs reply, waiting and information only. Give the sender, subject, reason, suggested next action and original link. Draft replies only for the three highest-priority items. Do not send, archive or delete.

Prepare a one-page brief for [meeting]. Include the objective, attendees, relevant previous decisions, open questions and three talking points. Use [sources] and link them. Do not contact attendees or change the calendar.

Draft answers to [five customer questions] using only [approved knowledge base]. Cite the supporting section. Flag gaps instead of inventing policies, refunds or promises. Return drafts for a human to review.

Compare [competitor pages] with the supplied previous snapshot. Report changes to pricing, offers, positioning or products, with links and dates. If no earlier snapshot exists, create a baseline and do not claim a change. Keep the report to five useful observations.

Summarise [reporting file] for [period]. Calculate agreed KPIs, compare with the equivalent previous period and show the calculation inputs. Explain notable changes without claiming unproven causes. End with three suggested actions and data gaps.

Turn [source] into a newsletter outline, three social drafts and five short-video ideas for [audience]. Keep the central facts consistent. Give each format its own purpose and link every draft back to the original. Do not publish.

Use [signed scope and intake form] to prepare a welcome email draft, missing-information checklist and proposed milestones. Do not invent deliverables or delivery dates. Flag anything that conflicts with the signed scope.

Help me learn [topic] for [practical goal]. Ask about my starting level, then create a seven-day plan with one small exercise per day and a way to check it. Start with the first exercise and wait for my attempt before moving on.

Want bots to work together? Create a group chat with New chat and add two to six bots in the To: field. Type @ to give one bot the job. Handoffs happen through messages and shared files, and every bot reply counts toward your usage, so keep groups to the bots a task actually needs.

@[Chief of Staff name] In this group, plan [task] with @[Specialist 1] and @[Specialist 2]. Give each Bot one clear stage, in order, with one owner per stage. Each Bot posts its output or a link here when its stage is done, and the next Bot starts only after that. Reply in this group, not in direct messages. Do not run stages in parallel unless I ask. When finished, post one summary with links and anything waiting on me.

5. Connect tools and give useful context

A bot needs both instructions and access. A link to a private document won’t work if the bot can’t open it. Open Connect apps from the sidebar (or use the Connect card a bot shows in chat), add the app and finish the sign-in. Grok Bot connects to apps including Gmail, Google Calendar, Google Drive, Google Sheets, Slack and Notion. For other websites, the bot uses its own computer and signs in with your Vault, a connected password manager like 1Password, or a Secure Form, which fills the details without the bot seeing them. If a step needs you, such as single sign-on, a passkey or a CAPTCHA, choose Take over, do just that step, then choose I’m done. Never paste passwords or codes into chat. Official app connection guidance.

• Start with the account and folder needed for one task, such as the content folder for your writer.
• Provide a short business brief, audience description, two or three examples of good work and any rules that matter.
• Ask the bot to prove it can read the intended source before assigning a large job.
• Use a permitted export or upload if a connection is unavailable. An inaccessible source should appear as missing, not as invented data.

For this task, list the apps and sources you need. Check the exact account, inbox, folder or document I specified, rather than assuming a default.

For each one, report: connected and readable, needs connecting in Connect apps, needs a sign-in (Vault, password manager or Secure Form), needs me to take over, or not needed. Prove access with one non-sensitive item, such as a document title. Do not print private message contents to prove access.

If something isn't connected, show me the Connect card or tell me exactly what to connect. If access isn't possible, tell me which file or export I can give you instead. Do not change anything during this check.

For client work: separate bots are not a security boundary. All your bots share one cloud computer, including its files, browser sessions and logins, so they can hand work to each other. Keep each client’s material clearly separated in your instructions and folders, connect only what each job needs, and remember that deleting a bot doesn’t delete files on the shared computer. Read the security guidance.

6. Test the job before scheduling it

Give the specialist a small real task. Check the facts, format and usefulness. Correct it before multiplying the same mistake across a week of work.

Review the task you just completed against my original request. Check the facts against the sources, identify anything missing and distinguish completed actions from recommendations.

The parts I like are [details]. Change [details]. Use this as our preferred output format: [format]. Produce the corrected version and summarise the lasting preferences you will remember.

Once the result works, save the method as a skill. Then ask the same specialist to create its routine. A routine belongs to a bot; a dashboard does not itself schedule the underlying jobs.

Save the successful process we just used as a skill named [SKILL NAME]. Include the purpose, required inputs, source locations, steps, quality checks, output format, approval rules and what to do if a source is missing. Keep task-specific dates and examples separate from reusable instructions.

Before scheduling, check your time zone in Settings, under Bot, Timezone. Routines can run on a schedule or when something happens, like a new email or Slack message, and they keep running in the cloud while your laptop is closed. A new routine waits for its next scheduled time, so open it and choose Test on desktop to run it now (a test does real work, so use safe inputs). To stop one, use the switch on the routine or Pause. Official skills and routines guidance.

Create a routine for this Bot to run the [SKILL NAME] skill [every weekday at 8:00 AM / when a new email arrives from (sender) / other trigger]. Use my time zone, [named time zone, for example Australia/Sydney].

Use [exact source] and put the result in [destination]. Keep the task limited to [scope / maximum items]. Do not repeat items already processed; use [source IDs / dates / output register] to track them.

If an input fails, report the blocker instead of inventing an output. Leave any emails as drafts and keep publishing for my approval. Send one short completion summary, and let [Chief of Staff name] know only if something needs my decision.

Show me the routine's name, owning Bot, trigger and next run time. Do not create a duplicate. Remind me to open it and use Test on desktop.

7. Build your Bot HQ dashboard

Paste this into your chief of staff after at least one specialist has produced a useful result. The useful part is the overview: decisions, completed work, upcoming routines and a few meaningful numbers. The animated scene is your choice of presentation. Pages are saved in your bot’s Library, so you can come back to Bot HQ any time.

Grok Bot can create Pages from a chat request. Pages availability can vary by rollout and account; update the app and check the current Pages instructions if you do not see them. If unavailable, start with a simple report and create the Page later.

Build a Page called Bot HQ that brings my existing bots and work together. First inspect the bots, outputs, routines and sources you can access. Ask only for essential missing information. Do not create new specialist bots for the dashboard.

Include:
1. A visual team overview showing each bot's name, role, last known status and current or last task. Use working, waiting for me, idle, failed, not activated and unknown where appropriate.
2. Up to five useful metrics for my work. Ask me which decisions these metrics should help with. Use only accessible sources and show the reporting period, source and last refreshed time.
3. Waiting on me: the decision, bot, reason and a real link to the relevant conversation or output where available.
4. Recent work: completed tasks, completion time and links to results.
5. Upcoming routines: owner, purpose, schedule, time zone and next run when available.
6. Optional activity history if trustworthy data exists. Explain what is counted; do not call messages or transcript events completed tasks.

Use real data only. Display Not connected, No data yet or Unknown where needed. Zero should mean a measured zero. Do not invent links, completion events, stats or bot activity.

Explain whether each section is a snapshot, refreshed by a routine, or genuinely live. A moving character or clock does not mean task data is live. Put a visible last-updated timestamp on snapshots. Do not infer that a bot is working merely because it has a schedule.

Use a responsive layout, readable labels and text status indicators as well as colour. Provide a plain list alongside the visual scene and respect reduced-motion preferences. Keep private email bodies and sensitive commercial details out of overview cards.

Use only controls that actually work. If a button cannot approve or cancel a task, link to the relevant bot or routine instead; never imply that a visual change has changed the underlying job.

Before building, ask me to choose: office, mission control, airport departures board, pixel village or aquarium. Keep the operational sections the same whichever style I choose. Build the Page and explain how I can request an update.

Important: animation and data refresh are different. Pages can be updated by asking the bot, including through a routine. Ordinary Page filters and switches change the view, not the stored data. Keep task approval in the actual bot workflow. How Pages work.

8. Pick one of five dashboard styles

After the master prompt has created Bot HQ, paste one style prompt below into the same conversation. You can swap the visual later without rebuilding your team. These are design requests; the exact result depends on what your account can render.

Apply an office theme to the existing Bot HQ Page. Keep all operational sections, source links, timestamps and data rules.

Create a top-down 2D office. Give each bot a named desk and a distinct character. For a verified working status, show the character typing with a short task bubble. Show idle bots strolling or taking a break. Show bots waiting for me with a visible question marker. Keep failed, unknown and not-activated states distinct.

Place the chief of staff near a central team board. Make names readable and include a compact status list below. Use restrained animation and a reduced-motion mode. Movement reflects the last verified status; it must not pretend to be a live feed.

Restyle the existing Bot HQ visual as a space-station mission-control deck. Preserve the stats, decisions, recent work and routines below.

Each bot has a labelled console. A working console glows and shows a short verified task description. Idle consoles dim to standby. Waiting-for-me consoles show an amber decision indicator. Failed consoles show a clear error label; unknown status says Unknown.

Use dark navy, crisp text and restrained accent colours. Keep decorative animation separate from actual activity. Do not generate fake streaming logs. Show the source timestamp for the displayed statuses and retain an accessible list view.

Restyle the existing Bot HQ visual as an airport departures board with a gentle split-flap effect when displayed values change. Preserve the operational sections below.

Each bot is a flight. Columns: bot, task, status, next scheduled run and output link. Map verified states clearly: Boarding = queued, In flight = working, Landed = completed, Awaiting clearance = needs my decision. Also support Idle, Failed, Not activated and Unknown. Include a legend.

Only show a next-run time when known. Use task or routine links that actually exist. Keep letter animation brief and disable it for reduced-motion users. Do not invent gates, timers or progress to imply real operational data.

Restyle the existing Bot HQ visual as a small top-down pixel-art village. Give each bot a labelled house and its own character, with a central town hall for the chief of staff. Preserve the useful dashboard sections below.

Working bots carry crates or work outside their homes, with the last verified task in a short speech bubble. Idle bots rest on benches. Bots waiting for me display a question marker. Not-activated houses have a clear label; failed and unknown states remain distinct.

Keep houses and labels readable on small screens with a wrapped layout or scrollable scene. Include a text roster, last-updated timestamp and reduced-motion support. The scene represents recorded bot states, not a claim of live telemetry.

Restyle the existing Bot HQ visual as a calm aquarium. Give every bot a distinct fish with a persistent name label. Preserve the metrics, decisions, completed work and routines below.

Verified working fish swim a little faster and show bubbles with a short task description. Idle fish drift gently near the bottom. Fish needing my decision display a question bubble. Unknown, failed and not-activated bots have explicit text labels in the roster.

Avoid overlapping labels, frantic movement and relying only on colour. Include a plain status list and reduced-motion support. Do not show percentages or countdowns unless grounded in real data. Retain the last-refreshed time.

9. Keep the dashboard useful

Start with an on-demand refresh. Add one daily update only if it helps you act on the information. Refresh frequency and task frequency do not need to match.

Update the existing Bot HQ Page using the latest accessible bot outputs and source data. Preserve its style and structure. Update the timestamp and clearly flag missing or stale sources. Reuse existing completed outputs rather than rerunning every task. Do not create a second dashboard or claim a successful refresh for sources you could not read.

Set up one routine owned by my Chief of Staff to update the existing Bot HQ Page every weekday at [time] in [time zone], after my specialist routines normally finish.

Use their latest completed outputs and agreed metric sources. Do not rerun their jobs or start a repeated bot-to-bot polling loop. Show pending tasks as pending and unavailable data as unavailable. Keep the last successful refresh time for failed sources.

Summarise only new decisions, significant failures and useful completed work. Confirm the saved schedule and explain how to pause it. If updating Pages through a routine is not supported in my account, tell me and use an on-demand update instead.

Review my current goals and Bot HQ. Recommend three dashboard improvements that would help me make decisions, rather than just add decoration. For each, state the question it answers, source needed, refresh frequency and likely extra work. Recommend one to add first and wait for my choice.

10. Best practices for a team you will actually use

• Give each job one owner. The chief coordinates; the specialist produces the work. Avoid three bots researching the same thing.
• Define finished. “Find five sources and produce three ranked ideas” is easier to check than “help with marketing.”
• Provide examples. A strong previous post or report teaches more than “make it professional.”
• Keep facts traceable. Ask for source links, reporting periods and unknowns. Separate observations from guesses.
• Start with drafts. Review outputs before enabling any workflow that changes live work or contacts others.
• Choose a sensible frequency. Weekly reports rarely need hourly routines. Watch account usage as you expand.
• Improve the instructions when correcting work. Tell the bot which preference should apply next time.
• Review the team weekly. Keep useful roles, simplify overlap and pause routines you do not use.
• Watch the chatter. Every bot reply counts toward your weekly usage, including messages between bots. There’s no switch to stop bots messaging each other, so tell them when not to, and ask them to remember it.
• Keep secrets in Vault. Logins, cards and contact details belong in Vault, not in a bot’s memory or chat.

Review my team register, latest outputs and accessible routines. List which jobs produced useful work, which were blocked, which overlap and which I have not used. Identify stale sources and repeated failures.

Recommend what to keep, simplify or pause. Use actual usage data only if accessible; otherwise say usage is unavailable. Do not cancel routines or delete bots automatically. Give me the three changes most likely to improve usefulness or reduce unnecessary runs.

11. Fix common setup problems

Your first setup checklist

Your next step: pick the one repeated task you would most like off your plate, copy the chief-of-staff prompt and start there.

Want more practical AI workflows? Explore The Ai Impact guides or join the newsletter.

FAQs

Sources and update notes

Setup references checked 30 October 2026. This is an independent guide from The Ai Impact. Plan access, controls and feature availability can change; use the official links for the latest details. Prompts and dashboard concepts are our suggested workflow templates.

The Ai Impact is not affiliated with xAI.

• Grok Bot product overview and access
• Installation and first task
• Creating bots and choosing a primary bot
• Computer access and connected apps
• Skills, schedules and routine testing
• Pages and updating results