This Playbook walks you through getting started with Grok Bot. How to set it up, what to hand it first, and how to know it's working. You get the eight most useful ways entrepreneurs are using it right now, with the exact prompts to copy, the mistakes to skip, and the one-paragraph brief I use with clients. Start with one.
This playbook is a synthesis of the best Grok Bot use cases from Riley Brown, Alex Finn, Claire Vo, and Tenex, translated for entrepreneurs and the teams they lead. Use it as a menu. Pick the outcome you want, borrow the play, and adapt it to your world. For the foundation underneath every play, start with how to build AI agents.
Give an agent a clear responsibility, relevant context, and a recognizable finished result. Teach it your standards through examples and corrections. When the work deserves to repeat, give it a trigger and somewhere to keep track of what happened.
Five words that make the product easier to understand. Everything in Grok Bot is built from these. Once you can tell them apart, the marketplace stops feeling like noise.
A conversation with a defined role.
A reusable method for doing something well.
A package of tools, connections, or skills.
Instructions that run on a schedule or supported event.
Saved context the next run can use.
Every play stands on its own. Find the line that sounds like you and jump straight to it.
Each play has the idea, where it fits for you and for a client, a prompt you can copy, and what a good result looks like. The prompts are adaptations. Check the product features in your own account when you use them.
Alex's strongest setup idea: explain your world, then ask the bot how it could help. Riley adds a useful simplification. Keep one general conversation you can always return to. A focused project can become its own bot when that separation helps.
Provide your current priorities, a few key reference documents, and what you want more space for. Your existing notes and past AI conversations can supply context without becoming a migration project.
Ask what they repeatedly chase, avoid, reconstruct, or forget. Those answers reveal useful responsibilities.
Here is what I do, what matters to me, and what keeps taking my attention: [context]. Suggest three specific ways you could help. For each, show the finished result I would receive and what information you need. Recommend one. Ask me only the questions needed to make that recommendation useful.
A useful output becomes more valuable when the method can be repeated. Alex teaches voice from sent emails and newsletters. Riley saves his preferred Notion enrichment format as a skill. Tenex demonstrates teaching a task by showing the steps.
Start with a deliverable you already like. I use my Gold Mine extraction method and my voice standards. Give the bot examples, correct a real attempt, then preserve the improved method.
Capture their best proposal, onboarding process, or customer response as a repeatable standard.
Study these examples and my corrections. Describe what makes the result good in concrete terms. Turn the method into a reusable skill with the required inputs, steps, output format, and quality checks. Test it on this new example: [example].
The next result needs less explanation and less editing.
Riley's most transferable idea: name something you care about, define a meaningful change, and have the bot notify you when it happens. His temporary email watcher stops after the reply arrives. Claire emphasizes that you must verify an actual routine exists.
A speaking engagement awaiting logistics, a proposal awaiting a response, or a client project approaching a handoff.
A supplier confirmation, a customer escalation, or an important delivery.
Watch [specific situation] using [sources]. Notify me when [meaningful change] happens. Include what changed, the source, and the next action. Stay quiet when nothing relevant changes. Stop when [completion condition]. Show me the proposed trigger and instructions before activating the routine.
A useful alert that saves you from checking. Choose frequency around the situation's urgency. The presenters' frequent checks are examples, not defaults.
Alex's email workflow combines three jobs: identify worthwhile opportunities, research the sender, and prepare a response in his voice. The practical value is having the reply ready where he already sends email.
Speaking, coaching, partnerships, and referrals.
Quote requests, sales inquiries, customer questions, or follow-ups.
Review [business inbox and date range]. Find up to three conversations where a response could advance a real opportunity. Read the full thread, explain why each matters, and flag unanswered questions. Use these examples of my writing to prepare replies in Drafts. Check for existing drafts first. Leave sending to me.
A relevant conversation moves forward. Research supports your judgment. It cannot guarantee that a sender is legitimate.
Combine Alex's repurposing with Riley's Notion technique. Use your original material, learn your voice, and add research in a clearly labeled expandable section beside your outline.
Turn a coaching insight into a newsletter, strengthen a keynote outline, or enrich a Notion draft with supporting examples.
Turn expertise from a sales call or expert interview into a useful article or customer resource.
Read [source] and these examples of my voice. Find the strongest idea for [audience]. Draft one [format] around it. Show the source behind the main point and flag missing evidence. Keep the draft for review. For this Notion outline, propose a separate "Supporting context" section with relevant research and links while preserving my original writing.
One piece you would actually use. Choose additional formats when they have a real destination.
Riley mines saved bookmarks. Tenex curates bite-size learning. Alex investigates recurring customer problems. Their common value is turning information into something you can use.
Connect an AI development to a client lesson, an existing offer, or a worthwhile experiment. This is the same move as my Gold Mine method.
Identify recurring customer questions and propose an improvement to their service.
Research [question] using [sources]. Bring me the three findings most relevant to [goal]. For each: what you found, the evidence, why it matters to us, and one possible application. Separate recurring evidence from isolated anecdotes. Recommend what deserves attention and what we can ignore.
The research changes a decision or produces a usable lesson. An interesting finding alone does not require a new build.
Riley asks his main bot to gather updates from other bots. Tenex adds a useful refinement. Have specialists produce short status summaries so the coordinator reads a summary in place of every full conversation.
Connect research, content, and project work around one outcome.
Coordinate preparation for a launch or a customer delivery.
Ask the bots working on [project] for what is complete, what is blocked, and what needs my decision. Include links to their work. Give me one combined update with the next action and its owner. Identify any missing responses rather than treating them as completed work.
Less work relaying context between conversations. Define who owns the finished result. Connecting multiple tools to Notion alone does not establish an automatic handoff.
Claire's strongest personal lesson is to judge the agent by the attention it returns to your life. Her examples include a printed family briefing, subscription reviews, and shopping to exact requirements. Tenex applies the same idea to household requests.
Pick something you repeatedly look up, forget, or spend too long comparing.
Fewer forgotten details and less time spent managing the tool.
Create a one-page family brief from [calendar and school messages]: commitments, things to bring, conflicts, and one fun conversation starter. Flag anything unconfirmed. Review subscription receipts in [accounts]. List recurring charges and renewal dates with source links. Ask me about actual use before recommending cancellations. Make no account changes. Find three [products] matching [dimensions, preferences, and budget]. Compare fit, full price, and compromises. Give me purchase links for review.
Three plays that did not need a full prompt to be useful. Borrow the shape and adapt it.
Talk through an idea by voice. Ask one challenging question at a time. Finish with the decision, the uncertainty, and the next move. Useful for coaching preparation and entrepreneur clarity.
Answer from approved material, flag exceptions, and turn recurring questions into proposed documentation improvements. The help desk gets smarter every week it runs.
Delegate a concrete website change and request a preview, screenshots, and checks. Review the result before publication.
A featured bot gives you a starting role. A skill gives you a method. A plugin gives you capabilities. Choose the missing piece for the job you want done.
Before installing anything, answer four questions.
What result will I use this for?
What does it connect to or change, and does it add routines?
Does its method fit my business, my examples, and my standards?
Can I test it on one real case and recognize a good result?
For Notion work, begin with the relevant Notion connection. For email work, add the intended inbox. For research, choose the sources the question needs. A creator's entire tool stack is not a setup requirement.
Build your own skill when you have a method worth repeating. Adapt a template when its workflow already fits. Give work a separate bot when a continuing role or project benefits from its own context.
Three of these come straight from the setup guides. The last two are the checks I add for client work.
Vague authority. "Handle my email."
Name what the bot may do and what needs a decision. Research and draft freely. Ask before sending. Pair the instructions with the app's approval controls. Approval guidance
Scheduling a workflow after one good result.
Complete a task → save the method → test on fresh input → schedule. A demonstration produces a draft skill. Add the decisions and exceptions the recording could not show. Test runs perform real work. Skills and routines
Adding bots because the example team looks impressive.
Add a bot when a distinct responsibility benefits from its own context and produces something you use. There is no required team size. One useful bot is a valid starting point.
Letting missing information look like "nothing to report."
Specify what happens when a source fails or is stale. Example: "If you cannot read my calendar, say the brief is incomplete. Do not reuse yesterday's schedule."
Assuming different bots keep client access separate.
Current docs describe separate bot screens on one shared computer. Files and browser logins are shared. For client work, use each client's own account and approved connections. Bot names do not isolate access. Computer and apps
Owner. Trigger and time zone. Current source. Finished result. Approval point. Missing-data response. Add a stop condition for temporary work. This makes the assignment clear without requiring a timed setup plan.
You own [responsibility]. Start on [event or schedule with time zone]. Read [current sources] and produce [result] in [destination]. You may [allowed actions]. Ask before [actions needing my decision]. If a source is missing or stale, explain what is incomplete. On a retry, check what already exists so you do not duplicate it. Stop when [completion condition]. Show the configured routine and next run, then test it on a safe example.
Judge the result by whether it advances work or returns attention. A successful schedule by itself does not establish usefulness.
This is the one-paragraph brief I hand a client to turn a vague wish into a real assignment. Fill in every bracket, run it on a real example, then improve the instructions from the feedback.
I want help with [responsibility] so I can [desired benefit]. Use [sources]. A good finished result looks like [example] and belongs in [destination]. You may [allowed actions]. Bring [specific decisions] to me. Start when [request, schedule, or event] and stop when [completion condition]. We will judge usefulness by [observable result].
Run it on one real example.
Improve the instructions from their feedback.
Preserve the method when the result is worth repeating.
Am I reading this, acting on it, or getting meaningful work off my plate?
Every agent, routine, and skill gets measured against that line. Reading it is the lowest bar. Getting meaningful work off your plate is the whole point.
Want to run the same kind of process on your own recording? See how I edit videos with ChatGPT and Astra.
Undeniable Studio is a live building room where you co-create with AI every Tuesday and leave with something real. Playbooks like this one are what we build together.
Join The Studio →