Key takeaways
- Treat Grokbot as teammates you staff by job, not as one multi-threaded chatbot you interrogate.
- Scope each bot like a job description: tight swim lanes, related tasks only, one owner for research, positioning, web, ads, analysis, or project management.
- Give bots tool access (browser, computer, MCPs/APIs, Google Docs, Google Ads, repo) so they can work in the same systems you use.
- Kick jobs off in parallel. Bots can message each other for handoffs; you should not be the only router of context.
- Leave feedback in the artifact (comments in a Google Doc, not only in chat), then send the bot back to revise.
- Do not push live spend, credit cards, or “keys to the kingdom” on day one. Start simple and low-risk, then widen autonomy.
- When the campaign flow works, put a project-manager bot in front as your only point of contact so you stop context-switching across five chats.
- Invest in memory, feedback, and templates. Copy strong bots from the marketplace, or scrape templates into a bot you already prompted.
Who this is for (and who should skip)
This is for a founder, marketer, or ops lead who already runs campaigns and is tired of AI that only talks. You need a Grokbot workspace, willingness to connect real tools (at minimum Docs and an ad account; repo access if you want page shipping), and enough comfort to treat a bot like a junior teammate: assign work, review output, give notes, grant more access over time.
Skip this if you want another Q&A chatbot. Josh Kim’s point is blunt: Grokbot is weak if you use it like search or a single LLM thread. It is also a poor fit if you cannot connect tools, or if you need a signed-off European data-processing setup before any experiment — he said privacy modes exist, but security is handled case by case with their team.
What Grokbot changes for marketing teams
Most people still use AI the way they started: one or two chatbots as a thought partner. You ask questions, spar on strategy, edit copy. Copilots can take tasks, but you babysit them, especially on anything complex. They are doing partners only if you keep pushing.
You can think of Grokbot as always-on, asynchronous AI teammates that actually do and finish the work on your behalf.
The product difference is structural. You do not run one agent with a tangle of threads. You create an individual bot for each job — a singular teammate you delegate a task or a set of related tasks. You message bots the way you message Slack or Teams. Bots can talk to each other in group chats. As you work with them, they pick up how you work, your preferences, and memory. Josh’s rule is the same as managing people: more feedback and more investment, more they can take off your plate.
Grokbot also has its own computer and browser. Hook it to the MCPs, APIs, and tools you already use, and it can operate those tools the way you would — bidding habits in an ad platform, the way you write a campaign brief, even how you like to manage a machine. Because it is hosted in the cloud, it runs 24/7. You do not leave a laptop open or stand up a Mac mini for reporting pulls. The UI is meant to feel as familiar as iMessage, which is the point: any marketer, not only the most technical one, can put agents into daily work.
What you are trying to remove is the real tax of a campaign: context management, coordination, and orchestration across research, product marketing, web, performance, and analysis. The deliverables are the easy part. The handoffs are the job.
Scope Grokbot roles like job descriptions
Before you paste a giant prompt into one mega-bot, decide who is on the team. Josh’s first rule for starting with Grokbot: scope bots properly. Identify a job that needs doing, write it like a job description, and keep the bot specialized. Tight swim lanes are how you get efficiency out of a bot team.
His campaign roster — crowdsourced across his team, not all built by him — was:
- Market researcher — product and market understanding, competitor sites, positioning gaps.
- Product marketer — positioning brief, one-liners, packaging, value statements, surface examples, landing outline, search-ad variants.
- Website ops — take the brief, open a PR, ship a landing page, send progress screenshots.
- Performance marketer — shell campaign in Google Ads (in the demo, optimize for clicks for a copy test).
- Marketing analyst — pull experiment data, TL;DR, recommendations back into strategy and assets.
- Project manager — study the specialists, absorb where you had to interject, run the flow, become the only inbox you watch.
Keep roles always-on. Initiatives have a start and stop; these bots should compound. If you need a time-boxed push, do not replace the roles — group the specialists in a chat and give that group an initiative (Josh’s example: a tiger team on website conversion). You can also customize avatars and instruct a bot’s voice in its own memory. Personality is optional; specialization is not.
How to use Grokbot for a full marketing campaign
The working example was a net-new product, XAero, an airline site with route search, fares, and existing value props. The assignment: launch a campaign from scratch — research through positioning, landing page, ads, analysis — then automate the orchestration. Pin the bots in the sidebar the way you pin people in iMessage, then flip between them. You do not wait for one job to finish before starting the next.
| Use case | Bot / tool | Input needed | Output | Owner |
|---|---|---|---|---|
| Market and competitive research | Market researcher (own browser) | Product URL / site | Product read, competitor set, gaps, screenshots | You review; then hand off |
| Positioning and messaging | Product marketer + Google Docs (MCP) | Researcher analysis; your Doc comments | Brief, one-liners, landing outline, ad variants in a sheet | You comment in the Doc |
| Paid landing page | Website ops + repo / Cursor under the hood | Latest brief, write access to the marketing site | PR, preview, production URL | You watch screenshots; approve ship |
| Search campaign shell | Performance marketer + Google Ads | Ads access; copy from product marketer | Campaign built in-platform, click-optimized for a copy test | You own budget and go-live |
| Experiment readout | Marketing analyst + Google Ads API | Live or prior test data | TL;DR, winner, spend/CTR/CVR, recommendations | You decide what to accept |
| End-to-end orchestration | Project manager talking to all bots | Roles, past chats, where you interjected | Queued campaigns, screenshots, single point of contact | You direct; you do not traffic every handoff |
1. Market research while other work is already running
What you are trying to accomplish: a researcher who actually visits the product and competitor sites, not a generic “who are our competitors” essay.
Prepare the product URL and a market-researcher bot whose job is only this class of work. Dictate the assignment: study the site, understand the product and market, then competitive analysis — competitors’ marketing sites, positioning, and the gaps you can lean into.
In the demo, the bot scraped XAero, identified a familiar airline competitive set, and returned gaps the speaker wanted to use in messaging: useful time, day design, long hauls, and leading every message with useful time. It showed its own browser view and progress screenshots, including a pull of the website. That is what “good” looks like: source-backed analysis plus visible work in the browser, not a vibes summary.
You can stack another research job on the same bot. Josh instead moved to the next specialist so work could run in parallel. Cloud hosting is what makes that orchestration possible without you sitting in one thread.
2. Product-marketer handoff, brief, and your comments in the Doc
Tell the product marketer to go to the market researcher, take the analysis, and draft a positioning brief: go-to-market, one-liners, positioning and packaging, value statements, and two or three examples of how it shows up on marketing surfaces. The bot should show that it exchanged messages with the researcher. Over time Josh says that collaboration gets more organic and you do less prompting of the handoff.
In the demo the brief landed in a Google Doc connected through MCP: research handoff, target audience (long-haul travelers who want useful time back), positioning, go-to-market notes, one-liners by angle, value statements, and a drafted paid landing page. Josh left a comment in the Doc — lean into that audience in messaging — then sent the bot back to the file.
Second pass on the same bot, still in parallel with other work: read the comments, update the draft, build a full landing-page outline, and draft Google search campaigns for variant testing of the copy angles. Output to look for: comments resolved, a fuller outline you can implement, and a sheet with variant name, hypothesis, ad group name, URLs, and copy variants. That sheet is the handoff performance marketers actually use.
3. Website ops and performance ads at the same time
Website ops: take the landing page from the latest product-marketer brief, open a PR that pushes it as a new page, and send screenshots while working. That bot needs repo access to the marketing site. Under the hood in the demo it used Cursor and cloud agents to write the page from the outline. Nudge it to push to production when the PR is ready, then ask for the URL. Good looks like a preview, a production push, and a URL you can open from the same Grokbot workspace.
Performance marketer: create a shell campaign in Google Ads, optimizing for clicks, because the test is copy and messaging still being drafted. Connect the Google Ads account before you ask. The bot should find the drafted copy, show screenshots of building inside the platform, and structure the campaign so the messaging test can be trafficked in. Josh did not take the demo campaign live — that needs budget guidance and a credit card. He pointed the analyst at a campaign he had already launched that week instead.
Failure mode here is obvious: if Ads or the repo is not connected, you get a plan, not a campaign or a PR. If you skip screenshots, you lose the only monitoring loop he used while bots worked in the background.
4. Analyst readout, then a project manager as your only inbox
Analyst prompt in plain language: go into Google Ads, pull the last messaging experiment, give the TL;DR, then recommend how to fold it into strategy and into the other assets already in play. Good output in the demo: a clear winner across variants (brand, customer promise, hours in between, cheap that costs a day, and similar angles), the metrics you actually care about (spend, CTR, CVR), and recommendations you can debate with humans or, if you choose, let the bot act on.
Josh flagged the live debate: how much liberty you give an agent to make those decisions. In the demo the analyst recommended; he would still intake and discuss. The win he claimed was process: analysis without the usual dependencies and waiting.
Until this point, you are still the orchestrator — the expensive part. The last move is a project-manager bot. Tell it to study and talk to each specialist, learn their role in bringing a campaign to life, read the conversations for where you had to interject, weave that into how it works with them, then kick off and automate the flow (in the demo: three new campaigns). Require screenshots and progress updates. Most important instruction: it is now your only point of contact. You do not talk to the other bots.
Good looks like a queue (research, positioning, page build, and so on), specialists doing the work, and one chat keeping you in the loop. You zoom out and spend time where you still have leverage.
Share templates and steal good bots
A bot you have trained — context, memory, routines — can be turned into a template. Prompt it to create a template of itself you can share. The receiver pastes it into their Grokbot instance and gets a snapshot of how you work. Josh used that internally: he asked who had the best positioning bot, the best ads bot, and so on, then assembled the demo team.
There is a bot marketplace for the strongest templates. You can build from scratch, or copy. The specific marketing bots from this campaign were still being refined and were not listed yet; he said they would be pushed. Building a strong specialist, in his “lazy” version, was a stack: dictate how you think for a few minutes, pull tone and voice from the live website, point the bot at two or three marketplace template URLs and tell it to scrape that context into itself, then connect Superme (an experts marketplace, via a plugin) so the bot vets positioning work against experts instead of you running that research by hand.
Prompts, bot instructions, and workflows
These are the assignments Josh dictated, cleaned of filler, not rewritten into a different method. Swap XAero and tool names for your product.
Market researcher
Hey market researcher, study the XAero website. Get a deeper understanding of what the product is and what market we are operating in. Then do a competitive analysis: identify and deeply understand our competitors, look at their marketing websites, understand their positioning, and identify the gaps and opportunities we have to strategically position against them in our marketing strategy.Product marketer — first brief
Product marketer, go to the market researcher bot, do a handoff of the analysis it just performed, and draft a positioning brief. Identify how we should go to market, our one-liners, positioning and packaging, and value statements. Show examples of how this should come to life across two or three different marketing surfaces.Product marketer — after you comment in the Google Doc
Product marketer, I left comments inside the Google Doc. Go through the Doc, take the comments and feedback, and incorporate them into an updated draft. While you are at it, build a full outline of the landing page, and ideate and draft Google search campaigns to do variant testing between the copy angles you drafted.In-Doc comment he actually left: “This is great. Lean into this in our messaging.”
Performance marketer
Performance marketer, start creating a shell campaign inside Google Ads. Build it so it is optimizing for clicks. We are going to do copy and messaging testing that is being drafted by the product marketer right now.Website ops
Website ops, the landing page outline is drafted. Take the landing page from the latest brief the product marketer just drafted, spin up a PR to push that as a new landing page on the website, and send me screenshots as you work so I can monitor progress.Marketing analyst
Marketing analyst, go into Google Ads. Pull down the data from our last experiment on messaging, analyze it, tell me the TL;DR of the insights, and give me recommendations on how we should incorporate it into our marketing strategy and update the other assets you have seen.Project manager — take over orchestration
Project manager, study and talk to each of the bots on my team and understand each of their roles in bringing a campaign to life. Look at our conversations and where I had to interject, give guidance, or give feedback, and weave that into the way you work with them. Kick this off and automate it completely with three new campaigns. Send me screenshots and progress updates so I can stay in the loop. Be my point of contact: I do not want to talk to any other bots. I only want to talk to you.Template snapshot
Create a template of yourself that I can share with someone else.Reconstructed from the speaker’s description — first access pass
You are hooked up to my Slack and email. Study the context, the messages, and my organizational history. Tell me what you can do for me and take a job off my plate.On permissions, he also described a more liberal setting: if the bot keeps asking, you can tell it to always allow those actions. Use that after you have seen it deliver, not as the default on a fresh bot with Ads or production access.
Measurement and business impact
Josh did not give time-saved hours, cost, or lift numbers. The checkpoints he actually used:
- Research that cites real competitor sites and names gaps you would brief a human researcher to find.
- A positioning Doc you can comment in, plus a sheet of testable ad variants.
- A landing page that reaches production and returns a URL from the workspace.
- An Ads shell you can see in screenshots, structured for a messaging test.
- An analyst TL;DR with a winner and the metrics the team already cares about (spend, CTR, CVR), plus recommendations.
- A project-manager bot that absorbs the handoff tax so you are not the router across five chats.
The business claim is qualitative: always-on pulls for recurring reporting, no babysitting to keep work moving, and attention moved from trafficking to direction. Live budget and go-live stayed human in the demo.
Pitfalls and guardrails
It won’t be as valuable if you come into it and just treat it as another LLM.
That is the main failure mode. If you only ask questions, you get another chatbot. The shift he wants: trust it to do things. A pattern on his team is Slack as first-line triage — inbound questions and DMs get handled, and you get pinged when something is important, while the bot can message the other person back.
You’re not going to give them the keys to the kingdom immediately.
Treat a new bot like a new teammate. Start with a simple task and one or two tools. Confirm it delivers and improves as you prompt, give feedback, and curate it toward how you work. Then give it more. He said engineers on his team eventually run armies of bots that push code to production — that is an end state after rich prompting, memory, and context, not a day-one setting.
Keep a human in the loop where the demo kept one: comments on positioning, production nudges, ad spend and credit cards, and whether analyst recommendations become decisions. Customer-facing and paid surfaces are where over-automation hurts. Brand voice is not automatic; he pulled tone from the existing site and still edited the brief.
On data: there are privacy modes you can opt into or out of so you can choose whether data is used for training. He did not claim a blanket EU/B2B guarantee. Security is case by case; talk to their team if company guidelines are the blocker.
Other traps he implied: loosely scoped bots that try to do everyone’s job; you remaining the orchestrator after the specialists work; skipping marketplace or template reuse and rebuilding every bot from zero; grouping work only by short initiatives so nothing compounds.
7-day implementation plan
Day 1
Pick one painful, bounded job (competitive teardown, weekly Ads pull, brief draft). Write it as a job description. Create one Grokbot for that job only. Connect one or two tools that give it context fast. Do not build the six-bot org chart yet.
Days 2–3
Grant Slack and/or email if that is where work arrives. Ask the bot to study that history and propose a job it can take. Run one low-risk task end to end. Give feedback the way you would a teammate. If it asks permission for the same action repeatedly and the output is sound, tighten permissions only for that class of action.
Days 4–5
Add the next specialist in the chain (researcher then product marketer, or analyst if you already have campaigns). Practice an explicit handoff: “go get the analysis from that bot.” Put the brief in Google Docs and comment in the file, then send the bot back to the file. If you have Ads or a site repo, connect them and demand screenshots. Steal structure from marketplace templates: dictate your taste, pull site tone, point at two or three template URLs.
Days 6–7
Stand up a project-manager bot. Have it study the specialists and your interjections. Make it your only point of contact for one campaign-shaped workflow. Share your best bot as a template. Do not fully automate spend. Use the week’s artifacts — Doc, sheet, PR, Ads screenshots, analyst TL;DR — as the bar for whether the team is real.
By the end of the week you should have one specialized bot that finishes work in a real tool, one handoff that you did not copy-paste yourself, and a written rule for what still needs you.
Campaign orchestration is the craft. Staff Grokbot so research, positioning, web, ads, and analysis can run as teammates, then put one project manager in front of them so you direct instead of trafficking. Start with a single tightly scoped bot, give it access, and invest in feedback until the handoffs happen without you.
Start with one bot, one job description, and one connected tool — then watch the walkthrough when you are ready to staff the rest of the team.
FAQ
Do I need Grokbot if I already use ChatGPT-style chatbots?
Only if you want work finished in your tools, not answers in a thread. Josh’s contrast is thought partner versus always-on teammate: one bot per job, messaging like Slack, bots collaborating, a cloud computer and browser, 24/7 reporting pulls. If you keep treating it as search or a single LLM, he says it will not be that valuable. If you hook it up and throw it real tasks, that is the test.
Grokbot vs copilots you have to babysit — what is actually different?
Copilots, as he defined them, still need you to push complex work along. Grokbot is positioned as ambitious and proactive: you delegate a job, it keeps going without you reminding it, and it can complete work in connected platforms. You still review. You do not have to sit in the thread to keep the job alive.
Are the marketing bots from the demo on the marketplace?
Not at the time he presented. He had scraped and consolidated marketplace bots plus internal ones, and said the refined set would be pushed. Until then, copy adjacent templates, or build the way he did: dictate your approach, pull site voice, scrape template URLs into your bot, optionally vet through Superme.
Should I fully automate or keep a human in the loop?
Start gradual. Simple task, one or two tools, proof it delivers, then more scope. Do not hand over production, paid, or “always allow” on day one. He still used human comments on the brief, a human nudge to ship the page, and a human block on live budget. Analyst recommendations can stay recommendations until you trust the bot’s judgment.
What is the biggest blocker for marketers using this?
A wrong mental model. New users open the product and do not see what it can do, so they use it like Google or another LLM. The blocker is not a missing feature in his answer — it is refusing to let the bot do things, starting with messy real work like Slack triage.
Should I assign company roles or group bots by initiative?
Prefer always-on roles (researcher, product marketer, and so on) that get better over time. Initiatives start and stop. Put role-bots into a group chat when a project needs a tiger team, and give that chat the initiative. Do not rebuild a new bot for every campaign if the swim lane is the same.
Can I give Grokbot a consistent personality, or connect it to something like Openclaw?
Each bot has its own context and memory, so you can instruct voice (his joke example was Harry Potter calling you a wizard). Avatars can be customized and animated; he had run Mashimaro themes. You can point a Grokbot at an existing agent and ask it to replicate that setup. Native talk to Openclaw was not available; he pointed to import templates as the path to discuss.
Is my data used for training, and can I connect Grok CLI work?
There are data privacy modes to opt in or out of training. Company security is case by case — he sent regulated buyers to their team rather than promising a universal control. On Grok CLI / Grok build CLI: he had not done that exact hookup himself, but described the landing-page path (Cursor and cloud agents pushing a PR) as the pattern. Point Grokbot at existing work and treat it as a harness so you do not rebuild context.