Grok Bot for research comes down to one Bot with one research job, an output contract written into the prompt, and a routine you only switch on after a clean test run.
Here is the build order I'd follow, from the plan check to a scheduled Monday brief.
Every setting below maps to xAI's own pages: Use the computer and apps, Files and results, Skills and routines and the Grok Bot changelog.
Want my team to sanity-check the setup before you start? Get your Backlink Gameplan, or try Polar Rank free for 3 days.
Step 1: confirm access, then write down one research question
On The X Subscription, xAI's Grok Bot plan page names Premium, Super and Ultra as the plans that include the Bot, and Plus as the one that doesn't.
You sign in with a Cursor account, and the Bot's computer lives in Cursor's cloud, as the private networks page explains.
Next, narrow the job until you could check the answer.
"Do some research for me" can't be tested.
"Check three named rivals every week for launches and repricing, and link each item" can, and it follows the Competitive intelligence starter on the use cases page.
Pick one question of that shape and write it down before you open the app.
Step 2: map the three doors research comes through
The Bot collects material in three different ways, and each has its own catch.
| Source | How the Bot gets in | Watch out for |
|---|---|---|
| Public X | Built-in read and search, no X account linked | Read-only; your DMs, bookmarks and home timeline need the X plugin |
| Websites | The browser on its cloud computer | Paywalled or private pages need a sign-in or a connected app |
| Your own material | Attachments, Google Drive, Sheets and other connected apps | Say what each file is and whether the Bot may change it |
xAI says a connected app is more dependable than clicking through a website whenever one exists, so connect the app first and let the browser cover the gaps.
No page in the docs describes a separate web-search tool or a deep-research mode, so the web part of the job is the Bot steering its own browser.
Step 3: set up the browser side once
Every Bot on your account works on one persistent cloud computer, and each one gets a screen of its own there.
So a research Bot and a writing Bot can browse side by side, though a single Bot handles just one computer-use task on its screen at a time.
Logins persist and are shared, which means you sign in once to a trade portal or a paid source you're allowed to use, and the rest of your Bots can reach it too.
That sharing is the gotcha: the docs call the screens separate work surfaces, not separate security boundaries, so keep off the computer any login another Bot shouldn't touch.
When a site wants single sign-on, a passkey or a CAPTCHA, the Bot passes you the computer; clear only that step and tell it to continue.
Two changelog lines help here: v0.70.2 lets a Bot save a web page as a PDF from its browser, and from 0.75.2 browser screenshots stay full size, tall pages included.
The full browser walkthrough lives in our Grok Bot browser setup guide.
Step 4: put the output contract inside the prompt
Citations don't appear on their own.
The files docs tell you to name the result and how you'll judge it, for example a document with headings and source links, or a brief recommendation with the evidence underneath.
So the sourcing rule has to sit in your prompt, like this:
Research job: check Acme Plumbing, FlowFix and DrainPro for anything they launched, repriced or announced in the last 7 days.
Sources: their websites (open them in your browser), their public X accounts, and my Google Doc "Competitor notes".
Output: a Google Doc with one heading per competitor. After every claim, add a direct source link and the date you read it.
Then give five separate lists: facts found, your inferences, actions done, actions waiting for my OK, open questions.
End with a section called "Could not verify".
Boundaries: read only. Do not contact anyone, post anything or sign up for anything.
The five-list split is xAI's suggestion for consequential work, and the last section follows the evidence checklist on the same page.
🔥 Want research Bots that are already set up and tested? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members building research workflows like this one. → Join the AI Profit Boardroom
Step 5: run it by hand and audit the links, not the wording
Run the prompt once in chat before anything gets scheduled.
To watch it click, type and move between pages, open the computer preview in the conversation's right-hand panel.
When the doc arrives, pick three links at random and check that each page really says what the claim says.
The evidence list on the files page also covers timestamps with time zones, the names of input and output files, and a short log of actions, so request whichever of those this job needs.
For numbers that shift fast, a screenshot isn't enough on its own; keep a link or an export from the source too.
If something's wrong, fix the same file instead of starting a fresh one: "Update the doc you just made and add source links to the two claims under FlowFix."
Step 6: save the method as a skill
Once a run comes back clean, have the Bot save it as a skill.
The skills docs say a useful one covers when it applies, the inputs and access it needs, the order of steps, how to check the result, what to hand back and what needs approval.
Save what we just did as a skill called "Weekly competitor check". Include the three competitor names, the source list, the Google Doc layout, the rule that every claim gets a dated source link, the "Could not verify" section, and the rule that you never contact anyone.
Private skills live in a single library that all of your Bots share.
If showing beats explaining, Teach a task records a browser workflow for up to ten minutes where the control is switched on, and you get a draft skill to tighten afterwards.
Step 7: schedule it, then press Test
Ask the Bot that owns the job to run the skill on a plain-words schedule, such as "every Monday at 8:00 AM".
Open the routine and read When to run, and set the time zone under Settings, Bot, Timezone if the job belongs to another office.
A fresh routine waits for its next slot, so press Test to see a real run straight away.
Add a missing-data rule as well: if a rival's site won't load, report that instead of reusing last week's notes.
Routines can also start from events, such as a Slack message, an email to the Bot's own address, a webhook call, or a GitHub, Linear, Sentry or PagerDuty event.
Keep triggers tight, because each run spends usage, and the docs warn that a busy-channel trigger can eat a week of usage in a day.
The settings I check before the first scheduled run
| Setting | Where | What I set |
|---|---|---|
| Time zone | Settings, Bot, Timezone | The office the brief is for |
| Ask first rules | Auto Review, Custom Rules | Sending messages, posting and sign-ups always ask |
| Connected apps | Connect apps in the sidebar | Google Drive and Sheets only, for this job |
| Project folder | /workspace on the cloud computer | One folder per study, with clear file names |
| Routine schedule | The routine's When to run | Weekly, never on every new message |
An Ask first rule always stops a matching action, which is the backstop if a prompt ever drifts, as the approvals docs describe.
📞 Not sure which research job to hand over first? Get your Backlink Gameplan and my team will help you pick one and scope it, or try Polar Rank free for 3 days. Backlink Gameplan or try Polar Rank free for 3 days
Splitting a bigger study across three Bots
For a larger piece, the group chat docs show a kickoff with one owner per stage: a researcher that gathers material and links each claim, a writer that drafts from it, and a reviewer that checks the draft against the sources.
Put the three Bots in one group, tag each with @, and finish the kickoff with "Do not publish anything."
Two catches: approval requests never show inside a group, and handoffs from a Bot to the group carry text only, so have the researcher send screenshots straight to the reviewer.
Anything saved under /workspace is readable by every Bot, so give each study its own folder.
Worked example: a plumbing firm's Monday market check
- The owner creates a Bot called Market Watch and pastes the Step 4 prompt with three local rivals.
- The Bot opens each rival's website in its browser and reads the services and offers pages.
- It searches public X for each rival's name and keeps the posts it relies on, with links.
- It builds the Google Doc, one heading per rival, with a dated link after every claim.
- It lists anything it couldn't confirm, such as a page that refused to load.
- On Monday the owner reads the doc and decides whether anything changes.
Worked example: an agency's research pack for a new client
- The agency drops the client brief as a PDF and last quarter's survey as a CSV into the chat, saying what each file is.
- The Bot reads both, then browses the three rival sites the brief names.
- It fills a spreadsheet with fixed columns: request, which rival covers it, source link, source file.
- Where Pages is available, it builds a page with that table and a short recommendation.
- An account manager opens the key links before anything goes to the client.
Gotchas worth planning around
- Memory isn't a source: for figures that move, tell the Bot to look them up again rather than recall them.
- Pages can try to steer a Bot. xAI marks outside content as untrusted and says its defences reduce that risk without removing it, so keep research Bots read-only.
- Auto Review treats only your own messages as instructions; text on a web page or in an email doesn't count.
- Pages is rolling out gradually and isn't on Enterprise teams yet.
- Tagging @bot on X isn't open yet to X accounts registered in Australia, the United Kingdom, the European Union, Norway, Iceland or Liechtenstein.
- Saving to Google Drive always makes a new file, even when one with the same name is already there.
Where the Bot stops and you take over
The Bot gathers, reads, sorts and drafts.
You open the key sources, judge what they mean and make the decision.
That split is what the routines docs mean when they say to automate the preparation before the execution.
If you want help drawing that line in your own business, get your Backlink Gameplan from my team or try Polar Rank free for 3 days, or work through the trainings inside the AI Profit Boardroom.
FAQ: Grok Bot For Research
Does Grok Bot add citations automatically?
No. The docs tell you to ask for them, for example a document with headings and source links. Put the sourcing rule in the prompt and in the saved skill.
Do I need to connect my X account for research?
No. Bots can search and read public X without a linked account, and that access is read-only. Your DMs, bookmarks and home timeline need the X plugin.
Can two research Bots browse at the same time?
Yes. All your Bots share one cloud computer, but each gets its own screen, so several can use the browser in parallel. Each Bot runs one computer-use task at a time.
Why does my new routine not run straight away?
A new routine waits for its next scheduled time. Use Test on the routine to run it now, and check the result in the Bot's chat.
Where should research files live?
In a project folder under /workspace on the cloud computer, which every Bot can read. Put the final result or a link to it in the conversation as well.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery".
- My Udemy courses have taught over 29,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Write the output contract, test the routine once, and join the AI Profit Boardroom for the trainings that go with setting up Grok Bot for research.











