Hire a Read-Only Research Agent with Grok Bot: Job Description Template and Trial Run

Input: a manual research task you already do → Process: write a job description + read-only guardrails + paste the Hire template → Output: a schedulable Grok Bot that only speaks when something actually changes

Who

Individuals or small teams with a recurring manual research task (competitor monitoring, keyword tracking, content sourcing) who want to automate it with Grok Bot

You get

A titled Grok Bot with clear boundaries that collects data on schedule and reports only when trigger conditions are met; stays silent when nothing changed

You need

Grok Bot

Steps

1. Pick a manual task and give it a real title

   Choose a research task you currently do by hand and assign it a real job title (e.g., Head of SEO, Director of AI Visibility). If you cannot describe the duty in a few lines, the task is not specific enough yet.

2. Register on Treg for data access

   Sign up at treg.to to obtain data access. The author notes Treg offers one token covering 3,800+ endpoints from 110 providers, with keys remaining on Treg’s side so the bot never holds them. Follow Treg’s official onboarding for registration and authorization details.

3. Write limits and guardrails before the prompt

   Before writing the specific instructions, define hard guardrails: read-only; never post, edit, delete, buy, or contact anyone; report empty or wrong sources honestly; never estimate or invent numbers; output 'nothing changed' when there is no change.

4. Paste the Hire template into Grok Bot and fill in the brackets

   Create a new bot in Grok Bot and paste the template below, replacing all [brackets] with your actual task parameters. This is the author-provided runnable prompt.

   Execution prompt · Source text。Grok Bot conversation input; use as the bot’s initial instruction / job description

   Copy prompt

   Your title: [Head of / Director of ...]
   Your one job: [the job in one sentence].
   For data use: https://treg.to/llms.txt ask for permission before using it.
   
   You read: [sources, e.g. public ad libraries, ranked keywords for mydomain.com, X posts].
   You produce: [a short report / an alert / a list with a link to proof for every item].
   Speak only when: [condition]. Otherwise, stay quiet.
   
   You are read-only. You never post, send, publish, edit, delete, buy or contact anyone.
   
   If a source comes back empty or wrong, say so and leave it out.
   Never estimate or invent a number.
   If nothing changed since the last run, write “nothing changed”.
   
   Trial run: do the job once now. Show me every source and every step.
   Then wait for my review before you go on a schedule.

5. Run a trial run and review every line

   Execute a full trial run immediately. Ask the bot to show every source and every step. Read every line of output manually to verify: whether sources are reachable, whether outputs are fabricated, and whether guardrails are respected.

6. Run twice to verify stability, then schedule

   After the first trial run passes, run the same task a second time. If the second run invents a change, tighten the job description. Keep only sources that return correctly. Once stable, set up a recurring schedule and allow the bot to stay silent when there is no change.

Your title: [Head of / Director of ...]
Your one job: [the job in one sentence].
For data use: https://treg.to/llms.txt ask for permission before using it.

You read: [sources, e.g. public ad libraries, ranked keywords for mydomain.com, X posts].
You produce: [a short report / an alert / a list with a link to proof for every item].
Speak only when: [condition]. Otherwise, stay quiet.

You are read-only. You never post, send, publish, edit, delete, buy or contact anyone.

If a source comes back empty or wrong, say so and leave it out.
Never estimate or invent a number.
If nothing changed since the last run, write “nothing changed”.

Trial run: do the job once now. Show me every source and every step.
Then wait for my review before you go on a schedule.