Guide

Grok Bot for SDRs: the official guide in brief, plus our advice

By Atlas AI Resources · Published October 15, 2026 · How we make these guides

Outbound is a long list of small jobs: research the account, find the right people, write the first touch, check who is already in a sequence, create records, and follow up. This page summarizes how one SDR splits that work between Cursor and a small team of Grok Bots, sorts every contact into CLEAR or FLAG, and starts the day with a review list. Then it adds our own fill-in templates for intake checks, flag reviews, and the morning send review. In the guide, nothing auto-sends, and we suggest you keep it that way.

It is based on the official guide Grok Bot for SDRs (September 24, 2026) by Stefan Markarian, written by an SDR who runs outbound this way every morning. The workflows below are summarized in our words; read the guide for the author’s full account.

1. Split the work: Cursor writes, Grok Bot operates

The author’s old workflow lived across six tabs: Hex for product footprint, Salesforce and Gong for account history, Amplemarket for email, Nooks for sequences, and LinkedIn for connection requests. Research alone could take half the morning. In the new setup, Cursor cloud agents build the research packet and write first-touch copy, and Grok Bot takes that work into the sales tools, checks what is already happening there, and keeps working when the laptop is closed. The guide’s five steps:

• Research. Cursor cloud agents pull product usage, closed-lost opportunities, and past conversations from Hex, Salesforce, and Gong to build the account story.
• Prospect. Find people whose roles match that story. Warm contacts and reconnects come first; cold outreach to director-level people happens only when their actual responsibilities fit.
• Draft. Cursor uses writing and research skills to draft first touches in the SDR’s voice, opening with a real account signal rather than a generic template.
• Intake. Prospects arrive from a pipeline-generation list, the Cursor morning packet, or the ChatGTM inbox. Grok Bot checks whether each person is already active in Nooks, paused in Amplemarket, or attached to a Stage 0 opportunity. CLEAR contacts can be enrolled unsent; FLAGs wait for the SDR.
• Finalize. The SDR keeps and sends from Nooks. Blank LinkedIn requests go out as a separate step. Nothing auto-sends.

2. One job per bot

The author does not use one general bot for everything. Each bot has one job, and they hand work to each other so the SDR sees only results and decisions. Josh Allen, Contact Man, and John Ford are the author’s names for bots, not people involved in the guide.

• Chief of staff (named Josh Allen in the guide). Owns the morning process: prepares the next pipeline-generation list, puts final emails in front of the SDR, tracks demo-request follow-ups, fixes contacts assigned to the wrong sequence, and sends a morning update on what changed and what needs attention.
• Contact Man. Owns the prospecting sheet: the keeper list, everyone added to a sequence, and which rows are staged. When the chief of staff enrolls someone, Contact Man updates the sheet so nobody is worked twice.
• John Ford. Owns LinkedIn connection requests, which go out with no note. It checks the daily cap and the number of pending requests first.
• SFDC Create and Sequencer. Handle contact creation and sequence changes when the normal CRM path gets messy.

In a single batch, the chief of staff can ask Contact Man to update the sheet and John Ford to handle LinkedIn. The coordinator keeps the morning moving while each specialist owns its lane. The guide links a shareable version of each bot. Those listings say the bots were created by a third-party user, and their descriptions do not always match the guide word for word, so read each bot’s job description and stop points before you add it, and adjust them to your own team.

3. A morning with CLEAR and FLAG

In the author’s example, Cursor finished a research packet and a batch of drafts before the day started. The chief of staff then checked each contact against Nooks, Amplemarket, and Salesforce, and put every contact in one of two buckets:

• CLEAR: safe to draft and enroll without sending.
• FLAG: something is unclear, and the SDR decides.

CLEAR drafts were checked against the writing rules and enrolled unsent. By the author’s account, six people were already active elsewhere and were left alone, thirty-four had edge cases and went to FLAG review, and one was fully clear, so the SDR reviewed and sent it. The edge cases were the kind only a person should settle: a paused contact in Amplemarket, an early opportunity in Salesforce, or two nearly identical emails.

The same process decides who to prospect. It compares product seats with the size of the account’s software-engineering organization. When the footprint is strong enough, the team targets directors and above from its Golden List; otherwise it works managers and above. The bots also clear the backlog: the author describes 24 contacts with emails in the wrong sequence being cleaned up in one pass, with first calls scheduled for the next afternoon, while the SDR still owned the drafts and final approval.

The guide sums it up: “CLEAR contacts keep moving. FLAGS stop.” Our guide on Grok Bot for founder-led sales uses a similar sort (KEEP, HOLD, DQ) for a founder doing their own outbound.

4. The guide’s rules, in brief

• Never auto-send. Bots research, draft, create records, and stage work. The SDR reviews and sends.
• Warm history beats a cold template. If Gong shows a real earlier conversation, the email starts there.
• Footprint beats title. Hex usage decides when an account has earned director-level outreach; a senior title alone is not enough.
• LinkedIn requests stay blank. If the bot cannot verify the profile URL, it flags the contact instead of guessing.
• Repeated feedback becomes a rule. Writing style, research process, sequence logic, and email format live in reusable skills across Cursor and Grok Bot. If the same problem happens twice, update the instructions.

The guide is clear that none of this replaces the decisions that matter: whether to keep a flagged prospect, how to reconnect when there is real Gong history, and whether two similar contacts are actually duplicates. The SDR still sells.

5. Atlas’s advice: three fill-in templates

What follows is Atlas’s own advice, not the guide’s. Adapt the wording to your team and your tools.

Atlas’s advice: the intake check for each contact

Ask your intake bot to fill one of these per contact, link the record behind each answer, and write “couldn’t check” rather than guess:

• Contact: ____ (title, account, where the lead came from).
• Already active? Sequencing tool: ____ · Email tool: ____ · Open opportunity in the CRM: ____ (yes, no, or couldn’t check).
• Possible duplicate: ____ (similar names, emails, or records found).
• Profile verified: yes or no, with the URL.
• Fit: how the role matches the account story, with the source.
• Warm history: any earlier conversation, with a link.
• Result: CLEAR or FLAG, plus a one-line reason for any FLAG.

Make “couldn’t check” an automatic FLAG. And spell out what CLEAR allows: enroll the draft unsent, nothing more. A bot being sure a contact is clear is not the same as your yes to send, so CLEAR should never unlock a send on its own.

Atlas’s advice: the FLAG review card

• What is unclear: ____ (one sentence).
• Evidence: links to the records the bot checked.
• Options: keep and enroll unsent · skip · merge as duplicate · reconnect from the earlier conversation · something else.
• Bot’s suggestion: ____ (labeled as a suggestion).
• Ask: Which option? Reply with one.

When you settle the same kind of flag twice, decide whether it should become a standing rule. That is the guide’s “repeated feedback becomes a rule” in practice. For copy-ready prompts that keep drafts in your voice and catch changed facts, see the anti-slop pass.

Atlas’s advice: the morning send review

Have the bot that owns the morning send one update you can act on from your phone:

• Overnight: ____ enrolled unsent · ____ left alone (already active) · ____ flags waiting.
• Ready for your review: each draft, with the account signal it opens with.
• LinkedIn requests: ____ queued · today’s cap ____ · pending ____.
• Backlog fixed: ____ (what changed, with links).
• Needs your decision: ____.
• Sent: nothing. Every send is yours.

For a ready-made set of daily digest prompts, see Grok Bot chief of staff prompts.

6. Keep sends behind your yes

We recommend draft-then-approve for anything that reaches a prospect: emails, sequence steps, LinkedIn messages, and replies. The guide’s John Ford sends blank connection requests itself, while the shareable John Ford listing describes drafting LinkedIn messages and waiting for review before any send, so decide up front whether connection requests also wait for you. Grok Bot’s guardrails are plain-language rules checked by a review agent, not code. The current docs put them in Settings → General → Auto-review, as Ask first and Allow automatically rules (the 101 guide calls this Settings > General > Agent, with allow and block lists steering the reviewer). Spell out each bot’s stop points (“enroll unsent only,” “ask before any send,” “flag if the profile can’t be verified”). An approval controls the action it is attached to; it does not undo work already done.

SDR bots read a lot of text written by other people: prospect websites, LinkedIn profiles, inbound requests, and replies. Any of it can contain instructions aimed at your bot, so treat that content as data to read, not orders to follow. Our guide on keeping AI agents safe from prompt injection covers this in more depth.

Our advice: all of your bots share one computer, so a login to LinkedIn or your CRM that one bot uses is reachable by the others. Give each bot only the access its job needs, and check your company’s outreach policy before a bot works in a new tool.

General architecture advice, not a description of how Grok Bot works: if you also run your own scripts against a CRM or sequencing API, enforce hard limits such as daily caps, allowed sequences, and write scopes with exact checks in that code.

7. Where to start

The guide’s own advice: connect the tools you already use, put the research and drafts in Cursor, give one bot ownership of the morning process, keep the final send for yourself, and let it run overnight so the morning starts with a review list. It suggests the GTM Account Research bot from the Bot Marketplace as a first step. Our suggestions for rolling it out:

• Start with one job. An intake check that only reads your tools and sorts contacts into CLEAR or FLAG is a low-risk first bot. Add enrolling unsent once a few weeks of its sorting match yours.
• Connect tools read-only first, then add write access one system at a time, starting with your own prospecting sheet rather than the CRM.
• Keep every message as a draft, and note each correction as a standing rule.
• Save what works as a skill, then schedule it. The overnight run is a routine; prove it by hand first and use safe inputs for the test run, since a test run does real work. Our guide on skills before routines walks through this.
• Add the next bot only when the first one is reliable, then let them hand work to each other so you are not the glue.
• Judge it by finished work: first touches you would actually send, flags you can settle quickly, and nobody worked twice, not how many contacts a bot enrolled.

The same draft-then-approve pattern shows up in sourcing, too: our Grok Bot for recruiting page covers a bot that drafts candidate outreach a recruiter approves. Read the author’s full account in Grok Bot for SDRs.

Useful public resources

Related: Grok Bot for founder-led sales · Grok Bot for post-sales · Grok Bot for recruiting · Skills before routines · Keeping AI agents safe from prompt injection · Grok Bot for PMs · Grok Bot for GTM