Test a Grok Bot Gmail Connector

August 11, 2026 - Test two Grok Bot Gmail drafts and write an evidence verdict.

Introduction

30 Second Summary

Weekly hand-copy routines feel dependable because they are familiar. They also consume time without proving whether software could handle the same job.

In this project, you will test whether Grok Bot can create two named drafts in Gmail without manual copying. Your evidence packet turns two direct Gmail readings into an ADOPT, REJECT, WATCH, or PARK verdict.

What You'll Build

Picture opening Gmail after two connector tasks and finding the exact named drafts ready for a measured decision about your weekly routine.

By the end of this project, you'll have:

A timed evidence packet that shows exactly what happened from connector authorization to the final decision.
Two separately issued draft requests that test whether the connector can repeat the same kind of work without sending an email.
A path-specific operational verdict backed by two Gmail readings, numeric results, wall time, and a reversal trigger.
Secret Mission: Audit each present draft's recipient, subject, and body without repairing the evidence. Then retain or revise the verdict.

Are there any prerequisites?

Your setup needs an internet-connected Windows device where Grok Bot is already installed and signed in.

You also need an existing Gmail account that you have permission to authorize for the connector.

Before We Start

This checkpoint fixes the evidence standard before the Gmail connector can influence your result. It also starts the clock for your 30-minute run.

Install and Authorize the Gmail Connector

The test needs one controlled path from Grok Bot to Gmail. Adding another connector would change the setup you are measuring.

In this step, you will add the Gmail connector to the existing signed-in Grok Bot desktop app. You will complete one browser sign-in for its connector authorization.

The displayed grant becomes part of your Connector Decision Record. Nothing else is installed in this measured path.

In this step, get ready to:

Add Gmail as the only connector in Grok Bot.
Authorize Gmail through one browser sign-in.
Record the displayed grant in your Connector Decision Record.

Install the Gmail connector

Grok Bot groups its connectors under Plugins. Your test plan allows Gmail as the only addition.

Bring the signed-in Grok Bot app from earlier into view.
Select Plugins in the sidebar.
Find Gmail among the available connectors.

Choosing Add may hand the next stage to your browser. This handoff counts as the single browser sign-in in your plan.

Choose Add on the Gmail listing.

You should see an authorization page in your browser or Waiting for authorization in Grok Bot. This confirms that the connector flow has started.

Authorize the connector

Authorization can feel sensitive because it links your Gmail account to the connector. Your evidence record captures only the details shown on screen.

Sign in through the browser with the Gmail account you plan to test.
Capture the displayed Gmail account here: your authenticated Gmail account.
Review the provider name shown on the authorization screen.
Copy the displayed permissions into the Connector Decision Record exactly as shown.
Record permissions not displayed: unknown if no permission list appears.
Complete Authorize or Authenticate in the browser.
Choose Reopen if Grok Bot shows Waiting for authorization.
Return to the Grok Bot app after the browser confirms authorization.

The sensitive part is complete: the browser authorization has finished. The next check ties that grant to the installed connector.

Browser flow did not complete?

If Grok Bot remains on Waiting for authorization, choose Reopen. Complete the browser flow with the same Gmail account.

Use this authorization troubleshooting prompt for help with the screen you see. You can also ask for help inside NextWork without sharing account details.

Record the grant and confirm installation

Record Gmail as the provider in the Connector Decision Record.
Record your authenticated Gmail account as the authenticated account.
Replace authorization: not yet granted with authorization: granted.
Preserve every measurement field below the authorization entry exactly as it appears.

Before the final check, where should the Gmail connector appear if this setup worked?

Select Plugins in Grok Bot.
Open Installed.
Confirm that Gmail is listed.

You should see Gmail under Installed. That is the controlled setup complete: one connector is authorized without adding anything outside the plan.

Your Connector Decision Record now identifies Gmail as the provider. It names the authenticated account.

The record also contains the displayed permissions or permissions not displayed: unknown. Its authorization line now reads authorization: granted.

The destination fields remain unmeasured. The manual-move count remains 0.

Gmail not listed under Installed?

Return to the browser to confirm that the authorization flow finished. Use Reopen if Grok Bot still shows Waiting for authorization.

Ask for help with a missing Gmail connector if the browser completed but Gmail remains absent.

Your measured connector path is ready. Next, you will issue the first named draft task without opening Gmail.

Issue the First Named Draft Task

Your authorized Gmail connector is ready inside Grok Bot. Now you can issue one controlled request for a named draft without sending a message.

This request creates a binary destination target with a unique subject. Gmail remains unopened so the later inspection can measure the connector's work without manual interference.

In this step, get ready to:

Attach Gmail explicitly to Task 1 with @.
Issue the exact unsent-draft request within the one-time approval boundary.
Record Task 1 as issued while leaving the destination result unmeasured.

Attach Gmail

A connector attachment routes this specific message through Gmail. Explicitly attaching it keeps the test path visible in the conversation.

Switch back to the Grok Bot conversation from earlier.
Click the message composer.
Type @ to display the available connectors.
Select Gmail to attach it to Task 1.

You'll see Gmail attached to the pending message. This proves that Task 1 is routed through the one connector under test.

Gmail Missing From the Attachment List?

Return to the Plugins area from earlier.
Confirm Gmail still appears under Installed.

If Gmail still does not appear in the attachment list, ask for help before issuing Task 1. Help me confirm why my installed Gmail connector is missing from Grok Bot's attachment list.

Issue Task 1

A uniquely named draft gives you a binary target to inspect later. The exact body helps distinguish the requested object from unrelated drafts.

Enter the Gmail account from your Connector Decision Record here: your authenticated Gmail address.
Prepare Task 1 by copying this request into the Grok Bot composer:

Using Gmail, create and save an unsent Gmail draft addressed to [[GMAIL_ADDRESS="your authenticated Gmail address"]]. Use the exact subject GROK-BOT-CONNECTOR-TEST-1 and the exact body Object one created by Grok Bot. Do not send it. The required result is a draft object inside Gmail, not text returned in this conversation.

What Does This Request Control?

The unique subject GROK-BOT-CONNECTOR-TEST-1 gives Task 1 a searchable destination target.
The exact body Object one created by Grok Bot. defines the requested content.
The instruction Do not send it. keeps the requested result inside Gmail Drafts.
The final sentence defines a Gmail draft as the required result.

Check the recipient against the Gmail account in your Connector Decision Record.
Send Task 1 from the Grok Bot composer.

An approval can feel high-stakes, so keep this one limited to creating the requested unsent draft.

If an approval appears, inspect its proposed target.
Inspect the proposed inputs.
Select Allow once only when the operation creates the requested unsent draft.
Select Deny if the operation sends a message.

Approval Does Not Match the Draft?

Choose Deny if the target or inputs include sending a message. Do not manually create the draft to replace a denied operation.

If the proposed action is unclear, pause the test before approving it. Help me review this Grok Bot approval against the Task 1 no-send boundary.

Record Task 1 as issued in the Connector Decision Record.
Keep Gmail closed.
Leave every draft unchanged.
Keep first look: not run unchanged.
Keep named objects produced: not measured unchanged.
Keep manual moves: 0 unchanged.

Grok Bot may report completion in the conversation. That report does not count as destination evidence.

Verify Task 1

The issued message is the checkpoint for this step. Draft existence remains deliberately unmeasured.

Before you inspect the message, do you expect all three issuance markers to be visible?

Review the visible Task 1 message in the Grok Bot conversation.
Confirm Gmail is explicitly attached.
Confirm the request contains the exact subject GROK-BOT-CONNECTOR-TEST-1.
Confirm the request contains the instruction Do not send it..
Confirm the Connector Decision Record says Task 1 was issued.

You'll see Gmail attached to the issued message. You'll also see the exact subject inside the request.

The request contains the no-send instruction. No destination result appears in the Connector Decision Record.

Your first controlled connector task is now issued with its safety boundary visible.

The first binary target is set. Next, you'll issue a second independent task to test whether the connector repeats the same path.

Issue the Second Named Draft Task

Task 1 gave you one named target. A weekly workflow needs repeatable evidence from the same connector path.

Now you'll issue a second independent request in the existing Grok Bot conversation. Gmail stays closed until both requests are on record.

In this step, get ready to:

Attach Gmail explicitly to a new task.
Issue the second exact draft request.
Preserve the zero-manual-move boundary.

Attach Gmail to the new task

Explicit attachment keeps the second request on the same connector path as Task 1. It also makes the issued message auditable later.

Return to the existing Grok Bot conversation from Task 1.
Select the empty message field.
Type @ to display the available connectors.
Select Gmail from the displayed list.

You should see Gmail attached to the new task before any text is sent.

Gmail Does Not Appear?

Keep the task unsent. Type @ again.

Select Gmail before continuing. Help me attach Gmail without issuing an extra task.

Prepare the second draft request

The second subject gives this run a distinct binary target. Its body supports the later integrity audit.

Store the same authenticated address for this request: your Gmail address.
Paste the task into the message field using the block below:

Using Gmail, create and save an unsent Gmail draft addressed to [[GMAIL_ADDRESS="your Gmail address"]]. Use the exact subject `GROK-BOT-CONNECTOR-TEST-2` and the exact body `Object two created by Grok Bot.` Do not send it. The required result is a draft object inside Gmail, not text returned in this conversation.

What Does This Task Specify?

The subject GROK-BOT-CONNECTOR-TEST-2 identifies only the second test object.
The body gives the object a known payload for later auditing.
The no-send sentence keeps the request inside the draft boundary.
The final sentence makes a Gmail draft the required artifact.

An approval can authorize an external mailbox action, so pause long enough to read it. Your boundary permits one unsent-draft creation.

Submit the task as a separate message.
Inspect the proposed target if an approval appears.
Inspect the proposed inputs if an approval appears.
Choose Allow once only for creation of the requested unsent draft.
Choose Deny for any send operation.
Keep Gmail closed.
Leave every draft object untouched.

Confirm the second task is issued

The visible conversation records your two test inputs before destination inspection. Each request needs its own issued message.

Before you check, ask yourself how many separately issued Gmail tasks you expect to see.
Count the separately issued Gmail tasks in the conversation.
Confirm that the count is 2.
Confirm that Task 2 shows Gmail attached.
Confirm that Task 2 contains the exact subject GROK-BOT-CONNECTOR-TEST-2.
Confirm that Task 2 contains an instruction not to send the draft.
Update the Connector Decision Record with tasks issued: 2.
Keep manual moves: 0 unchanged.

You should see exactly two separately issued messages. Task 2 shows Gmail as its attached connector.

Its text includes GROK-BOT-CONNECTOR-TEST-2. It also says not to send the draft.

Task 2 Does Not Match?

If the message is still unsent, correct the Gmail attachment first. Check the prompt again before submitting it.

If the wrong message is already issued, pause before creating another task. Preserve the two-task evidence boundary while you record the divergence.

Help me preserve the evidence boundary after a Task 2 problem.

That is the repeatability input locked in. Next, you'll inspect Gmail directly for the first destination count.

Inspect Gmail Directly

Your two named draft tasks are now issued through the authorized Gmail connector. The next question is whether those objects reached Gmail.

A Grok Bot completion report cannot prove the drafts exist in the destination. This step gets the first reading directly from Gmail without changing the evidence.

In this step, get ready to:

Open Gmail independently from the Grok Bot conversation.
Search Drafts for both exact subject lines.
Record the first numeric destination reading without changing any draft.

Open Gmail as the evidence source

Destination-side verification uses the system that should contain the result as the evidence source. For this run, only Gmail can prove whether the requested drafts exist.

In your browser, open Gmail.
Sign in with the same authenticated Gmail account if prompted.
Select Drafts in the left sidebar.
Update the Connector Decision Record with Destination opened and inspected directly: yes.

You should see the Drafts view for the authenticated Gmail account.

Why Does Gmail Count as Evidence?

A completion report describes what Grok Bot says it attempted. Gmail shows whether the requested object reached the destination.

This evidence boundary keeps the first reading tied to objects you can inspect directly.

Search both exact subjects

Each exact-subject search turns one requested draft into a binary artifact check. The two results combine into a first reading out of two.

Before you search, how many of the two named draft objects do you expect Gmail to show?

Search for the first draft by entering subject:"GROK-BOT-CONNECTOR-TEST-1" in Gmail's search bar.
Count Task 1 as present only when Gmail shows the named draft under Drafts.
Select Drafts again in the left sidebar.
Search for the second draft by entering subject:"GROK-BOT-CONNECTOR-TEST-2" in Gmail's search bar.
Count Task 2 as present only when Gmail shows the named draft under Drafts.

Your two searches now give you an observed count of zero, one, or two named draft objects.

Search Results Look Unclear?

Return to Drafts before repeating each exact-subject search.
Count only a draft object whose subject exactly matches the search.

Help me make sense of my Gmail Drafts search results.

Record the first destination reading

This first look preserves the destination count at this moment. The next step checks whether that count stays stable.

Exclude the Grok Bot report from the count.
Exclude any screenshot of the Grok Bot conversation from the count.
Exclude copied Bot text from the count.
Leave every draft object unchanged.
Confirm that manual moves: 0 remains unchanged in the Connector Decision Record.
Replace First look: not run with First look: 0, 1, or 2/2, based only on named draft objects present in Gmail after exact-subject searches for "GROK-BOT-CONNECTOR-TEST-1" and "GROK-BOT-CONNECTOR-TEST-2".

You should see a numeric first-look count in the Connector Decision Record. The destination confirmation should read Destination opened and inspected directly: yes.

That is your first defensible destination reading. It comes from Gmail itself.

Frame one exact-subject Gmail result so personal mail details stay outside the screenshot.

Your first destination reading is recorded. Next up, you will repeat both searches from scratch before writing the verdict.

Take the Second Look and Write the Verdict

Your first Gmail reading gave you direct destination evidence for both named tasks. The clock remains running because one early reading cannot close the decision.

A second reading checks whether the count holds when you repeat the search from scratch. You will turn both readings into a reversible verdict for this one connector path.

In this step, get ready to:

Confirm the destination count with a second Gmail inspection.
Reconcile the evidence packet with the actual run.
Write one evidence-based verdict with its timing and reversal rule.

Confirm the count on a second look

Repeated observations reveal whether a destination result stays stable. Both readings remain in the record because a changed count is part of the evidence.

Before the second look, ask yourself whether the count will match your first reading.

Reload the existing Gmail tab from your first look.
Select Drafts from Gmail's left sidebar.
Enter subject:"GROK-BOT-CONNECTOR-TEST-1" in the Gmail search box.
Record whether the first named draft object is present.
Clear the first search query.
Enter subject:"GROK-BOT-CONNECTOR-TEST-2" in the Gmail search box.
Record whether the second named draft object is present.
Record the combined count as second look: X/2 in the Connector Decision Record.

What does the second look prove?

Matching counts show that the repeated destination check held steady.

Different counts reveal instability that must stay visible in the verdict. The original first reading remains part of the evidence.

Compare the second-look count with first look: X/2.
Record second look confirmed in destination: yes when the counts match.
Record second look confirmed in destination: unstable when the counts differ.
Preserve the original first reading when the counts differ.

Reconcile the run with the pre-run record

The pre-run record protects your result from hindsight. Reconciliation makes the packet describe the path you actually tested.

Record named objects produced: X/2 using the second-look count.
Record manual moves: N using the actual number of objects manually created, copied, moved, or repaired during the run.

Your measurement section now contains a numeric task result plus a separate manual-move count.

Compare the actual connector path with the original diagram.
Compare the actual run with the original plan.
Update the diagram if the path diverged.
Update the plan if the run diverged.

The diagram and plan now describe the observed run instead of a reconstructed version.

Record any divergence you found.
Preserve the dated publication anchor from the pre-run record unchanged.
Add the scope limit one connector on one service, no class-wide conclusion.

Why keep the publication question open?

The dated Cursor Gmail listing publishes Search, read, draft, and manage email..

The listing leaves these two named task outcomes unanswered. Your destination evidence answers only this controlled run.

Calculate the wall time and choose the verdict

An operational verdict converts the measurements into a reversible decision. Each outcome has a numeric threshold plus a specific condition for reconsidering it.

Which verdict fits the evidence?

Use ADOPT when both direct readings are 2/2 with 0 manual moves. Reverse it when an adopted weekly run produces fewer than 2/2 complete objects or requires more than 0 manual moves.
Use REJECT when two stable readings are below 2/2 or the run requires more than 0 manual moves. Reverse it when a fresh controlled run produces two stable 2/2 readings with 0 manual moves.
Use WATCH when the readings are unstable or work remains visibly in progress. Set the re-check date to 2026-09-06. Reverse it when two stable destination readings establish either 2/2 with 0 manual moves or a lower result.
Use PARK when connector authorization or destination access prevents a valid run. Set the re-check date to 2026-09-30. Reverse it when the blocker is removed and the complete two-task run can be repeated.

Record the finish time in the Connector Decision Record.
Subtract the recorded start time from the finish time.
Write the result as total wall time: N minutes.

The timed run now has a measured endpoint. Connector setup remains inside that total.

Choose exactly one verdict from ADOPT, REJECT, WATCH, or PARK.
Record the reversal trigger that belongs to your chosen verdict.

Your decision now states what the current evidence supports plus what future observation would change it.

Record not required as the re-check date for ADOPT or REJECT.
Record 2026-09-06 as the re-check date for WATCH.
Record 2026-09-30 as the re-check date for PARK.
Add a completed version of this required evidence line using your measured values:

Total wall time: N minutes | Setup installed: 1 Gmail connector; browser sign-ins: 1; anything else installed: 0 | Platform: Windows; macOS: reasoning-only, not tested; Linux: not tested, desktop app unavailable in checked documentation | Deliverables: 4/4 in one project | Tasks issued: 2 | Named objects produced: X/2 | Manual moves: N | Destination opened and inspected directly: yes | Second look confirmed: yes or unstable | Published anchor: Cursor Gmail plugin listing checked 2026-08-30, publication date not displayed, published `Search, read, draft, and manage email.`, task-completion question unanswered before run | Limit: one connector on one service, no class-wide conclusion | Authorization: recorded or unknown, retained or revoked | VERDICT: ADOPT, REJECT, WATCH, or PARK | Reversal trigger: recorded | Re-check date: not required or YYYY-MM-DD.

What does this evidence line capture?

The timing fields keep connector setup inside the measured run.
The destination fields preserve both direct observations plus their stability status.
The publication field separates the published capability from the unanswered task-completion question.
The verdict fields connect the decision to its reversal trigger plus any required re-check date.

Save the completed Connector Decision Record.

Before you review the finished record, ask yourself whether every decision field traces back to direct Gmail evidence.

Confirm that the record contains both destination readings plus their confirmation status.
Confirm that the record contains the numeric task result plus the numeric manual-move count.
Confirm that the record contains the finish time plus total wall time.
Confirm that the record contains one verdict plus its reversal trigger and re-check date.
Confirm that the record retains the dated publication anchor plus the one-connector scope limit.

You should see one completed evidence packet with no empty measurement, timing, or decision fields.

Verdict still unclear?

Use the two destination readings plus the manual-move count as the decision inputs. Preserve any instability instead of averaging it away.

Check that only one verdict rule matches the evidence recorded in your packet.

Help me map my readings to the correct verdict rule.

That closes the timed run. Your verdict now rests on two direct Gmail readings with a measurable path for changing the decision.

Secret mission

Audit the Objects Without Repairing Them

A matching subject can hide a wrong recipient or body. Open each present draft to audit every field. Preserve the verdict or revise it downward without repairing the evidence.

Clean Up Your Resources

This project adds no ongoing usage cost through your existing Grok Bot and Gmail access. Choose whether to keep the tested path, preserve it until a re-check date, or remove its evidence and connector authorization.

Resources you used:

The completed Connector Decision Record with the object-integrity audit and final verdict.
The account-wide authenticated Gmail connector, if it remains visible under Installed.
Up to two unsent Gmail drafts named GROK-BOT-CONNECTOR-TEST-1 and GROK-BOT-CONNECTOR-TEST-2.

Keep everything running

No action needed. Choose this after an ADOPT verdict if you plan to keep using the tested weekly path.

Retain the Gmail connector while the tested path continues to produce 2/2 complete objects with 0 manual moves.
Preserve the Connector Decision Record as evidence for later weekly runs.
Leave the audited test drafts unchanged if you want to retain the destination evidence.

Pause - I'll come back to this later

No project process needs to be stopped. Choose this after a WATCH or PARK verdict while you wait for the recorded re-check date.

Preserve the Connector Decision Record without changing its measurements.
Leave each audited Gmail draft untouched.
Keep the Gmail connector in the state recorded by the verdict.
Return on 2026-09-06 for a WATCH verdict.
Return on 2026-09-30 for a PARK verdict.

Delete - I don't want to use this again

Remove the project resources after a REJECT verdict or when you no longer need the tested path. This clears the test drafts and removes the connector access.

Preserve the Revocation Record

Deletion is final. Keep the Connector Decision Record until you have recorded the authorization revocation.

Return to Gmail in your browser.
Select Drafts from the left menu.
Search for subject:"GROK-BOT-CONNECTOR-TEST-1".
Use Gmail's delete control to discard the matching unsent draft if it is present.
Search for subject:"GROK-BOT-CONNECTOR-TEST-2".
Use Gmail's delete control to discard the matching unsent draft if it is present.

The test drafts are now cleared from Drafts. The connector connection and its authorization still need separate removal.

Switch back to the existing Grok Bot desktop app.
Select Plugins.
Check whether Gmail still appears under Installed.
Use the connector's uninstall control if Gmail is still listed.
Return to the source service recorded during browser authorization.
Use that service's connected-app access controls to revoke the Gmail connector authorization.
Record authorization: revoked in the Connector Decision Record.
Delete the Connector Decision Record from the location where you saved it if you no longer need the evidence.

Nice Work!

You made it! You tested whether Grok Bot can replace a weekly Gmail hand-copy step. Your evidence packet now supports a scoped operational verdict from direct destination checks.

You've learned how to:

Created a dated evidence packet for the Windows run. The record preserves the authorization details. It also preserves the dated publication anchor. It finishes with the measured wall time.
Issued exactly two named draft tasks through one authorized Gmail connector. Each task kept sending outside the approval boundary. Manual copying stayed outside the test.
Confirmed the outcome through two direct Gmail readings. You used the measured result to write a reversible operational verdict. The record includes its reversal trigger. A WATCH or PARK result also includes its re-check date.
Secret Mission: Audited complete draft integrity by opening each present object. You compared its recipient with the request. You repeated the check for its exact subject. You also checked its exact body. You preserved the destination evidence without repair.

Ready to quiz yourself?