ChatGPT Prompts for Sales Pipeline Management: A Weekly Review Kit

Written by

It is Monday morning, the CRM says you are on track, and you have a quiet feeling that half of those deals are not real. The fix is not a new tool. It is a short weekly review where ChatGPT does the sorting and you make the calls. Below are the ChatGPT prompts for sales pipeline management I would paste, in this order: check the export, find the stalled deals, give every deal a real next step, sanity-check the forecast, and write the follow-ups that move deals forward.

Imagine it is 9:30 on Monday, your coffee is still warm, and you already know the five deals that need you this week, the three you can close out, and the follow-up drafts are waiting in your outbox.

Want one practical AI workflow like this every week? Start free at RobotsHelp Launchpad.

No password needed – your reset opens instantly.

No password. Just your email, and you’re in.

What pipeline management with ChatGPT actually looks like

ChatGPT does not replace your CRM, and it should not update it for you. Its job in this routine is narrow: read a pipeline export, spot the gaps a tired human skims past, and draft the next messages. You stay the one who decides which deal gets pushed, parked or closed.

It works with an export from HubSpot, Pipedrive, Salesforce or a plain spreadsheet. OpenAI’s own guide to data analysis with ChatGPT says it can read CSV and Excel files and works best with clear column headers in the first row and one record per row. So one row per deal, plain column names, no merged cells.

Before you upload anything, strip what you do not need: personal phone numbers, private email addresses, anything your company rules keep inside the CRM. Deal name, stage, amount, dates and a short note are enough for every prompt below.

Step 1: Export the pipeline and let ChatGPT check the data

Export your open deals with these columns: Deal, Company, Stage, Amount, Close date, Last activity, Next step, Notes. Then start with a pure data check. No advice yet, just the facts that are broken.

You are my sales operations assistant. I am uploading my open pipeline as a CSV.
Columns: Deal, Company, Stage, Amount, Close date, Last activity, Next step, Notes.

Check the data only, no advice yet:
1) deals with a close date in the past
2) deals with an empty or vague next step ("follow up", "check in")
3) deals with no activity in the last 14 days
4) possible duplicate deals (same company, similar name)

Give one short table per check. Do not guess missing values. If a column is unclear, ask me one question before you start.

This first pass alone usually shows the problem. Close dates that already passed are the classic sign of a pipeline that looks healthier than it is.

Step 2: Test each deal against its stage

A stage name only means something if you know what has to be true to leave it. HubSpot’s sales pipeline walkthrough puts it well: each stage needs an objective exit criterion, such as “discovery completed with decision maker”, not “feels warm”. Write yours down once, then let ChatGPT check every deal against them.

Here are my pipeline stages and what must be true to leave each one:
- Discovery: we know the problem, the budget range and who decides.
- Proposal sent: the decision maker has seen the proposal.
- Negotiation: price and start date are being discussed in writing.

For each open deal, check whether the Notes show evidence that it meets the criteria of its CURRENT stage.
Mark each deal: Supported, Unclear or Not supported.
Quote the sentence from the notes that proves it. If there is none, say "no evidence".

Deals marked Not supported are not failures. They are just sitting one stage too high, and moving them back makes every number after this more honest.

Step 3: Sort the stalled deals into push, park or close

Most dead deals do not die with a no. They go quiet. This prompt forces a decision on every quiet one, so they stop inflating the total.

Take the deals with no activity in the last 14 days and sort them into three groups:

PUSH: clear buyer interest in the notes, but no next step. Say what is missing.
PARK: real, but the timing is later. Suggest a new close date and a check-in date.
CLOSE: no reply after two or more attempts, or the buyer said no or chose someone else.

One line per deal with the reason, based only on the notes. Do not invent buyer intent.

Step 4: Give every deal a real next step

The HubSpot walkthrough has a simple rule for pipeline reviews: if you cannot name the one scheduled next step for a deal, it does not belong in the forecast. “Follow up” is not a next step. “Client reviews the proposal with their CFO by Thursday” is.

For every deal marked Supported or Unclear, rewrite the Next step field as ONE buyer action with a date.
Good: "Anna reviews the proposal with her CFO by Thu 22 Oct."
Bad: "Follow up", "Send info", "Waiting".

If the notes do not show a buyer action, write "MISSING" and suggest the single question I should ask the buyer to get one.
Output a table: Deal, Old next step, New next step or MISSING, Question to ask.

Copy the new next steps into your CRM yourself. That two minutes of typing is also your sanity check that the suggestions make sense.

Step 5: Sanity-check the forecast

Your CRM probably shows a weighted number. In HubSpot’s default setup, each deal stage has a probability, and the weighted amount is the stage total multiplied by that probability. That is useful, but it trusts the stage. This prompt checks the evidence instead.

Take the deals with a close date in this month and split them into three groups:
COMMIT: decision maker named, price discussed, and a dated next step agreed by the buyer.
BEST CASE: two of those three.
PIPELINE: the rest.

Show the total amount per group. Then list every deal whose stage in my CRM looks more advanced than its group here, with the missing evidence.
Show your calculation so I can check it.

That last line matters. OpenAI recommends reviewing the code, outputs and assumptions whenever ChatGPT runs calculations on your data, and asking it to use a specific method when the result depends on one. If a total looks off, ask it to show the rows it summed.

Salesperson prompts for ChatGPT discovery call follow-up emails

The fastest way to move a deal out of discovery is a follow-up that lands the same day, in the buyer’s own words, with a clear next step. Paste your rough call notes straight in, typos and all.

Here are my notes from a discovery call: [paste notes].

Write a follow-up email to [name, role] that:
- thanks them in one line, no flattery
- sums up their problem in their own words (quote them where the notes allow)
- names the one priority I heard
- proposes the next step with two time options
- ends with one open question about who else should be involved

Under 150 words. Plain and warm, no hype words, no "I hope this finds you well".

For a deal that went quiet after the proposal, you need a different email. Many salespeople find that giving the buyer an easy way out gets more honest replies than another “just checking in”.

Write a short check-in for [name] about the [deal] proposal I sent on [date]. No reply since.
Three sentences max:
1) one specific detail from our last conversation
2) a simple yes or no question about the next step
3) an easy out: "If the timing changed, just reply 'later' and I will check back in [month]."
No guilt, no fake urgency.

This is for deals already in your pipeline. For cold outreach and multi-step sequences, see ChatGPT for sales outreach sequences.

Step 6: The Monday summary you send yourself

Finish the review with one short summary. Keep last week’s summary in the same chat, or paste it in, so ChatGPT can compare.

Summarize this pipeline review in 8 lines max:
- total open pipeline and the change since last week's summary (pasted below)
- deals moved forward, moved back, closed
- the 5 deals that need me this week, each with one action
- one risk I should not ignore

Plain text. Use only the data in this chat.

If you want that summary to arrive by itself every Monday, the n8n CRM automation workflows show how to build it as a small automation. And if your deal notes start life as messy call notes, summarizing meeting notes with ChatGPT gives you cleaner input for every prompt above.

What one deal looks like before and after

A made-up example, so you can see the shape of the output:

BEFORE
Deal: Northwind, onboarding package | Stage: Proposal sent | Amount: 8,400
Last activity: 19 days ago | Next step: follow up

AFTER THE REVIEW
Stage check: Unclear (no note that the decision maker saw the proposal)
Group: PUSH
Next step: MISSING. Ask: "Has the proposal reached your operations director yet, or should I walk them through it in 15 minutes?"
Forecast: PIPELINE, not COMMIT

Same deal, same CRM. The difference is that you now know what to do with it today.

Mistakes that make the review useless

Letting ChatGPT guess. Every prompt above says “based only on the notes” for a reason. Without that, it fills gaps with plausible buyer intent.
Reviewing every deal in detail. Let the prompts sort all of them, then spend your own time on the largest and the late-stage ones.
Skipping the stage criteria. Step 2 is the one people drop, and it is the step that makes the forecast honest.
Retyping the prompts every week. Save them once in a note or text snippet tool, and the whole review takes about the time of one coffee.

FAQ

Can I do this on the free ChatGPT plan?

For a small pipeline, yes: you can paste the table as text instead of uploading a file. File uploads and limits depend on your plan, so check OpenAI’s data analysis guide for what yours supports.

Do these prompts work in Claude or Gemini?

Yes. They are plain instructions with no ChatGPT-only features, so they work in any chat assistant that can read a table.

How often should I run a pipeline review?

Weekly, on the same day, works for most small teams and solo sellers. Daily turns into busywork, and monthly finds the stalled deals too late.

Is it safe to upload CRM data?

Follow your company’s rules first. Remove personal contact details you do not need, and check your ChatGPT workspace settings for how chats and files are stored.

Start with one review this week

Export the pipeline, run Step 1 and Step 3, and close out the deals that are already gone. That alone gives you a cleaner Monday. Want more workflows like this, one practical win at a time? Join free at RobotsHelp.