ChatGPT for Sales: What It Does Well, the Prompts and the CRM Connection

ChatGPT is good at the writing and thinking work around a sale: account research, first drafts of emails, discovery questions, objection answers and turning messy call notes into a clean summary. It is only as good as the facts you give it, and on its own it does not keep your pipeline, remember what a colleague promised a client, or send anything. Teams get the most out of it with saved prompts in a Project, a business plan for anything that contains customer data, and, if ChatGPT should also update records, a CRM connected through MCP.

This guide covers prompts by deal stage, a call-note workflow, the data rules per ChatGPT plan, and what OpenAI currently requires before ChatGPT can write to a CRM. Customermates publishes it and appears in one labelled section near the end.

Where ChatGPT helps in sales, and where it does not

The rule that follows from the table: ChatGPT prepares, you decide and send. A wrong fact in an internal note costs a minute. A wrong fact in an email to a buyer can cost the deal.

Set it up once: a sales Project

Before the prompts, give ChatGPT the context it would otherwise ask for every time. In a Project or a custom GPT, add:

• your offer in five sentences, with prices if you name them;
• your ideal client: industry, size, the role that buys, the problem that triggers a purchase;
• two or three short client stories you are allowed to mention;
• the objections you hear most and how you answer them;
• tone rules, for example: "Short sentences. No 'I hope this finds you well'. One question per email."

OpenAI's pricing page lists Projects on every personal plan, Free included, and custom GPTs among the Plus features. Keep one Project for outbound and one for account work, because a Project stuffed with everything gives vaguer answers.

Prompts by deal stage

Replace the brackets with real facts. The more specific the input, the less the output sounds like a template.

Research brief before a first call

I have a first call with [name], [role] at [company], [size], [industry].
Here is what I found: [paste LinkedIn about section, company page, recent news].
Give me: 3 likely problems they have that our offer solves, 3 open questions
for the first 10 minutes, and the one client story from this Project that fits best.
Mark anything you are guessing as a guess.

First-touch email

Write a first email to [name] at [company]. Trigger: [new hire, funding, post, event].
What they care about: [paste]. Our relevance in one line: [line].
90 words at most, plain language, one specific question at the end.
No compliments, no "quick call", no claims about results we have not given you.

Objection answer

The prospect wrote: "[exact words]".
Our usual answer: [paste]. Facts we can use: [paste].
Give me three replies under 80 words: one that asks what is behind the objection,
one that answers it directly, one that offers a smaller first step.
Use only the facts above.

Follow-up after a meeting

Here are my notes from today's meeting with [name] at [company]: [paste].
Draft a follow-up under 150 words: thank them in one line, restate the two or three
points we agreed, list next steps with owner and date as stated in the notes,
and end with the next appointment. If an owner or date is missing, write [missing].

Pipeline review

Here are my open deals with stage, value, last activity date and next step: [paste].
List the deals with no activity for more than 14 days, the deals without a next step,
and for each one a single question I should answer before Monday's review.
Do not estimate close probabilities.

The sales follow-up email guide has more follow-up structures if that is the step where your deals stall.

From call notes to a CRM entry

Summarizing call notes is the most common daily use, and it is where a fixed output format pays off. Paste a transcript or your own notes and ask for fields that match your CRM:

Summarize these call notes for the CRM. Return exactly these fields:
- Summary: 3 sentences (context, what was decided, what happens next)
- Pain points: bullet list
- Decision makers mentioned: names and roles
- Next steps: action, owner, date
- Stage suggestion: one of [Prospecting, Qualification, Demo, Proposal, Negotiation]
If something is not in the notes, write "not stated". Do not add facts.

Notes:
[paste]

Before: "Call with Hanna Roth, Roth Steuerberatung. 6 people, client list in Excel, getting messy. Wants WhatsApp and email with clients in one place. Must clear it with her partner next week. Send proposal by Tuesday."

After:

Three habits make the summaries reliable. Label speakers in transcripts ("Rep:", "Client:"), because attribution of action items depends on it. Put numbers and dates on their own line in your notes. And read the result before you paste it into the record, because a model fills gaps when it is not told otherwise.

If a meeting tool already records your calls, its own summary is usually generic. Paste its transcript into the prompt above to get the fields your CRM uses.

Customer data: which ChatGPT plan to use

Most sales prompts contain personal data: names, roles, what someone said on a call. What happens to it depends on the plan.

OpenAI offers the DPA for ChatGPT Business, ChatGPT Enterprise and the API. Its consumer terms for Europe include a separate addendum for business use, so using ChatGPT commercially is allowed; the question for a company is which terms and which data you put into it.

Under GDPR, that leaves a short checklist for a sales team:

• Use a business plan with a signed DPA for anything that contains client data.
• Paste the conversation, not the whole client record. Leave out phone numbers and addresses the task does not need.
• Tell your team which plan and which Projects are approved, and which are not.
• Mention AI processing in your privacy notice if you process prospects' data this way.

This is a working checklist, not legal advice. Your data protection officer decides the details.

ChatGPT alone or a dedicated sales tool?

ChatGPT does not replace a system that stores the deal history. Its memory belongs to one user, and a chat history is not something a colleague can pick up when you are on holiday.

Connecting ChatGPT to a CRM

There are three ways, from simplest to most direct.

1. Copy and paste. Works on every plan. Slow, but nothing changes in the CRM unless you change it.
2. An automation tool. n8n, Zapier or Make send a transcript or a new record to OpenAI's API and write the result back. You pay for the automation tool and the API separately, and you maintain the workflow. The n8n beginner guide shows the basics.
3. An MCP app. ChatGPT calls the CRM's tools directly: search contacts, update a deal, add a note.

For the third route, OpenAI's help center (checked 8 October 2026) says:

• Full MCP support, including write actions, is rolling out in beta to ChatGPT Business, Enterprise and Edu.
• It runs on ChatGPT web only, not on mobile.
• On Business, only admins and owners can turn on developer mode and create an app, under Workspace settings › Apps › Create. Enterprise admins can grant it to selected members.
• Pro users can connect MCP servers with read and fetch permissions in developer mode, without write actions.
• ChatGPT may ask for confirmation before write actions, depending on the app's permissions and the action.

So a small team that wants ChatGPT to update the CRM needs ChatGPT Business, and an admin who adds the app once. Claude's connector rules differ; the Claude CRM guide covers them.

Two risks matter once it works. A connected AI acts with the permissions of the person who connected it. And if your CRM can send email or chat messages, a send call sends. Ask for drafts and send them yourself.

Where Customermates fits

Disclosure: Customermates publishes this guide. Customermates is an open-source, AI-native CRM built in Germany for small client-facing teams such as consultants and agencies. It works with ChatGPT in two ways.

Your ChatGPT, through MCP. The MCP endpoint is part of every plan and of the self-hosted core. In ChatGPT you add an app with the URL of your Customermates address followed by /api/v1/mcp, choose OAuth, sign in and click Approve on the Authorize access screen. The connector guide has the steps and an API-key fallback. ChatGPT then has 53 tools, among them tools for contacts, organizations, deals, services, tasks and your own lists, their notes and fields. The model runs on your ChatGPT plan, so this uses no Customermates AI credits. Customermates does not stop a tool call to ask for confirmation, so set the app's permissions in ChatGPT and ask for drafts rather than sends.

Prompts to test the connection on a new workspace, whose default deal stages are Prospecting, Qualification, Demo, Proposal, Negotiation, Won and Lost:

1. "List my open deals in the Proposal stage with organization, value and assigned users." (read)
2. "Add this call summary to the notes of the Roth Steuerberatung deal and set Stage to Proposal: [paste]." (update)
3. "Create a task assigned to me to send the proposal, linked to the Roth Steuerberatung deal." (create and link)

Mate, in Customermates Cloud. If your team has no ChatGPT Business plan, Mate is the assistant built into Customermates Cloud. It uses the monthly hosted AI credits per active user that come with each plan (see pricing), prepares summaries and reply drafts, and runs Routines: saved instructions on a schedule or on record events. An example routine, on the schedule Every week, Monday at 08:00:

List my open deals in the stages Proposal and Negotiation. For each one, give
the organization, the value, the latest note in one line and the next step.
Flag deals without a next step. Do not send messages and do not change records.

A routine runs as its owner and may send messages when its instructions ask for it, so the last sentence matters. Run limits are on the routines page. On Pro and higher, the unified inbox puts email, LinkedIn, WhatsApp, Instagram and Telegram conversations next to the records, where Mate or your connected ChatGPT can read a thread and save a reply draft.

What it costs. Five people on Pro pay €145 per month, billed monthly, which includes the inbox and Mate's hosted credits. If you connect ChatGPT instead, add the ChatGPT Business seats from OpenAI's pricing page. Starter at €12 per user covers records and MCP without the inbox.

Not a fit if you need email sequences, a phone dialer or a native mobile app. A sales engagement tool covers sequences; ChatGPT can still write the drafts for it.

Frequently asked questions

OpenAI details checked on 8 October 2026 against Developer mode and MCP apps in ChatGPT, Connected apps in ChatGPT, Data controls FAQ, Enterprise privacy, ChatGPT pricing and the EU Terms of Use.