AI-first GTM: How to find buyers and message them automatically
Use Claude or ChatGPT to pick your audience, find people to talk to, and write messages you’d actually send.
You can do a lot of this in Claude or ChatGPT already. Give it some context about your business, work out who you should talk to, then write a message that starts a conversation.
This works if you’re new to a team or already working in growth or sales. BDRs can use it to build a prospect list. AEs can use it to research accounts before reaching out.
Start small. Ten profiles are enough to notice when your targeting is off.
1. Give your AI some context
There’s a lot of ways to make a target audience list, but honestly you shouldn’t overthink it at this point and just get started with something.
Open Claude or ChatGPT. Paste your website and explain who gets the most value from your product. If you’ve just joined the team, use a customer case study or ask a teammate which customers they’d want more of.
A URL helps, but your AI might not be able to read the page. Paste the relevant text when it asks.
Pick your first audience
Help me pick one audience for my outreach. My website: [URL] What we sell: [one sentence] Our best customers: [who they are and why they use us] The problem we solve: [in our customers’ words] Where we can sell: [countries or regions] Read the website if you can access it. Ask me to paste the relevant text if you can’t. Suggest one audience to start with. Return a table of job titles, company type, company size, and location filters. Separate facts I gave you from assumptions. Tell me who to leave out and what to check on the first ten profiles. Ask about any missing context before making the list broader.
Say you sell onboarding software to SaaS customer success teams. Here’s how that conversation could look. The examples below use fictional companies and sample data.
| Filter | Start with | What to check |
|---|---|---|
| Job title | Head or VP of Customer Success | Do they own onboarding? |
| Company type | B2B SaaS | Does the product need a customer onboarding process? |
| Company size | 51–200 employees | This is a starting assumption, not proof of fit. |
| CS team size | 5–20 people | Check the team rather than guessing from company size. |
LinkedIn Sales Nav is pretty good for turning this into a list because it has many fine-grained filters. Job title and seniority are helpful, although they’re not 100% accurate. Check whether people actually post or use LinkedIn before making it your main way to reach them.
2. Check who actually fits
Give your AI a small list to review. You can paste your research notes or upload a CSV with company names and profile links.
Ask it to explain each decision. A person having the right job title doesn’t mean they need your product.
Review your first ten prospects
Review this list against the audience we chose. [Paste company names, profile URLs, and research notes, or upload a CSV.] Return a table with: person/company, fit, evidence, source URL, and next step. Use “good fit”, “not a fit”, or “needs research”. Cite the source for each factual claim. If the information came from my notes, say so. Don’t invent people, profile links, team sizes, or problems. Mark missing facts as unknown. Flag duplicates, existing customers, and opt-outs when that information is in the data I supplied.
| Sample account | Fit | Evidence from the supplied notes | Next step |
|---|---|---|---|
| Northstar Demo | Good fit | 8 people in CS; onboarding runs in spreadsheets | Check the CS lead’s profile |
| Harbor Demo | Not a fit | Marketing agency | Leave out |
| Maple Demo | Needs research | B2B SaaS; CS team size and process unknown | Check the missing details |
Click through the sources before using the list. With web search available, your AI can help research public pages. It still needs to show where the information came from.
3. Write a message you’d actually send
A lot of people overcomplicate this and write long, AI-generated messages. No one is going to read those, so save your tokens.
One line about you, then one question asking if they’re facing the problem you help solve. If they agree, you have a reason to continue the conversation.
You don’t need an impressive background to ask a useful question. Say what you work on. Use a customer result only when it’s real and relevant.
Write your first message
Help me write a short first LinkedIn message for the audience we chose. What I do: [one plain sentence] A real reason to trust me, if relevant: [experience or a result I can back up] The problem I help with: [in my customer’s words] How I normally write: [paste a message you’ve sent] Write two versions, each under 45 words. One line about me, then one question asking whether they have this problem. Don’t assume they have the problem. Don’t add flattery, invented results, a meeting link, or a long product pitch. Return a table with the version and full message. Show me both; don’t send anything.
| Version | Example message |
|---|---|
| A | I work on onboarding software for SaaS teams. Are you still using spreadsheets to track new customer onboarding? |
| B | How does your CS team keep track of new customer onboarding? We make software for this, and I’m curious how you’re handling it today. |
Don’t paste a whole AI paragraph just because it sounds polished. Edit it like you’d edit a message to someone you know.
4. Connect PumpGTM to find people and send follow-ups
So far, you’ve used your AI to plan and write. You can take the list and send the messages yourself.
I built PumpGTM so I could also find people, manage outreach, and check results from my AI. It connects to Claude or ChatGPT and uses your connected accounts to send messages and follow-ups.
| What you want to do | With Claude or ChatGPT | With PumpGTM connected |
|---|---|---|
| Pick an audience | Work from your website and customer context | Use that context to prepare a people search |
| Find and review people | Research available sources or review a list you provide | Run discovery and review candidates in your workspace |
| Write messages | Draft and revise in chat | Use approved messages in an outreach sequence |
| Follow up and check results | Send separately and bring the results back | Send through connected accounts and read available outreach results |
Connect PumpGTM to Claude or ChatGPT, then try this in your chat:
Find people with PumpGTM
Use PumpGTM to prepare a people search for the audience we chose. Show me the roles, company filters, location, and exclusions before running discovery. Ask me to confirm the targeting. After I confirm, run discovery and show the candidates with the available evidence and profile links. Help me review the first ten. Don’t invent missing details. Don’t create a sequence, enroll anyone, or contact anyone yet.
Once you’re happy with the people and your first message, ask for a sequence preview:
Preview your outreach before approving it
Preview a LinkedIn sequence for the good-fit people I reviewed. Connect first. Send my chosen message one day after acceptance. Suggest one short follow-up three days later if they haven’t replied. Show the sender, audience, full messages, and timing in this chat. Stop later steps when someone replies. Flag any missing account connection or setting I need to check. Don’t call any tool that creates or saves a sequence, enrolls people, or sends messages. Wait for my explicit approval of these details.
Keep the preview in chat until you’re ready. Creating a sequence in PumpGTM can make it live; people enrolled in a live sequence can be contacted. The prompt above asks your AI to wait for your approval before creating or enrolling anything.
5. Post something useful for your audience
Talk about the way you think about the problem and what you’ve built to solve it. You don’t need a big announcement.
A question from a customer, a short demo, or something you learned this week is a good place to start. This helps the people you’ve connected with get to know how you think.
Turn this week’s work into a post
Help me turn something from this week into a LinkedIn post. Who I want to reach: [audience] What happened: [a customer conversation, product change, or something I learned] What I can show: [screenshot, demo, or a specific example] How I normally write: [paste one of your posts] Draft a short post explaining what happened and why it matters to this audience. Keep my wording where it works. Don’t add a dramatic hook, invented results, or a sales pitch at the end. Point out anything private I should remove before posting.
Make sure you have permission before sharing a customer’s message or screenshot. Remove anything private.
6. Work out what to repeat
Keep track of what you did and what happened. Read the replies as well as the numbers. “Not interested” counts as a reply, but it doesn’t mean the message is working.
You can paste your results into Claude or ChatGPT. With PumpGTM connected, ask your AI to read the available outreach results first.
Decide what to try next week
Help me decide what to repeat next week. [Paste the date range, audience, message versions, unique people messaged, replies, positive replies, meetings, and revenue.] If PumpGTM is connected, read the available results and state the date range. Ask me for any missing meeting or revenue data. Return a table by message version. Calculate reply and meeting rates using unique people messaged as the denominator. Show missing numbers as unknown, not zero. Read the replies I provide and distinguish interest from polite replies or rejections. Explain what the sample supports. Don’t call a small difference a proven winner. Suggest one thing to try next week and what to measure.
| Message | People messaged | Replies | Meetings |
|---|---|---|---|
| A | 40 | 6 (15%) | 2 (5%) |
| B | 40 | 3 (7.5%) | 1 (2.5%) |
A did better in this example, but 40 people per version is a small sample. Keep the audience similar while you collect more replies. Changing the audience and the message at once makes it harder to tell what helped.
Your checklist for this week
Start with one audience and one message. Tick these off as you go. Your progress saves in this browser.
Has LinkedIn been useful for you? Feel free to reply to me on X.
keep shipping,
Namanyay