How to use AI for sales prospecting

To use AI for sales prospecting, give an assistant such as ChatGPT or Claude your ideal customer profile. Connect it to a real B2B data source through an MCP server or API, and let it search companies, find the right roles, verify work emails and research each account. A person should review the list and every message before anything is sent.

Facts about other products checked on 2026-11-09 · sources listed below

What AI does well in prospecting, and what it does not

AI assistants such as ChatGPT and Claude are good at the parts of prospecting that are slow for people:

Turning a loose description of your customer into precise search filters.
Reading company websites, careers pages and job posts, then summarising what matters.
Running repetitive steps in order: search, enrich, verify, export.
Drafting a specific first line or angle for each prospect.

They are not good at facts they were never given. A language model on its own does not know who holds a role today or what that person's work email is. If you ask it for an address, it may produce one that looks plausible and does not exist. That is the main rule of AI prospecting: the model plans and reasons, a data source supplies the facts, and a person approves what goes out.

Step 1: define your ideal customer profile with the assistant

Start by having the assistant interview you. Ask it to question you about your best current customers: industry, headcount, region, the job titles that sign, the titles that use the product, and the events that tend to come before a purchase (a new sales leader, a funding round, hiring for a specific role).

Ask it to write the result as a short, structured profile with explicit filters, for example "US B2B SaaS, 50 to 200 employees, VP or Head of Sales, currently hiring SDRs". Vague profiles produce vague lists, so push it to name what is out of scope too. Our guide on how to create an ideal customer profile covers the inputs in more detail.

Step 2: connect the assistant to real B2B data

The practical way to give an assistant facts is a Model Context Protocol (MCP) server: a set of tools the assistant can call directly, such as "search people" or "verify email". See what an MCP server is if the idea is new. Several B2B data vendors now publish one. Based on each company's own pages, checked 2026-11-09:

Apollo says it offers an official remote MCP server and native connectors in Claude, ChatGPT and Perplexity on any plan, including Free, at no extra cost.
Lusha offers an official MCP server with setup guides for Claude, ChatGPT, n8n and Cursor; it uses the same credit balance as the rest of the product.
Wiza runs a remote MCP server at mcp.wiza.co that uses existing plan or API credits.
ZoomInfo has an MCP server that requires a ZoomInfo licence (GTM.ai self-serve or Enterprise).
Clay offers an MCP for Claude, ChatGPT and Codex; plan availability is not published.
Leadsets is a remote MCP server at https://mcp.leadsets.ai and a REST API, priced as a flat monthly plan from $29 with 1,000 free records to start and no credits.

Pick based on what you need. If you need phone numbers, Apollo, Lusha, Wiza and ZoomInfo all list them; Leadsets does not offer phone numbers yet. If you need SOC 2 today, note that Leadsets has it planned, not done. If you want one place for search, enrichment, email verification and company website scraping with a predictable monthly bill, Leadsets is built for that.

Step 3: build, enrich and verify the list

With data connected, a good agent works in a fixed order:

Find accounts that match the profile (industry, size, location, technology).
Add a timing signal, such as open roles on the careers page. A company hiring SDRs this month is often a better target than a perfect fit with no change underway.
Find people in the right function and seniority at those accounts, ideally with a reason each one matched.
Enrich and verify work emails. Treat "verified", "catch-all" and "unknown" as different results. A catch-all email cannot be confirmed for a specific mailbox, so send to it at lower volume or hold it back.
Export to your sequencer, CRM or a sheet.

Ask the agent to report what it could not confirm rather than filling gaps. Free, open-source playbooks such as the build an outbound prospect list skill encode these steps so the assistant runs them the same way every time.

Step 4: research accounts and draft outreach

Once the list is clean, use the assistant for research it can check against a source: what the company sells, who it sells to, which roles it is hiring for. Ask for one true, specific observation per account and a short reason your product is relevant. Then have it draft messages using those notes, following the basics of how to write a cold email.

Keep a person in the loop here. Read a sample of every batch, check that each first line is accurate, and make sure your outreach follows the rules that apply to you; see is cold email legal.

Example: one prompt, start to finish

Maya Chen runs sales at Northwind Analytics (northwind.co), which sells inventory software to mid-market distributors. She connects a B2B data MCP server to Claude and writes:

"Find US logistics software companies with 50 to 200 employees that posted SDR roles in the last 30 days. For each, find the VP or Head of Sales with a verified work email. Put catch-alls on a separate tab."

The agent searches companies, reads careers pages to confirm the hiring signal at accounts such as Stackfold and Parcelry, finds the sales leaders, verifies each email and returns a sheet with a match reason per row. Maya reviews the catch-all tab, removes two accounts she already works with, and asks the assistant for one opening line per company based on its homepage. She edits the lines before loading them into her sequencer. (Illustrative example; names are fictional.)

Using ChatGPT for sales prospecting

Without any connector, ChatGPT is still useful for the thinking work: building your ICP, writing persona notes, brainstorming buying triggers, summarising pages you paste in and drafting message variations. Do not ask it to produce names or email addresses from memory.

With an MCP connector added, ChatGPT can call a data provider's tools the same way Claude can, and the workflow above applies. Common mistakes to avoid:

Sending to emails the model generated rather than ones a verifier returned.
Treating catch-all or unknown addresses as safe, which raises bounce rates.
Letting the agent send without review.
Skipping the timing signal and messaging everyone who fits on paper.

Questions and answers

Can ChatGPT find leads for sales?

On its own, ChatGPT can help define your ideal customer and research companies you name, but it does not have a live, verified contact database. To find real people and work emails, connect it to a B2B data provider through an MCP connector or API, then review the results before sending.

Can AI find someone's work email address?

Only if it is connected to a data source that returns and verifies emails. A language model asked directly may invent a plausible address that does not exist. Use a provider that labels results as verified, catch-all, unknown or invalid, and only send freely to verified ones.

What is an MCP server in sales prospecting?

An MCP server exposes tools an AI assistant can call, such as searching people, enriching a contact or verifying an email. Adding one to Claude, ChatGPT or Cursor lets the assistant build prospect lists from real data instead of guessing. Vendors including Apollo, Lusha, Wiza, ZoomInfo, Clay and Leadsets publish MCP servers.

Will AI replace SDRs in prospecting?

AI can take over list building, research and first drafts, which are a large share of an SDR's time. People are still needed to set the target, judge fit, review messages and handle conversations once prospects reply. Most teams use AI to let each person cover more accounts with better research.

How much does AI sales prospecting cost?

The assistant itself is usually a standard subscription, and the main cost is the data source behind it. Pricing varies by vendor: some charge credits per contact, some per seat, and Leadsets charges a flat monthly fee from $29 with 1,000 free records to start. Check whether phone numbers, verification and API or MCP access are included in the plan you choose.

Is it legal to use AI for cold outreach?

Using AI to find and research prospects does not change the rules that apply to the outreach itself. You still need to follow laws such as CAN-SPAM, GDPR or CASL where they apply, honour unsubscribes and keep a suppression list. Use data from providers that source it lawfully and offer an opt-out.

Sources · checked 2026-11-09

Prices and features change. If something here is out of date, email support@leadsets.ai and we'll fix it.

Go deeper