How to Use Claude Code for Sales Prospecting

Most reps spend more time researching accounts than talking to them. Claude Code can flip that ratio.

This is a practical walkthrough, not a theory post. By the end you will have a prompt structure you can copy, a clear picture of what Claude Code can and cannot do for prospecting, and a path to automating the loop entirely.

What Claude Code Actually Does Here

Claude Code is Anthropic's agentic coding environment. You run it in a terminal, point it at files and URLs, and it can read, write, browse, and reason across all of it in one session.

For sales prospecting, that means you can hand it a CSV of target accounts, give it a set of instructions, and have it pull publicly available signals, compare those signals against your ICP criteria, score each account, and draft a personalized opener, all without switching tabs.

It is not a CRM integration out of the box. It does not have access to paid data enrichment services unless you wire those in. What it does well is structured reasoning over public information and your own context, faster and more consistently than doing it manually.

The Prospecting Workflow, Step by Step

Here is the pattern that works.

Step 1: Prepare your account list

Start with a CSV that has at minimum: company name, website, LinkedIn URL if you have it, and any CRM fields you already own (ARR, industry, headcount estimate, last touched date).

Name the file accounts.csv and drop it in your working directory.

Step 2: Define your ICP in a text file

Create icp.md. Write out your ICP criteria in plain language. Be specific. Vague criteria produce vague scores.

A good icp.md looks like this:

# Ideal Customer Profile

## Firmographics
- Industry: B2B SaaS, tech-enabled services, professional services
- Headcount: 50-500 employees
- Geography: North America, UK, Australia

## Signals that raise fit score
- Recent funding round (Series A or B in the last 18 months)
- Active hiring in sales or revenue operations
- Leadership change in VP Sales or CRO role in last 6 months
- Company mentions a specific pain: pipeline visibility, rep ramp time, meeting quality

## Signals that lower fit score
- Primarily e-commerce or consumer
- Fewer than 10 salespeople
- No outbound sales motion visible

The more honest you are here, the more useful the output.

Step 3: Write the master prompt

This is the part reps skip and then wonder why the output is generic. Paste this into your Claude Code session after loading your files:

You are a B2B sales researcher. I have loaded two files:
- accounts.csv: a list of target accounts with company name, website, and other CRM fields
- icp.md: my ideal customer profile criteria

For each account in the CSV, do the following in order:

1. RESEARCH: Visit the company website and LinkedIn page. Note what the company does, who they sell to, approximate headcount signals, and any recent news visible in public sources (press releases, blog posts, LinkedIn activity from leadership).

2. SIGNAL PULL: Identify any of the positive or negative fit signals listed in icp.md. Quote the source for each signal you find. If you cannot confirm a signal, say so explicitly rather than guessing.

3. SCORE: Give each account an ICP fit score from 1-10. Show your reasoning in two to three sentences. Do not give a score above 7 unless you found at least two confirmed positive signals.

4. DRAFT OPENER: Write a 3-sentence cold email opener for accounts scoring 6 or above. The opener must cite at least one specific signal you found. Do not use generic phrases like "I came across your company" or "I hope this finds you well." Lead with the signal.

Output everything in a structured markdown table followed by the drafted openers beneath it.

Start with the first 5 accounts and ask me to confirm before continuing.

That last line matters. Claude Code can hallucinate if left to run unsupervised across 200 rows. Running in batches of 5-10 and reviewing before continuing keeps the output honest.

What Good Output Looks Like

A scored row for a strong-fit account might read:

Acme Revenue Co | Score: 8/10 | Signals found: Job posting for VP of Sales (LinkedIn, confirmed), Series B announcement March 2026 (company blog), CEO posted about "rep productivity" challenges (LinkedIn, last week). Fit note: Three confirmed positive signals; headcount estimate 120 based on LinkedIn employee count; North America HQ confirmed.

The opener for that account:

Your Series B close and the VP of Sales hire you posted last month usually mean one thing: you are scaling meetings fast and ramp time becomes the bottleneck. We work with teams at that stage to reduce no-shows and route deals to the right reps from day one. Worth 20 minutes to see if it applies?

That opener is short, cites the actual signal, and makes a specific claim. It is not perfect, but it is a better first draft than most reps write from scratch.

Where This Breaks Down

Claude Code prospecting is good for structured reasoning over public signals. It is not a replacement for:

• Paid data enrichment (it cannot access ZoomInfo, Apollo, or LinkedIn Sales Navigator without API connections)
• Intent data or third-party behavioral signals
• Internal CRM history beyond what you feed it

It also requires supervision. Signals can be outdated. Websites do not always reflect current company state. Treat the output as a strong first pass, not a final source of truth.

Connecting Research to the Meeting

Research is only valuable if it converts to a booked meeting and then a closed deal. That is where most prospecting workflows lose the thread.

In one B2B sales case study measuring 2,420 meetings across five reps and 1,281 deals, the overall close rate was 52.9%, but the gap between the best and worst rep was nearly 30 percentage points. The same deal, routed to the wrong rep, had a dramatically different outcome. The no-show rate in that study averaged 28.1%, meaning more than one in four meetings never happened at all.

Good prospecting research raises reply rates. It does not fix what happens after the reply. That requires routing the right prospect to the right rep and making sure the meeting actually takes place.

You can read more about how those mechanics compound in our case study here.

Getting the Files

We put together a starter pack of the files referenced in this post: the icp.md template, the master prompt, and a sample accounts.csv structure. You can grab the free Claude Code prospecting files below and adapt them to your ICP.

FAQ

Do I need to know how to code to use Claude Code for sales prospecting?

No. Claude Code runs in a terminal and you interact with it in plain English. The setup takes a few minutes if you follow Anthropic's installation guide. The prompts in this post are copy-paste ready.

How accurate is the ICP scoring Claude Code produces?

Accuracy depends on two things: the quality of your icp.md and whether Claude Code can actually find confirming signals on public pages. When you require it to cite sources, the scoring is reasonably reliable for accounts with a visible web presence. For smaller or less active companies, you will often get honest "could not confirm" responses rather than fabricated signals.

Can Claude Code write to my CRM directly?

Not out of the box. You would need to either export the markdown output and import it manually, or build a light script that parses the output and pushes it via your CRM's API. That is a one-time setup if you run this workflow regularly.

Is this the same as using Claude in a browser?

Claude Code is different from claude.ai in a browser. It is a local agentic environment that can read files from your machine, run code, and browse URLs in sequence as part of a longer task. For prospecting workflows that span multiple accounts and files, Claude Code is significantly more capable than a standard chat session.

Better research gets more replies, smarter routing closes more of them, and protecting the meetings in between is what turns a good pipeline into a great quarter. More revenue. Same pipeline.