B2B prospecting: a 5-step process with AI prompts and live data

TL;DR

• B2B prospecting is finding the businesses and people most likely to buy from you, then researching them before the first contact. It runs in five steps: define your ideal customer, list matching accounts, find the buyers, check buying signals, and write a relevant first line.
• Most of that work is research, and research needs current data. Funding, hiring and job titles change every month, so a list built from memory or an old export goes stale fast.
• Firecrawl gives ChatGPT, Claude and any MCP-compatible AI tool that current data. Firecrawl /search and /scrape read the live web. Firecrawl Alexandria adds company and people data from providers such as Apollo, People Data Labs, FullEnrich and Fireround.
• AI sales prospecting goes further when an agent runs the whole job. An AI prospecting agent in Grok Bot, Claude Code or Codex uses the same Firecrawl tools to build the list, qualify it and draft openers while you do other work.

Most B2B prospecting guides explain the process and stop there. Below is the process in five steps, plus a prompt for each step that you can paste into ChatGPT, Claude or any AI tool connected to the Firecrawl MCP server.

B2B prospecting means finding companies that fit what you sell and the people inside them who own the decision. It also covers learning enough about each account to start a relevant conversation. It is outbound by nature. The seller picks the account and makes the first move instead of waiting for a form fill.

Prospecting and lead generation are often used as synonyms, but they start from opposite ends:

B2B prospecting follows five steps that most sales teams share, whatever tools they use. Each step narrows the list and adds context. By the last step you have a short list of the right people and a reason to talk to each.

1. Define your ideal customer profile (ICP). Describe the companies that get the most value from you: industry, size, stage, tech stack, location.
2. Build the account list. Find real companies that match the profile.
3. Find the buyers. Identify the people at each account who would own the purchase, and their work email.
4. Check buying signals. Look for funding, hiring, launches and leadership changes that make now a good time.
5. Write the first touch. Open with something specific about their situation that you found in steps 2 to 4.

ChatGPT and Claude can plan a prospecting process, score accounts against an ICP and draft outreach. On their own they can't build a reliable list, because they answer from training data. They don't know last week's funding rounds, current job openings or who holds a title today, and they will guess email addresses.

The fix is to connect a data source. The Firecrawl plugin for ChatGPT, the Firecrawl connector for Claude and the Firecrawl MCP server for other AI tools give the assistant three things:

• Firecrawl /search to find current pages, news and announcements on the web.
• Firecrawl /scrape to read a company's own site as clean Markdown or a short summary.
• Firecrawl Alexandria to query structured company, people and funding data from official data providers.

Firecrawl Alexandria is Firecrawl's library of data providers and specialized indexes that AI agents can call through one Firecrawl account. For B2B prospecting, it brings firmographics, funding rounds, people search, work emails and job postings into the same chat where you plan the outreach. Our guide to data enrichment for AI agents covers which source fits each company and people field. Finding a tool is free, and running one costs its listed Firecrawl credits.

These are the Firecrawl Alexandria tools that cover each prospecting job:

Prospecting is one use. Firecrawl Alexandria groups its providers into categories, and each one supplies a different kind of data:

Setting up takes a free Firecrawl account, a Firecrawl connection in your AI tool, and accepted terms for each data provider you want to use. ChatGPT and Claude have a one-click Firecrawl plugin and connector, and every other MCP client uses the Firecrawl MCP server. None of them needs an API key, because each one signs in to your Firecrawl account.

1. Install Firecrawl. In ChatGPT, open Plugins, search for Firecrawl and select Install. In Claude, add the Firecrawl connector from the connector directory, then turn it on from the + menu under Connectors.
2. Sign in and approve the connection. New users can create a free account at this step.
3. Accept provider terms. An admin on your Firecrawl team accepts the terms for Apollo, People Data Labs and FullEnrich in the dashboard's Data Sources settings.
4. Start a new chat and mention Firecrawl in your prompt (type @Firecrawl in ChatGPT).

Any AI agent that supports MCP can use the same Firecrawl tools, Firecrawl Alexandria included, through the Firecrawl MCP server. Add the server, then sign in to your Firecrawl account in the browser window that opens.

• Grok Bot: install Firecrawl from the plugin directory, the same listing as the Cursor Marketplace. You can also message your bot: Add this MCP server: https://mcp.firecrawl.dev/v2/mcp-oauth.
• Claude Code: run claude mcp add --transport http firecrawl https://mcp.firecrawl.dev/v2/mcp-oauth, then finish signing in from /mcp.
• Codex: install the Firecrawl plugin in the Codex app, or run codex mcp add firecrawl --url https://mcp.firecrawl.dev/v2/mcp-oauth in the Codex CLI, then codex mcp login firecrawl.
• Cursor or any client with a JSON config: add the server below to your MCP settings.

{
  "mcpServers": {
    "firecrawl": {
      "url": "https://mcp.firecrawl.dev/v2/mcp-oauth"
    }
  }
}

Grok Bot builders already pair the two. One, mr fundman on X, connected his Grok bots to Firecrawl for web data and to mem0 for memory.

The prompts below work the same way in these tools. Ask the assistant to use Firecrawl instead of typing @Firecrawl. The Firecrawl MCP server docs cover API key setup and more clients.

Firecrawl Alexandria isn't part of Firecrawl's keyless tier, so data provider calls need a signed-in account. The free plan includes 1,000 Firecrawl credits a month.

With Firecrawl connected, ChatGPT, Claude or your MCP client can run each prospecting step from one prompt, and each prompt builds on the chat before it. Replace the bracketed parts with your own details. You don't have to name the providers, because the assistant picks the Firecrawl Alexandria tools that fit the data you ask for.

Your best customers already describe your ICP. Firecrawl /scrape reads their websites, so the assistant can find what they have in common instead of you guessing.

@Firecrawl read the websites of my five best customers: [domain1], [domain2],
[domain3], [domain4], [domain5]. For each, tell me what they sell, who they
sell to, their rough size and any tech they mention. Then write an ideal
customer profile from what they have in common, and list the traits I should
filter on.

For headcount and funding on the same five companies, ask it to add People Data Labs company records.

An account list is the set of real companies that match your profile. Fireround finds startups by sector, stage and date of their last round. Apollo and People Data Labs search companies by attributes such as industry, size and location, which covers established businesses too. In Firecrawl's Company Enrichment Benchmark of 200 US companies, Apollo scored 89.3% on headquarters, employee count, founding year and LinkedIn page, and People Data Labs scored 80.6%.

@Firecrawl using the ICP above, find 20 companies that match it. Use
Fireround for [sector] companies that raised a seed or Series A in the last
60 days, and Apollo company search for [industry] companies with
[employee range] employees in [region]. Remove duplicates and give me a table
with company, domain, size, last round and date.

Ask the assistant to show the Firecrawl credit cost before it runs a paid search, and to stop at the page size you need.

The same idea works for qualifying accounts you already have. Andrew Allsop, a B2B SaaS go-to-market practitioner, built a Claude workflow with Firecrawl, Cloudflare and the Companies House API that checks whether new signups are single sites, multi-site groups or consultants, a job that used to take someone 30 minutes a day. He shared the setup on X.

The buyer is the person who owns the problem you solve, usually a founder at a startup and a department head at a larger company. Apollo's people search shows who holds a title at a company for free. FullEnrich then finds a work email, and you only pay for the people you shortlist.

@Firecrawl for each company in the table, run Apollo's free people search for
[titles, for example founder, CTO, head of engineering]. Show me the previews
without revealing anyone. Then find work emails with FullEnrich for the one
most senior match at the top 10 accounts only, and tell me the total Firecrawl
credits before you run it.

These people tools are part of Directory, a collection in Firecrawl Alexandria with more than 240 million professional profiles, including employment histories and work contact information. It launched for recruiting, and the same profiles help you find buyers.

Every provider that can find people is listed in the Firecrawl Alexandria people category. To look people up in Claude specifically, our guide to Claude for people search covers more search types.

Buying signals tell you which accounts to contact first. Firecrawl /search finds recent news and announcements, Firecrawl /scrape reads the company's careers or product pages, and Apollo returns current job postings.

@Firecrawl for the top 10 accounts, find what changed in the last 90 days:
funding, new executives, product launches, and roles they're hiring for. Use
web search for news, read each company's careers page, and use Apollo job
postings where available. Rank the accounts by how strong the signal is and
cite a source link for each one.

Signals matter because buyers arrive informed. Gartner surveyed 645 B2B buyers and published the results in May 2026. Buyers used an average of seven information sources, and 69% preferred to validate AI-generated insights with a sales rep. A rep who knows the account's latest news is useful to that buyer, and one who hasn't done the research adds little.

The first line should be about the prospect's situation, built from a signal you can cite. Belkins analyzed 7.5 million cold emails sent in 2025 and reported an average reply rate of 0.45%, so a generic opener rarely earns a reply.

For each of the top 10 accounts, write a two-sentence opener for [name or
role]. Start with the signal you found and link it to the problem
[your product] solves. No flattery and no product pitch in the first
sentence. Put the source link under each opener so I can check it.

Read every opener and source before you send anything. The assistant drafts, and you decide.

Firecrawl prices each step by the tool it calls, and every Firecrawl Alexandria tool shows its price in Firecrawl credits before it runs. Finding tools and reading their inputs is free. The cheap steps come first in the process above, so you pay for contact data only on accounts that already passed your filters.

As of November 2, 2026, Apollo people search costs 0 Firecrawl credits and a Fireround funding search costs 5. A FullEnrich work email costs 30 per email found. The Firecrawl pricing page lists the plans beyond the free 1,000 Firecrawl credits a month.

An AI prospecting agent can run all five steps as one job. It researches your product, builds the list, qualifies it and drafts the openers, then picks up again when you ask for the next batch. Grok Bot works on its own cloud computer, so the job keeps running after you close the app. Claude Code and Codex run the same job from your terminal and save the results to a file. All three use Firecrawl for the data, so every company, person and signal comes from the live web and Firecrawl Alexandria instead of the model's memory.

The fastest-growing sales teams are already moving this way. McKinsey's 2026 B2B Pulse Survey of nearly 4,000 buyers and sellers in 13 countries found that high-growth B2B companies were three times as likely to raise AI investment by double digits in 2026 (71% versus 25%). McKinsey also reports that financial services companies that rebuilt prospecting and relationship management around AI agents earned 3 to 15% more revenue per relationship manager.

Sales teams are already working this way. Krista, from SpaceXAI's go-to-market team, recorded a 53-minute webinar walking through the Grok Bots she set up for her own sales work. Each bot has one job and keeps working while she's offline. She still uses many of the same sales tools, "but I'm just not touching the UI anymore."

Her prospecting bot is the closest match to this guide. She gives it five accounts and five people, and it runs overnight:

• It checks intent signals, such as AI mentions in job posts.
• It pulls personal hooks from public posts and podcasts.
• It logs everything to a Google Sheet.
• It drafts outreach in Gmail for her to approve.

Her other bots handle her inbox, call slides, Salesforce updates, travel and customer research. Her advice is to onboard each bot like a new colleague, teach it a task by showing it once, keep scheduled routines to a minimum, and review every draft before anything goes out.

Each of those prospecting jobs needs current data, and Firecrawl gives a bot like hers a source for every one:

A prospecting agent needs five things from you before it starts:

1. A role and limits. Say what the job is, which tools hold the facts, and what the agent must never do, such as invent a contact.
2. Your website. The agent reads it with Firecrawl /scrape to learn what you sell before it looks for buyers.
3. One ICP to start with. A single segment first tells you which audience produced the replies.
4. A definition of a qualified prospect. Sorting accounts into high, possible and poor fit keeps weak matches out of your outreach.
5. A finish line. Give a number of accounts and the file or format you want back.

This brief covers all five. Paste it into a new Grok Bot, Claude Code or Codex session with Firecrawl connected:

You are my B2B prospecting agent. Use Firecrawl for every fact about a
company or a person: Firecrawl /search and /scrape for the web, and
Firecrawl Alexandria for funding, company and people data. Never invent
a company, a person or an email address.
 
1. Read [your website] and tell me what we sell, who it helps and the
   problem it solves.
2. Suggest a primary ICP, a secondary ICP and who to exclude. Wait for
   me to confirm before you build anything.
3. Find 25 accounts that match the primary ICP. For each one, find the
   most likely buyer with Apollo's free people search and one buying
   signal from the last 90 days, with a source link.
4. Sort the accounts into high fit, possible fit and poor fit, with one
   line on why.
5. For high-fit accounts only, find a work email with FullEnrich and
   write a two-sentence opener based on the signal.
6. Save the results to prospects.csv. Before any step that costs more
   than 100 Firecrawl credits, tell me the cost and wait for my go-ahead.

Keep working in the same session once the first list is back. The agent already knows your product, your ICP and what it found, so short follow-ups are enough:

• "Find 25 more accounts, but only companies with 50 to 200 employees."
• "Founders replied more than sales leaders. Put founders first in the next batch."
• "Check last week's accounts for new funding, hiring or leadership news and flag the ones that changed."

Developers can build the same workflow into their own product with Firecrawl's SDKs, the REST API, the Firecrawl CLI and no-code tools such as n8n, Zapier and Make. The Firecrawl Alexandria docs cover tool discovery and the request format.

Our own deployment team prospects with Firecrawl. When a rep wants to reach companies working on a specific problem, our internal system starts with the accounts we already know. Then it uses Firecrawl /search to find more companies that match and Firecrawl /scrape to read their sites and understand what each one does. Every email gets written from what the company publishes about itself, not from a bought contact list. The same setup pulls posts from developer forums where people describe the problems they're trying to solve, and it reaches structured sources like app store reviews and Q&A sites through Firecrawl Alexandria. Reps start each week with fresh, sourced reasons to reach out, and every claim links back to the page it came from.

Other teams run this kind of research at scale. 11x, an AI growth platform for revenue teams, builds company research briefs for outbound with Firecrawl. It has processed more than 11 million requests across 175,000 unique domains:

Firecrawl gathers and structures information from across the web to give us a detailed picture of any company, from its position in the market down to the tools in its tech stack. That depth of research is what makes our outbound so context-rich.

Jeson Patel, CTO at 11x (read the 11x story)

• How to use ChatGPT for lead generation walks through local business leads with Google Maps and work emails.
• How to use ChatGPT to generate B2B leads builds a lead list from Apollo and FullEnrich in one chat.
• People enrichment in Firecrawl Alexandria covers what each people data provider adds.
• Data enrichment for AI agents explains how agents fill company, people and financial fields with a source for each one.
• The best company enrichment APIs compares company data accuracy across providers.
• How to build an AI SDR turns this research into a Python agent that writes outreach.