# Account-Based Marketing with Claude AI (Step-by-Step) # Account-Based Marketing with Claude AI: The Complete Guide (2026) > **Quick answer:** Account-based marketing with Claude AI means giving an AI agent a list of target accounts and having it research every company, find every decision-maker, identify their specific pain points, write hyper-personalized outreach using their exact language, and send the connection request — without you switching tabs. GrowthSpree runs this as a **four-phase workflow in Claude Cowork** (Anthropic’s AI agent for non-developers): **1) mine contacts** from LinkedIn into a structured spreadsheet, **2) deep-research** each prospect, **3) draft personalized outreach**, and **4) send it** via Claude in Chrome. It compresses what used to take a BDR team days into a single session. > **TL;DR:** ABM with AI is no longer about plugging ChatGPT into a cold-email sequence. Using Claude Cowork’s browser automation and research capabilities, you can mine LinkedIn for decision-maker contacts across 15+ companies, build a structured spreadsheet, deep-research every prospect’s pain points and language, draft outreach that references what they actually care about, and have Claude physically send the LinkedIn connection request — the full cycle from account list to executed outreach in one agentic workflow. This guide breaks down all four phases as we run them at GrowthSpree, the setup you need (Claude Cowork on the Max plan plus the Claude in Chrome extension), where humans stay in the loop, and the mistakes that turn AI personalization back into spam. ## The workflow at a glance | Phase | What Claude does | Output | |---|---|---| | 1. Contact mining | Browses each target company’s LinkedIn People tab, finds relevant roles | Live spreadsheet of names, titles, profile URLs | | 2. Deep research | Researches each prospect — posts, priorities, pain points, exact language | Research notes per contact | | 3. Personalization | Drafts outreach referencing each prospect’s specifics | Personalized messages | | 4. Execution | Sends the connection request with note via Claude in Chrome | Sent outreach, logged | *Workflow as run at GrowthSpree with Claude Cowork; time compression versus a manual BDR process varies by list size and depth of research.* Traditional ABM execution dies in the gap between strategy and follow-through: the list gets built, but the research and personalization for dozens of contacts is so tedious that teams fall back to template merge fields. An agentic workflow removes that gap — the same session that finds the contact also researches them, writes to them, and reaches out. ## What you need Three pieces: [Claude Cowork](https://www.anthropic.com/news/claude-cowork) (Anthropic’s agentic workspace for non-developers, available on Claude’s Max plan), the **Claude in Chrome** extension (so Claude can physically browse and act on LinkedIn), and optionally the **xlsx and docx skills** for structured spreadsheet and document outputs. The broader AI infrastructure behind this approach is powered by [MCP servers](https://modelcontextprotocol.io/docs/getting-started/intro) — our [complete MCP guide](https://www.growthspreeofficial.com/blogs/mcp-servers-b2b-saas-marketing-complete-guide) covers every platform in the stack. ## Phase 1: Contact mining — from account list to spreadsheet Contact mining is systematically identifying and extracting decision-maker names, titles, and LinkedIn profile URLs from your target account list. It’s traditionally the most tedious step in ABM — and the one Claude automates most dramatically. 1. Start with a list of target company LinkedIn URLs — for example, 15 Series A/B SaaS companies in your ICP. 1. Give Claude a single prompt: visit each company page, open the People tab, search for marketing and growth roles, and add every match to a spreadsheet. 1. Claude uses Claude in Chrome to physically browse each page, click into People, search the relevant roles, scroll through results, and compile everything into a live .xlsx file. > **Key takeaway:** The output isn’t a screenshot or a pasted blob — it’s a structured, live spreadsheet you can filter, dedupe, and hand to the next phase. Garbage in still applies: the tighter your target-account list, the better everything downstream. ## Phase 2: Deep research on every prospect With contacts in the sheet, Claude researches each one: their posts and comments, what they’ve published, the initiatives they talk about, and the *exact language* they use for their problems. This is the step human teams skip at scale — and it’s where real personalization comes from. In our demonstration workflow, this phase surfaced a prospect’s specific attribution philosophy, which later became the hook of the outreach note. Depth matters more than breadth: one genuinely specific detail beats five generic ones. ## Phase 3: Hyper-personalized outreach Claude drafts outreach for each contact using the research — referencing their actual priorities in their own vocabulary, not “I loved your recent post.” The difference from template ABM is structural: merge fields personalize the greeting; research-based drafting personalizes the argument. Review the drafts — this is a natural human checkpoint to confirm ICP fit and tone before anything goes out. ## Phase 4: Execution — Claude sends it This is the phase that separates the workflow from every “AI writes your cold email” tool: Claude doesn’t just write the outreach — it **sends it**. Using Claude in Chrome, it physically sends the LinkedIn connection request with the personalized note. In GrowthSpree’s demonstration, the sent note referenced the prospect’s attribution philosophy — the detail extracted in Phase 2. The entire cycle from target account list to sent connection request happened inside Claude Cowork: research, personalization, and execution in a single agentic workflow. > **Key takeaway:** You’re not using AI to write outreach copy — you’re running an end-to-end ABM pipeline where research, personalization, and execution happen in one session, with you supervising rather than tab-switching. ## Where humans stay in the loop Agentic doesn’t mean unsupervised. The judgment calls stay human: - **The account list.** You define the ICP and pick the targets — the workflow scales whatever list you give it. - **Draft review.** Approve outreach before it’s sent, especially early, until the quality bar is proven. - **Volume and pacing.** Keep connection-request volume human-plausible and respect LinkedIn’s norms — the goal is better outreach, not more spam faster. - **The conversation.** When a prospect replies, a human takes over — the AI earned the meeting; you run it. This is the AI-native pattern — operators directing agents — that we cover in the [AI-native vs automation agency framework](https://www.growthspreeofficial.com/blogs/ai-automation-agency-vs-ai-native-marketing-agency-b2b-saas-b2b-2026-eight-differences). ## Scaling it up: from workflow to program This guide covers the hands-on workflow. To run it as a full program — signal-based account selection, intent scoring, multi-channel activation, and pipeline attribution — layer it into the architecture from our [ABM with AI agents execution blueprint](https://www.growthspreeofficial.com/blogs/account-based-marketing-ai-agents-execution-2026). For the drip-sequence extension of this exact workflow, see [ABM personalization with Claude + Cowork + Drips](https://www.growthspreeofficial.com/blogs/abm-personalization-claude-cowork-drips). And to target the same accounts with paid, pair it with [LinkedIn Ads MCP](https://www.growthspreeofficial.com/blogs/linkedin-ads-mcp-the-ai-powered-linkedin-ads-analytics-engine-for-b2b-saas). ## Common mistakes to avoid - **Skipping the research phase.** Outreach personalized only by name and company is template spam with extra steps. - **Blasting volume.** The workflow’s advantage is quality at scale — keep pacing human and respect platform norms. - **A loose account list.** AI scales whatever you feed it; a bad ICP list just produces personalized outreach to the wrong people. - **No human review early.** Approve drafts until the quality bar is consistently met. - **Stopping at the send.** Track replies through to pipeline so you learn which personalization angles convert. ## Frequently Asked Questions ### Q1. What is account-based marketing with Claude AI? Giving an AI agent a target-account list and having it research each company, find decision-makers, identify their pain points, write hyper-personalized outreach in their language, and send it — an end-to-end ABM workflow run in Claude Cowork rather than a copy-writing assist. ### Q2. What is Claude Cowork? Anthropic’s agentic AI workspace for non-developers, available on Claude’s Max plan. It can browse (via Claude in Chrome), research, produce real files like spreadsheets and documents, and execute multi-step workflows in one session. ### Q3. What are the four phases of the workflow? 1) Contact mining — Claude browses each target company’s LinkedIn People tab and compiles decision-makers into a spreadsheet; 2) deep research on every prospect; 3) hyper-personalized outreach drafting; 4) execution — sending the connection request via Claude in Chrome. ### Q4. What do I need to set it up? Claude Cowork (Max plan), the Claude in Chrome extension, and optionally the xlsx and docx skills for structured outputs. ### Q5. Does Claude actually send the outreach? Yes. Using Claude in Chrome, it physically sends the LinkedIn connection request with the personalized note — completing the cycle from account list to executed outreach in one session. ### Q6. How is this different from ChatGPT for cold email? A writing assistant drafts copy from what you paste in. This is an agentic pipeline: the same session finds the contacts, researches them, drafts from that research, and executes the send — no tab-switching or copy-paste. ### Q7. How many accounts can it handle? The demonstration workflow mined contacts across 15+ companies in one session, and the pattern scales with your list. For programmatic scale (hundreds of accounts with signal scoring), see our AI agents execution blueprint. ### Q8. Is this safe for my LinkedIn account? Keep volume and pacing human-plausible and respect LinkedIn’s norms. The workflow’s advantage is research-quality outreach, not volume — using it to blast connection requests recreates the spam problem it solves. ### Q9. Where do humans stay in the loop? Defining the ICP and account list, reviewing drafts before send (especially early), setting volume and pacing, and taking over the conversation when a prospect replies. ### Q10. What makes the personalization actually good? Phase 2. Claude researches each prospect’s posts, priorities, and exact language, so the outreach references what they genuinely care about — like a specific attribution philosophy — rather than a generic compliment. ### Q11. Can I extend this into a drip sequence? Yes — the same research and personalization feed multi-touch sequences. See our ABM personalization with Claude + Cowork + Drips guide for that extension. ### Q12. Do I need MCP servers for this? Not for the core workflow — Cowork plus Claude in Chrome covers it. MCP servers power the surrounding infrastructure (ad platforms, CRM, analytics) when you scale this into a full signal-based ABM program. ## Run your first session Pick 10–15 ICP accounts, open Claude Cowork, and run Phase 1 today — you’ll have a structured contact sheet within the session. Then layer in research and personalization before you send anything. For the program-level architecture around it, read the [ABM with AI agents blueprint](https://www.growthspreeofficial.com/blogs/account-based-marketing-ai-agents-execution-2026), or book a strategy call and we’ll show you the workflow live on your ICP. --- **About the author:** Ishan Manchanda is Co-Founder at GrowthSpree, a B2B SaaS marketing agency (Google Partner, HubSpot Solutions Partner, 4.9/5 on G2). GrowthSpree runs AI-native ABM — Claude Cowork workflows plus the QLA Signal Stack’s 15+ intent signals — across 300+ B2B SaaS accounts and $60M+ in managed spend, with senior operators supervising every send.