# Personalized Outreach at Scale: Building a Claude Agent That Researches Before It Reaches > Generic cold email gets ignored. Specific cold email gets replies. Here is how to build a Claude agent that reads a prospect company website, recent content, and job postings before writing a cold email -- at scale. - **URL**: https://www.maximalstudio.in/blog/personalized-outreach-agent --- Maximal StudioApproachResourcesBlogToolsGet In Touch<- Back to BlogPersonalized Outreach at Scale: Building a Claude Agent That Researches Before It ReachesJun 29, 2026-Shubham RasalGeneric cold email gets ignored. Specific cold email gets replies. Here is how to build a Claude agent that reads a prospect company website, recent content, and job postings before writing a cold email -- at scale.The average cold email response rate is around 1-3%. The difference between 1% and 8% is almost always specificity -- emails that reference something real about the recipient's situation rather than merge-tag personalization ("Hi , I noticed you work at "). Real specificity requires research. Research takes time. At scale, the math doesn't work -- you can research 10 prospects thoroughly or send 200 generic emails. Agents break this tradeoff. Here's how to build one that does the research automatically and writes the email from it. What Real Personalization Looks Like Generic: "Hi Sarah, I see you're the Head of Marketing at Acme. We help marketing teams like yours automate their content." Real: "Hi Sarah, I read your post on how Acme is pushing into the enterprise segment -- the point about sales cycle length was sharp. We've helped two other companies making that same transition set up automated qualification workflows. Worth a quick call?" The second email took 3 minutes to write manually. With an agent, it takes 15 seconds. What the Agent Reads For each prospect, the agent fetches and analyzes: Company homepage -- products, positioning, language they use about themselves Blog / content -- recent posts reveal strategic priorities and vocabulary Job postings -- what they're hiring for signals where they're investing LinkedIn (founder/target's recent posts) -- what they personally care about publicly From these four sources, Claude identifies: the one thing this company is actively working on right now that your service is relevant to. The Research Prompt Research this company for outreach: [company URL] Fetch and analyze: 1. Homepage: what do they do, who do they serve, how do they describe themselves 2. Blog (if exists): what did they publish in the last 30 days, what topics 3. Jobs page: what roles are they hiring for right now 4. Find one specific detail -- a challenge they mentioned, a market they're entering, a feature they just launched -- that is genuinely relevant to [your service/offer] Output: - Company summary (2 sentences) - The specific hook: one thing that's true about them right now that connects to our offer - Do not invent details. If no specific hook exists, say "no strong hook found." The Email Writing Prompt Write a cold outreach email using this research: Company: [name] Contact: [name], [title] Hook: [output from research prompt] Our offer: [one sentence on what you do] Email requirements: - 3 sentences maximum - First sentence: the specific observation (the hook) - Second sentence: why that's relevant to what we do - Third sentence: one soft ask (reply, 15-min call, specific question) - No "I hope this finds you well", no generic opener - Tone: peer-to-peer, not vendor-to-prospect Subject line: [write one, 6 words max, no clickbait] The Scale Architecture Wire this into a list-based workflow: Input: CSV with [company_url, contact_name, contact_email, contact_title] For each row: 1. Research agent -> company hook 2. If hook found: email writing agent -> draft email 3. If no hook: flag as "skip -- no angle" 4. Append to output CSV: [contact_email, subject, body, hook_used] Output: ready-to-send CSV for your email tool (Instantly, Lemlist, etc.) A list of 100 prospects takes about 15 minutes to run. The output is 70-85 personalized emails (some will have no strong hook and get flagged) and 15-30 flagged skips. Quality Control Don't send everything the agent produces. Before running any list through your email tool: Spot-check 10 emails manually -- read them as the recipient would Check for anything that's technically true but sounds creepy ("I noticed you posted about...") Verify the hook is actually specific and not generic-sounding After a few iterations, you'll know which prompt produces good output for your offer and which sources reliably surface useful hooks. Response Rates in Practice Teams running this approach consistently see 5-12% reply rates on cold outreach, compared to 1-3% for template-based campaigns. The difference is entirely in the specificity of the hook -- generic hooks produce generic results even with a good email template. Want This Built for Your Outreach? We build personalized outreach agents for agencies, SaaS companies, and founders who need qualified meetings, not just sent emails. Let's talk about your pipeline.Keep exploringWork with usBuild your AI product ->We ship AI integrations, dev tools, and full products for teams.Free ToolsAI Calculators & Utilities ->ROI calculator, LLM cost estimator, workflow tools.Case StudiesReal-world AI builds ->See how we've shipped AI automation for real businesses.BlogMore posts ->Practical guides on AI, automation, and building fast.Maximal StudioAI & automation for builders.PagesToolsBlogCase StudiesApproachResourcesOfficeIndiaBangaluru, Karnataka, IndiaConnectLinkedInXEmail© 2026 Maximal Studio. All rights reserved. --- Maximal Studio builds custom AI systems, automations, and growth tooling.