AI for Recruiters That Remembers Every Role

Updated January 2026 | 6 min read

You're recruiting for five different roles. A senior engineer for a fintech startup. A sales director for a SaaS company. A marketing manager for an e-commerce brand. A data analyst for a healthcare company. A product designer for a B2B platform.

You ask ChatGPT to write InMail outreach for the engineering role. It generates something generic about "exciting opportunities" and "cutting-edge technology." It doesn't mention the company's Series B funding. It doesn't mention they're building payment infrastructure. It doesn't mention the salary range. It sounds like every other recruiter message candidates ignore.

You're on message 47 today. You don't have time to rewrite AI's garbage.

Why Generic AI Fails Recruiters

ChatGPT can write a recruiter message template. It can't write your recruiter message because it doesn't know the role, the company, or what makes this opportunity different from the 10 other messages that candidate got this week.

When you ask AI to help with candidate outreach, it gives you the same boilerplate everyone else is using:

• "I came across your profile and was impressed by your background"
• "We're looking for a talented professional to join our growing team"
• "This is an exciting opportunity to make an impact"
• "Competitive salary and benefits package"

Candidates can spot this from a mile away. They don't respond because it's obvious you didn't write it and probably didn't even read their profile.

Generic AI doesn't know:

• What the client company actually does
• What makes this role different from similar roles
• What the real requirements are vs nice-to-haves
• What the comp range and benefits actually are
• What the client culture is like and who thrives there
• What the interview process looks like
• Why this role is open and what happened to the last person

You're juggling multiple searches for different clients. AI can't keep them straight. It mixes up details. It gives you engineering outreach that sounds like sales outreach. It uses the wrong company name.

What Recruiters Actually Need From AI

You need AI that already knows your active searches. Not AI that makes you re-explain every role every time you want to write a message.

That means storing:

• Client company profiles and what they actually do
• Role details including real requirements and deal-breakers
• Comp ranges and benefits packages
• Company culture and what kind of person fits
• Interview process and timeline
• Why the role is open and what success looks like
• Key selling points that differentiate this opportunity

When you're sending 30 outreach messages a day across six different searches, you can't afford to manually customize every AI-generated message because it doesn't know which role you're talking about.

How CLAUDE.md Fixes This

CLAUDE.md is a markdown file in your Obsidian vault where you document each active search once. Claude Code reads it every time you ask for help with that search.

You write something like this:

## Active Searches

### Client: PayFlow (Fintech Startup)
**Role:** Senior Backend Engineer
**Location:** San Francisco (hybrid 3 days/week)
**Comp:** $180K-$220K base + equity (0.1-0.25%)
**Stage:** Series B ($40M raised, growing 200% YoY)

**What They Do:**
Payment infrastructure for B2B SaaS companies. Think Stripe but built for recurring revenue businesses. Handling $500M+ in transaction volume.

**Tech Stack:**
- Go, Python, PostgreSQL, Redis
- AWS infrastructure
- Kafka for event streaming
- React for internal tools

**Real Requirements:**
- 5+ years backend development (must have)
- Experience with payment systems or high-transaction systems (must have)
- Go experience (nice-to-have, willing to train)
- B2B SaaS background helpful

**Culture:**
- Fast-paced startup (not a 9-5 job)
- Ownership-focused (they want builders, not ticket-takers)
- Small team (15 engineers, everyone ships code)
- Technical founders (CEO and CTO both engineers)

**Why Role Is Open:**
Scaling team after Series B. Not a backfill. New headcount to build fraud detection system.

**Selling Points:**
- Solving hard technical problems (payment infrastructure at scale)
- Equity that's actually worth something (strong growth metrics)
- Work directly with founders
- Impact: what you build affects millions in transaction volume

**Interview Process:**
1. 30-min recruiter screen (me)
2. 60-min technical screen with engineering manager
3. 4-hour onsite (2 technical, 1 systems design, 1 culture fit)
4. Offer usually within 48 hours

**Deal-Breakers:**
- Can't/won't work hybrid (they tried full remote, it didn't work)
- Not interested in startup risk
- Looking for stable 9-5 environment
      

Now when you ask Claude Code to "write InMail for a backend engineer who has payment processing experience," it knows the PayFlow role. It knows the tech stack. It knows the selling points. It generates outreach that's specific to this opportunity.

When you're switching between the PayFlow search and a different search, Claude knows which context to use. No mixing up details. No wrong company names.

Before and After CLAUDE.md

Before: You find a great candidate for the PayFlow role. You ask ChatGPT to write InMail. It generates: "I came across your impressive profile and wanted to reach out about an exciting Senior Backend Engineer opportunity at a fast-growing fintech company. They're looking for talented engineers to join their team and work on cutting-edge technology."

It's generic garbage. You rewrite it from scratch. 10 minutes wasted.

After: You ask Claude Code to write InMail for a backend engineer with payment processing experience. It reads your CLAUDE.md. It generates: "Saw you built the payment reconciliation system at [their company]. PayFlow is hiring a Senior Backend Engineer to build fraud detection for a payment platform processing $500M+ annually. Series B funded, technical founders, Go/Python stack. $180K-$220K + equity. Worth a conversation?"

Specific. Relevant. Human. You send it as-is.

Before: You're working on five searches. You ask AI to draft job descriptions for all of them. It gives you five nearly identical job descriptions with different titles. They all say "fast-paced environment" and "competitive compensation." They don't reflect what the roles actually are.

After: You ask Claude Code for job descriptions for your active searches. It reads your CLAUDE.md for each role. It generates five different descriptions that actually describe the companies, the roles, the requirements, and the selling points. Each one sounds different because they are different.

You post them with minimal editing.

Before: A candidate responds to your outreach and asks about the interview process. You dig through your email to find the client's hiring manager notes from three weeks ago. You write back 20 minutes later.

After: A candidate asks about the interview process. You ask Claude Code: "What's the PayFlow interview process?" It tells you immediately: recruiter screen, technical screen, 4-hour onsite, offer within 48 hours. You respond in 30 seconds.

What This Actually Looks Like in Practice

Monday morning. You're sourcing for three different roles. You find a strong backend engineer candidate on LinkedIn.

You open Claude Code: "Write InMail for PayFlow backend engineer role. Candidate has payment processing experience at Square, currently at a Series A startup."

Claude reads your CLAUDE.md. It knows PayFlow is Series B. It knows they want payment processing experience. It knows the comp range and tech stack.

It generates a message that mentions the candidate's Square experience, explains why PayFlow is the next logical step (Series B vs Series A, bigger scale challenges), lists the tech stack, includes comp range.

You send it. 2 minutes total.

You find another candidate. Different search. Sales director role for a SaaS company.

You ask Claude: "Write InMail for CloudMetrics VP Sales role. Candidate currently running West Coast sales at a competitor."

Claude switches context. It reads the CloudMetrics section of your CLAUDE.md. It generates completely different outreach focused on career progression, team size, comp structure.

No mixing up details. No reusing the engineering message template.

Wednesday afternoon. The PayFlow hiring manager asks you to schedule interviews for two candidates. You need to send confirmation emails with interview details.

You ask Claude: "Draft interview confirmation email for PayFlow candidates. Technical screen with Sarah, next Tuesday at 2pm."

It knows the interview process. It knows what to expect. It generates clear emails explaining the format, duration, and what to prepare.

You send them.

The $997 Investment That Makes Outreach Actually Work

Every recruiter message you send needs to be specific, relevant, and human. Generic AI can't do that. You end up rewriting everything or sending boilerplate that doesn't work.

If you send 30 messages a day and AI wastes 5 minutes per message because you have to fix its generic output, you're losing 2.5 hours a day. That's 12+ hours a week.

CLAUDE.md fixes that. You document each search once. AI remembers it. Every message it generates is specific to the role, the company, and the candidate.

We build your CLAUDE.md structure. We work with you to document your active searches. We set up Claude Code and Obsidian. We train you on maintaining it as searches change.

You get a working system that makes AI useful for recruiting, not another tool that creates more work.