Kno2gether kno2gether.com ↗ Get the Clay outreach prompt + setup →
Playbook

Get Paid: Let Claude + Clay Find Your Clients and Write the Outreach

Part 5 of 5, Claude on the Payroll. Every skill in this series is something people pay for. The bottleneck was never the work — it was finding the right people and writing each of them personally, hundreds of times. This is the copy-paste system: connect Clay for verified leads, hand Claude one prompt that spins up a researcher agent per creator, drafts a real 3-sentence DM, and a checker verifies it — while you sleep. You wake up to a lead list you only have to review and send.

See how Knotie does white-label AI
01

Why outreach is the bottleneckThe part everyone gets stuck on

By now in this series you can make Claude build tools, save money on bookkeeping, produce ad creatives, and run deep research. Every one of those is a skill someone will pay you for. So the real question stops being 'can Claude do the work?' and becomes 'how do I get paid?' To get paid you need an offer and you need clients. Finding clients is where almost everyone stalls — not because the work is hard, but because the sourcing is. To sell a service you have to find the right people, research each one individually (what they make, what they're missing, where you could actually help), prove you understand them with a message that isn't a generic template, and then do that hundreds of times. Done by hand it's soul-crushing, so it never gets done. This guide hands that entire job to two tools working together: Clay finds and verifies the people; Claude researches each one and drafts the message. You stay the human who reviews and sends.

Attribution: this workflow is rebuilt from a walkthrough by creator Samin Yasar (YouTube, '8 Insane Claude Fable Use Cases'). Demo specifics shown in that video are his; figures and screen results there are illustrative of what's possible, not a guaranteed result for you.
02

An offer = who + what + proofStep 1 — Define the offer before you touch a tool

The automation is worthless without a crisp offer, because the whole point is to find people who match it. Don't open Clay until you can write your offer in one sentence. The cleanest offers in this series reuse a skill you already proved: help creators make better videos, build their data visualizations, edit their footage, or generate their ad creatives — the exact things earlier parts taught you to do. Pick ONE outcome and one audience. Specific and checkable always beats 'AI marketing help'.

  1. Name the audience precisely: e.g. 'YouTube creators in my niche with 10k–200k subscribers'. Tight ranges make the lead search returnable.
  2. Name the single outcome you sell: 'I edit your long-form into 3 short clips a week' or 'I turn your raw footage into a captioned explainer' or 'I build the data visualizations for your videos'.
  3. Name your proof: the skill from Parts 1–4 you can actually deliver. Your message will be only as strong as the work behind it — don't claim what you can't ship.
  4. Write the gap you fix. For a creator that might be 'great content, no thumbnails / no shorts / flat data slides'. The agent will look for exactly this gap in each lead's real work.
The honest order is: build the skill first (Parts 1–4), THEN sell it. Outreach that promises work you can't deliver burns the lead and your name. Good skills back the claim.
03

Verified leads, two clicksStep 2 — Connect Clay to Claude

Clay is a sales-data platform that finds people and returns their verified emails, social profiles, and company info — pulling from a large third-party data network (Clay's own marketing cites 150+ providers). The big advantage of pairing it with Claude: Anthropic and Clay built an official connector, so the data is already verified — you don't scrape anything or guess whether an email is real. Connecting is genuinely two steps. Note the honest scope below.

  1. In Claude, open the Connectors page (the + under the message box → Connectors, or Settings → Connectors). Find Clay in the connector directory.
  2. Click Connect → choose your Clay workspace → Authorize. If you don't have a Clay account, one is created automatically and you get starter credits.
  3. Back in the chat, confirm Clay's tools are enabled (click into the connector menu and check it's toggled on / 'good to go') before you prompt — a missing toggle is the #1 reason the prompt 'can't find any contacts'.
  4. That's it. No API key, no scraper, no spreadsheet. Clay now hands Claude verified contacts on demand.
Honest scope: the Clay connector is READ-FOCUSED — search people, research accounts, pull verified contact details. Bulk enrichment jobs (enriching columns across thousands of rows) still run inside Clay's own UI, not through the connector. So scope your offer as 'researched leads with drafted outreach,' not 'a full enrichment pipeline.'
04

Where the endurance mattersStep 3 — The prompt: one agent that spins up an agent per lead

Here is the whole system in one prompt. It does four things: (1) uses Clay to find N matching creators, (2) spins up a researcher sub-agent per lead to study their recent work and pull their verified email, (3) has each researcher draft a 3-sentence DM — reference a real video, name the gap, offer one specific fix — and (4) runs a checker agent to verify every draft cites that person's real work, not a hallucination. This is the long, boring, hundred-task job: where older models would quit or start hallucinating after ~40 leads, a long-endurance model (Claude Fable 5 in the demo) is built to grind all the way through. Copy the block, swap the bracketed bits, and paste it into Claude with Clay connected.

  1. Set N small the first run (5–10 leads) so you can sanity-check quality before scaling to 50–100.
  2. Tighten the audience and the gap to YOUR offer — vague audience = vague leads.
  3. Run it and let it work. The longer it runs, the more it's doing the research you'd otherwise do by hand.
Why two layers of agents: the researcher per lead keeps each message grounded in one specific creator; the single checker enforces a consistent quality bar across all of them. That separation is what stops generic-template drift.
05

Swap the [brackets], paste into ClaudeThe copy-paste outreach prompt

Paste this into Claude with the Clay connector enabled. Everything in [brackets] is yours to edit.

  • Use Clay to find [10] [YouTube creators] in my niche with roughly [10k–200k subscribers] who make [educational tech videos]. For EACH person found, spin up a dedicated researcher agent that does the following independently:
  • 1) Go through their recent videos and pull what their content is actually about — their style, their best-performing topics, and one specific recent video by title.
  • 2) Get their verified email and main social handle from Clay (do not guess or fabricate — if Clay has no verified email, mark the lead 'no verified contact' and skip drafting).
  • 3) Draft a 3-sentence direct message that: references that ONE real recent video by name, names the specific gap I fix ([their data slides are flat / they have no short-form clips]), and offers ONE concrete fix I can deliver ([3 captioned shorts a week]). No generic flattery, no template language.
  • Then run a single checker agent over ALL drafts: verify each message references that person's REAL work (cross-check the video title and topic actually belong to them), flag any draft that cites a detail you can't confirm, and reject any message that reads like it could be sent to anyone.
  • Output a table: name | verified email | social | recent video referenced | the 3-sentence draft | checker verdict (pass/flag). Save it to a file called outreach-list.md. Do NOT send anything — leave every message as a draft for me to review.
The last line is load-bearing: 'Do NOT send anything.' Keep it in every version of this prompt. Drafted is not sent.
06

How to grade the outputWhat 'good' looks like

When the run finishes you should be able to scan the list and trust it. Use this as your acceptance checklist — if the output fails any of these, refine the prompt and re-run rather than sending.

  • Each row names a REAL, specific recent video by title — not 'your recent content' or a made-up title.
  • The gap named is true for that creator (flat data slides, no shorts, weak thumbnails) — spot-check 2–3 against their actual channel.
  • The fix is something YOU can actually deliver from a skill in this series.
  • The email is marked verified by Clay — and leads with no verified contact are skipped, not faked.
  • The DM is 3 sentences, specific, and would obviously not work if copy-pasted to anyone else. If a message would fit any creator, it failed.
  • The checker column flagged the weak ones honestly. If everything is 'pass', be suspicious and re-check manually.
A flagged or skipped lead is a feature, not a failure — it's the system telling you where it wasn't sure, which is exactly where a hallucinated message would have embarrassed you.
07

The line you do not crossThe guardrail: review before you send. Always.

This is the highest-value and highest-risk piece in the whole series. The agent finds real people and writes to them beautifully — which is exactly why it must never hit send on its own. One confidently-wrong line to a real prospect costs more than the entire list is worth. This is the same review-before-action discipline the series has flagged twice: in Part 2, where a computer-use agent that can cancel a subscription needs a permission gate before it clicks; and here, where outreach needs a human between draft and send.

  1. Have the agent write drafts to a FILE or sheet — never wire it to an email or DM API that actually sends. Drafted ≠ sent.
  2. Keep the checker agent step: a second pass that verifies each draft cites the person's real work, not an invented detail.
  3. YOU read every message before it goes out. Fix tone, kill anything that feels off, delete the flagged ones.
  4. You are the send button. Send them yourself, spaced out, like a human — not a blast.
  5. Respect each platform's rules: automated cold DMs are banned in many places. Personalized messages you wrote-with-help and send yourself are a different thing entirely.
This is the difference between a tool that makes you faster and a tool that torches your reputation. The automation does the hundred hours of research and drafting; the human keeps the send button. Never give that up.
08

From one run to a real serviceProductize it: the repeatable shape

Run this once and you have a lead list. Write down the shape and you have a service you can quote and repeat. Every sellable version follows the same four beats — the same productize pattern used across this series.

  1. Scope — one outcome, one audience, in a sentence ('25 researched [niche] creators with a personalized drafted DM each, referencing their real work').
  2. Connect — Clay for verified contacts; that's the data layer you don't have to build.
  3. Workflow — the copy-paste prompt above is your reusable recipe: find → research-per-lead → draft → check.
  4. Handoff — define the artifact: outreach-list.md (or a sheet). State explicitly what you deliver (researched, drafted leads) and what stays human (the review and the send).
Do this for one offer this week: pick the outcome, connect Clay, run the prompt on a small N, review, send a handful yourself. That's a productized client-getting system — not a someday plan.
09

All 5 parts, one machineWhere this sits in the series — and the full system

Claude on the Payroll is five jobs you can hand one capable, long-endurance agent. Each part stands alone, but they're designed to wire into a single repeatable machine: the agent builds the skill, saves you money running it, markets it, researches the market, and — this part — finds the clients and drafts the outreach to sell it. That loop is where this was heading all along.

  • Part 1 — Build: ship a working tool from a plain-English goal. guides.kno2gether.com/fable5-payroll-goal
  • Part 2 — Bookkeeper: a computer-use agent that reconciles spend and cancels waste — with a permission gate before it clicks. guides.kno2gether.com/fable5-payroll-bookkeeper
  • Part 3 — Ad-maker: generate and self-sort ad creatives into approved/rejected. guides.kno2gether.com/fable5-payroll-admaker
  • Part 4 — Red-team / research: deep, multi-source research rigour — a research method, NOT financial advice (any trading or money figures shown in the walkthrough are illustrative, not a recommendation). guides.kno2gether.com/fable5-payroll-redteam
  • Part 5 — Clients (this guide): Clay + Claude find the people and draft the outreach. guides.kno2gether.com/fable5-payroll-clients
Follow for the full system — all five parts wired into one machine that finds work, does the work, and gets paid for it. That's the one-person AI company this series has been building toward.

Get the next drop

New AI build guides + the occasional bonus template. No spam, unsubscribe anytime.

By submitting you agree to our Privacy Policy & Terms. Unsubscribe anytime.

Frequently asked questions

What does the Clay connector for Claude actually do?
It lets Claude search Clay's data network for real people and pull their verified contact details (email, socials), plus research target accounts — all from a natural-language prompt. It's read-focused: find, research, pull. Bulk enrichment across large datasets still runs inside Clay's own UI, not through the connector. Connecting is two steps: open Claude's Connectors page, click Connect, choose your workspace, and Authorize. New users get a Clay account and starter credits created automatically.
Why use a long-endurance model for this?
Outreach research is a hundreds-of-tasks job — find, research per lead, draft, verify, repeat. Older models tend to quit or start hallucinating after roughly 40 leads. A long-endurance model (Claude Fable 5 in the demo) is built to grind through long, repetitive jobs without degrading, which is exactly what a real lead list needs. The longer the job, the more this matters.
Will the messages actually be personalized, or just templated?
Personalized — if you prompt it right. The trick is spinning up a researcher agent per lead so each message is grounded in one specific creator's real recent video and a real gap, then running a checker agent over all drafts to flag anything generic or unverifiable. Grade the output with one test: if a message would also fit any other creator, it failed. Re-run rather than send weak drafts.
Is it safe to let the agent send the messages?
No — let it research and draft, never send. The agent writes drafts to a file; you review every one (verify it cites the person's real work, fix tone, drop the flagged ones) and you are the one who hits send. Automated cold DMs are banned on many platforms; personalized messages you review and send yourself are a different thing. Keep the human as the send button.
Is Part 4 telling me how to trade or invest?
No. Part 4 of this series is about research rigour — how to make an agent gather and cross-check multiple sources — not financial advice. Any trading percentages or money figures shown in the original walkthrough are illustrative of what the method can surface, not a recommendation or a guaranteed result. Do your own diligence; nothing here is investment advice.
Sources · Concept credit: the Claude Fable 5 + Clay outreach workflow demonstrated by Samin Yasar ('8 Insane Claude Fable Use Cases'). This guide is an original rebuild from the primary docs below, not a transcript; demo figures are attributed and not presented as guaranteed results. · Clay is now available as a connector in Claude — Clay blog (read-focused: account research, contact discovery, draft personalized outreach; new users get starter credits) · Clay MCP — prospect in ChatGPT and Claude (the '150+ data providers' figure is Clay's own marketing claim) · Get started with custom connectors using remote MCP — Claude Help Center · Introducing Claude Fable 5 — Anthropic (long-horizon / endurance capability) · Slash commands and workflows — Claude Code docs

You can draft the outreach. Selling the same service as a product is the part Knotie is built for.

This playbook gets Claude + Clay finding clients and drafting the messages — the hard sourcing problem, solved. The next step is turning the skill you sell into a product clients buy under your brand: AI voice and chat agents plus automations, with a customer portal, custom domain, and credit billing that meters their usage so you keep the margin — no code. That's exactly what Knotie does. You find the clients and productize the offer; Knotie handles the white-label plumbing so you can actually charge for it.

See how Knotie does white-label AI