Prospect research and first-touch emails: Multi-Step Agent Workflow for Claude Code, Codex or ChatGPT

Researches up to 10 companies in parallel with cited sources, grades each against your ideal customer profile, writes a 3-email sequence for the A and B prospects with a sourced hook in every first email, checks facts, length and spam words, and hands you a review sheet.

Prompt

Installs to .agents/skills/prospect-research-and-first-touch-emails-multi-step-agent-workfl/

mkdir -p .agents/skills/prospect-research-and-first-touch-emails-multi-step-agent-workfl && curl -sL "https://prompts-gpt.com/prompts/prospect-research-and-first-touch-emails-multi-step-agent-workflow-for-claude-code-codex-or-chatgpt/raw" -o .agents/skills/prospect-research-and-first-touch-emails-multi-step-agent-workfl/SKILL.md
SKILL.md
Preview
---
name: prospect-research-and-first-touch-emails-multi-step-agent-workfl
description: "Researches up to 10 companies in parallel with cited sources, grades each against your ideal customer profile, writes a 3-email sequence for the A and B prospects with a sourced hook in every first email, checks facts, length and spam words, and hands you a review sheet."
---

# Prospect research and first-touch emails: Multi-Step Agent Workflow for Claude Code, Codex or ChatGPT

Researches up to 10 companies in parallel with cited sources, grades each against your ideal customer profile, writes a 3-email sequence for the A and B prospects with a sourced hook in every first email, checks facts, length and spam words, and hands you a review sheet.

## Prompt

```
---
name: prospect-research-outreach
description: Researches up to 10 companies in parallel with cited sources, grades each against your ideal customer profile, writes a 3-email sequence for the A and B prospects with a sourced hook in every first email, checks facts, length and spam words, and hands you a review sheet. Nothing is sent. Use when asked to run the "Prospect research and first-touch emails" workflow.
---

# Prospect research and first-touch emails

Researches up to 10 companies in parallel with cited sources, grades each against your ideal customer profile, writes a 3-email sequence for the A and B prospects with a sourced hook in every first email, checks facts, length and spam words, and hands you a review sheet. Nothing is sent.

## Inputs

Ask for any of these that the user hasn't given:

- `icp` (e.g. "B2B SaaS, 50-300 employees, selling to mid-market, with a self-serve trial and a support team of 5+")
- `offer` (e.g. "An AI support-triage add-on for Zendesk that cuts first-response time")
- `proof_points` (e.g. "A 120-person fintech support team cut first-response time from 9h to 2h in 6 weeks.
Used by 40 SaaS support teams.
Setup takes under a day.")
- `prospects` (e.g. "pipedrive.com
front.com
calendly.com
loom.com")
- `sender` (e.g. "Priya Shah, founder, Triagely")
- `cta` (e.g. "Open to a 15-minute call next week?")

## Steps

Run the steps in order. Refer to saved results by name. Don't skip a step that says STOP.

1. **Manual start** — Start: Paste up to 10 companies in `prospects`, one per line (name or URL). Describe your `icp`, `offer` and `proof_points` (only results you can back up), plus `sender` and the `cta`.
2. **Research each prospect** — Launch these as parallel subagents in one message, then wait for all of them. Research the companies in your slot of `prospects` (count non-blank lines from the top). For each company, search the web and read its site, recent news and careers page, then report:
- Company: name, URL, what it sells and to whom (one line)
- Size: headcount range; funding or revenue only if a source states it
- Recent: launches, funding, leadership hires or news from the last 6 months, each with its date
- Hiring: open roles that relate to `offer` (title and link)
- Likely pains: up to 3, each tied to a fact above and marked 'inferred'
- Best hook: the single specific, recent, verifiable fact a cold email could open with
Every fact needs its source URL and date. If two companies share the name or you can't confirm which one is meant, say so and stop for that company. If your slot has no company, reply only 'Empty slot.' Save the result as `research`.
   - **Prospects 1-2**: Take lines 1 and 2 of the list.
   - **Prospects 3-4**: Take lines 3 and 4 of the list.
   - **Prospects 5-6**: Take lines 5 and 6 of the list.
   - **Prospects 7-8**: Take lines 7 and 8 of the list.
   - **Prospects 9-10**: Take lines 9 and 10 of the list.
3. **Score ICP fit** — Score each company in `research` against `icp`. For each, rate four criteria as yes, partly or no, citing the research line you used: (1) matches the firmographics in the ICP, (2) evidence of the pain `offer` solves, (3) a timing signal in the last 6 months (launch, funding, relevant hiring, new leader), (4) a sourced hook exists.
Grade A: yes on 1 and 2, plus a timing signal. Grade B: yes on 1 and at least partly on 2. Grade C: everything else, including 'no' on 2 and companies where nothing verifiable was found. For each C, say whether it's 'C: poor fit' or 'C: unverified'.
Return a table: company, grade, the four results, a one-line reason. Don't raise a grade because a company is well known. Save the result as `scores`.
4. **Decide what to write** — Look at the grades in `scores` and pick a route. 'Write sequences' if at least one company is A or B. 'Nothing qualifies' if every company is C and most are 'C: poor fit'. 'Research too thin' if every company is C and most are 'C: unverified'. Pick one route: Save the result as `sequences`.
   - **Write sequences**: For each A and B company in `scores` (skip C), write a 3-email sequence from `sender`, using `research` and `offer`.
Email 1 (day 0): subject of 2-6 words, no clickbait. Body 120 words max: open with the best hook (a specific fact, not flattery), one sentence linking it to a likely pain, one sentence on the offer, one proof point, and this ask: `cta`.
Email 2 (day 3): a different angle or pain and a different proof point, 80 words max.
Email 3 (day 8): a short close-the-loop note, 60 words max, no guilt-tripping.
Under each email add two lines:
Evidence: the fact used about the company and its URL
Proof used: the proof point quoted word for word from `proof_points`
Rules: no claims about the company beyond the research, no invented results or customers, no fake familiarity ('loved your recent post'), no 'just following up' or 'quick question'. Start each company with a header: company name and grade.
   - **Nothing qualifies**: Explain in under 200 words why no company in `scores` fits `icp`, what the list has in common, and 3 concrete ways to find better prospects (company traits to filter on and places to look). Don't write emails.
   - **Research too thin**: For each company in `scores`, list what couldn't be verified and where to check by hand (company site, careers page, LinkedIn company page, press page, the exact legal name or domain to paste on a re-run). Don't write emails: a cold email without a verified hook is the generic kind this workflow exists to avoid.
5. **Fact and spam check** — Review loop (max 3 passes): If the draft contains no emails (a 'nothing qualifies' note or a manual-check list), pass it. Otherwise pass only if, for every email:
1. Every statement about the prospect matches its Evidence line, and the Evidence line has a URL.
2. The only result or customer claimed is the one in its 'Proof used' line.
3. Body word count: Email 1 at most 120, Email 2 at most 80, Email 3 at most 60.
4. No spam-style wording: free, guarantee, risk-free, act now, limited time, 100%, all-caps words, more than one exclamation mark, 'just following up', 'quick question', 'hope this finds you well'.
5. Subject line is 2-6 words with no 'Re:' or 'Fwd:' trick and no emoji.
6. One ask per email.
In feedback, name the company, the email number and the failing rule, and tell the reviser to keep the Evidence and Proof used lines. Save the result as `sequences`.
6. **Build the review sheet** — Turn `sequences` and `scores` into a review sheet. First a table: company, grade, hook (one line), evidence URL, Email 1 subject, status ('ready', or 'check' if any fact used is marked inferred or is older than 6 months). List C companies at the bottom with their reason. Then the full emails grouped by company, unchanged.
End with a 'Before you send' checklist: open every evidence URL; add your postal address and a one-line way to opt out (required for commercial email in the US under CAN-SPAM); if any prospect is in the EU or UK, check that you have a lawful basis for cold B2B email there (GDPR/PECR rules differ by country); send from a warmed-up domain with SPF, DKIM and DMARC set. Save the result as `outreach`.
7. **Check before sending** — STOP and ask the user to approve before continuing: Open every evidence link and work through the 'Before you send' checklist; this workflow doesn't send email. Approve to export the sheet; reject to stop.
8. **Save the review sheet** — Finish: Return `outreach`.

## Done when

Every step has run, each approval was given, and you've reported what was produced and where it was saved.
```

## Usage Notes

Run it in Claude Code or Codex: save as .claude/skills/prospect-research-outreach/SKILL.md (or .agents/skills/ for Codex) and say "run the Prospect research and first-touch emails workflow"; parallel steps run as subagents. Or paste it into ChatGPT or Claude and answer its questions. It asks for: icp, offer, proof_points, prospects, sender, cta. Build or run it step by step at prompts-gpt.com/workflows/prospect-research-outreach.

Example output

At a glanceProspect grades and hooks from the review sheetLedgerly and Shiftboard Cloud get emails; Parcelwise and Northdesk are graded C
ItemLedgerlyShiftboard Cloud
GradeAB
Hook3 Support Specialist roles, May 2Self-serve trial launch, Apr 18
Email 1 subjectThree new support hiresTrial launch and tickets
Proof usedFirst response 9h to 2h in 6 weeksUsed by 40 SaaS support teams
Statusreadycheck

Example for: Triagely (AI support-triage add-on for Zendesk), sender Mara Lindqvist, founder; 4 fictional prospects: Ledgerly, Shiftboard Cloud, Parcelwise, Northdesk.

Prospect Review Sheet: Triagely Outreach

CompanyGradeHookEvidence URLEmail 1 subjectStatus
LedgerlyAOpened 3 Support Specialist roles (posted May 2)ledgerly.example/careersThree new support hiresready
Shiftboard CloudBLaunched self-serve trial (blog, Apr 18)shiftboardcloud.example/blog/trialTrial launch and ticketscheck
ParcelwiseC: poor fit1,800 employees, enterprise-only sales
NorthdeskC: unverifiedTwo companies share the name; domain unconfirmed

Shiftboard Cloud is marked "check" because its pain (rising ticket volume after the trial launch) is inferred.


Ledgerly (Grade A)

Email 1 (day 0)
Subject: Three new support hires

Hi Dana,

Ledgerly posted three Support Specialist roles on May 2. When a team grows that quickly, triage usually turns into the bottleneck before headcount catches up. Triagely is an add-on for Zendesk that routes and prioritizes tickets automatically, which cuts first-response time. A 120-person fintech support team cut first-response time from 9h to 2h in 6 weeks.

Open to a 15-minute call next week?

Mara Lindqvist, founder, Triagely

Evidence: 3 Support Specialist openings, posted May 2 (ledgerly.example/careers)
Proof used: "A 120-person fintech support team cut first-response time from 9h to 2h in 6 weeks."

Email 2 (day 3)
Subject: Onboarding new support hires

Hi Dana,

New agents tend to lose time deciding which tickets come first. Triagely makes that call automatically, so new hires can start working the queue on day one. Setup takes under a day.

Worth a 15-minute look next week?

Mara

Evidence: Same careers page (ledgerly.example/careers)
Proof used: "Setup takes under a day."

Email 3 (day 8)
Subject: Closing the loop

Hi Dana,

I'll stop here. If ticket triage turns into a problem as the team grows, I'm happy to show you how Triagely handles it.

Mara


Shiftboard Cloud (Grade B)

Email 1 (day 0)
Subject: Trial launch and tickets

Hi Omar,

Shiftboard Cloud launched its self-serve trial on April 18. Trial users often bring a surge of how-to tickets that bury the urgent ones. Triagely sorts Zendesk tickets by urgency so your team answers the right ones first. It's used by 40 SaaS support teams.

Open to a 15-minute call next week?

Mara Lindqvist, founder, Triagely

Evidence: Self-serve trial launch, Apr 18 (shiftboardcloud.example/blog/trial)
Proof used: "Used by 40 SaaS support teams."

Email 2 (day 3) and Email 3 (day 8): written in the same format and passed all checks.


Fact and spam check: Pass 2 of 3 passed. Pass 1 failed one rule: Shiftboard Email 1 was over the word limit at 131 words.

Before you send

  • [ ] Open every evidence URL.
  • [ ] Add your postal address and a one-line way to opt out (required under CAN-SPAM).
  • [ ] If any prospect is in the EU or UK, confirm you have a lawful basis for cold B2B email there (GDPR/PECR).
  • [ ] Send from a warmed-up domain with SPF, DKIM and DMARC configured.

STOP: Approve to export this sheet, or reject to stop. Nothing has been sent.

Written by AI from this prompt. Your results will vary.

Run it in Claude Code or Codex: save as .claude/skills/prospect-research-outreach/SKILL.md (or .agents/skills/ for Codex) and say "run the Prospect research and first-touch emails workflow"; parallel steps run as subagents. Or paste it into ChatGPT or Claude and answer its questions. It asks for: icp, offer, proof_points, prospects, sender, cta. Build or run it step by step at prompts-gpt.com/workflows/prospect-research-outreach.

Make it repeatable

Keep using this prompt without pasting it in every time.

Prompt details

By
Prompts-GPT editors
Published
How prompts are written and checked

Did this prompt work?

Keep exploring

More in Multi-Agent Pipelines