# Cold Email — Marketing Skill for AI Agents

A cold email gets replies when it reads like a note from a peer who noticed something relevant, not a pitch. Use a 2–4 word lowercase subject that looks internal, open with an observation tied to the prospect's problem, give one proof point instead of a feature list, and end with one interest-based ask ("Worth exploring?") rather than a 30-minute call. Keep the first email under about 80 words, plain text, with no link. Follow up 3–4 times on widening gaps (day 3, 7–8, 14, 21–28), each with a new angle, and honor the breakup email. Send only from warmed secondary domains, verify lists to keep bounces under 2%, turn off open tracking, and measure replies.

> Maintained by Corey Haines · v2.2.4 · Updated 2026-11-09

Install: `npx skills add coreyhaines31/marketingskills`
Source: https://github.com/coreyhaines31/marketingskills/tree/main/skills/cold-email

## How it works

1. **Gather the five inputs** — The agent reads your product marketing context file if one exists, then fills gaps: who you're writing to and why them, the outcome you want, the problem you solve, your proof, and any research signals like funding, hiring, or stack changes. It doesn't block on missing inputs. It writes with what it has and notes what would make the email stronger.
2. **Pick a shape and personalize** — It chooses a structure that fits the situation (Observation → Problem → Proof → Ask, Trigger → Insight → Ask, PAS, and others) or writes freeform. Personalization has to connect to the problem: if you can delete the opening line and the email still makes sense, the personalization isn't working. When research finds no real hook, it falls back to a segment-level message instead of inventing one.
3. **Write the subject and follow-ups** — Subject lines stay short, lowercase, and colleague-looking, with no first name, numbers, or urgency. The agent then builds a 3–5 email sequence on increasing gaps, rotating one new angle per email: a fresh insight, a similar-company case study, a resource, then a breakup email that leaves the door open. Every email stands alone, and none say "just checking in."
4. **Check the sending setup** — Before the first send, the agent checks deliverability: secondary domains with SPF, DKIM, and DMARC, 2–3 mailboxes per domain at 20–30 sends a day, and 3–4 weeks of warmup. It verifies lists right before upload, turns off open and click tracking, adds a plain-text opt-out line and postal address, and sets stop-loss rules such as pausing any campaign above 2% bounces.
5. **Coordinate channels and plays** — For higher-tier accounts it lays out a multichannel cadence, such as the 21-day Tier 1 sequence mixing email, LinkedIn, calls, and short video. It can also pick a play: value-first teardown for agencies, product-led outbound on usage triggers, problem-led outreach with a free sample, or founder-led notes. A reply on any channel stops every channel.
6. **Triage and answer replies** — Each reply is classified as positive, information request, objection, not now, referral, out of office, unsubscribe, or bounce. Positive replies get an answer within minutes with two proposed times plus a booking link. The agent can handle out-of-office, unsubscribes, bounces, and not-now replies on its own, and drafts the rest for a human to send.

## When to use it

- "cold email"
- "cold outreach"
- "outbound"
- "prospecting email"
- "SDR emails"
- "follow-up sequence"
- "nobody's replying"
- "email deliverability"
- "sending domains"
- "warm up inboxes"

## What's inside

**The playbook covers:** Writing Principles, Voice & Tone, Structure, Subject Lines, Follow-Up Sequences, Before You Send, Beyond Email, When They Reply, Quality Check, What to Avoid, Tool Integrations

### Reference library

- [Benchmarks, Data & Expert Methods](https://github.com/coreyhaines31/marketingskills/blob/main/skills/cold-email/references/benchmarks.md) — Almost every public cold email benchmark comes from a vendor that sells sending, data, or agency services, and each measures differently (some count auto-replies as replies, some don't). Use these as ranges, and judge yourself against your own baseline.
- [Cold Email Deliverability](https://github.com/coreyhaines31/marketingskills/blob/main/skills/cold-email/references/deliverability.md) — The best cold email does nothing if it lands in spam. Deliverability is set up before the first send and watched every week after. This reference covers the sending setup, the mailbox providers' rules, volume limits, warmup, monitoring, and when to stop.
- [Follow-Up Sequences](https://github.com/coreyhaines31/marketingskills/blob/main/skills/cold-email/references/follow-up-sequences.md) — The first email draws most replies (about 58% in Instantly's 2026 benchmark), but the remaining ~42% arrive on follow-ups. A sequence that stops after one send gives up nearly half its replies.
- [Cold Email Copywriting Frameworks](https://github.com/coreyhaines31/marketingskills/blob/main/skills/cold-email/references/frameworks.md) — Frameworks beat templates — they teach thinking patterns, not copy-paste shortcuts.
- [LinkedIn Outreach](https://github.com/coreyhaines31/marketingskills/blob/main/skills/cold-email/references/linkedin-outreach.md) — LinkedIn is the second channel in most B2B outbound and the first for some audiences: people who ignore cold email often accept a relevant connection request. It also has the tightest limits and the highest account risk of any channel. This reference covers the profile, the limits, the sequence, the copy, and the automation decision.
- [Multichannel Cadence](https://github.com/coreyhaines31/marketingskills/blob/main/skills/cold-email/references/multichannel-cadence.md) — Most buyers need to see a name a few times, in more than one place, before they respond. A cadence coordinates email, LinkedIn, phone, and the occasional video or package so each touch builds on the last and nobody gets hit on four channels in one afternoon.
- [Outbound Plays](https://github.com/coreyhaines31/marketingskills/blob/main/skills/cold-email/references/outbound-plays.md) — A play is a repeatable outbound motion: a trigger, an offer, a sequence, and a definition of success. Plays keep outbound from becoming a pile of one-off emails. This reference covers four that work in 2026, with the conditions each needs.
- [Personalization at Scale](https://github.com/coreyhaines31/marketingskills/blob/main/skills/cold-email/references/personalization.md) — Personalization drives 50–250% more replies (Lavender). The key insight: if your personalization has nothing to do with the problem you solve, it's just an attention hack (Clay).
- [Handling Replies](https://github.com/coreyhaines31/marketingskills/blob/main/skills/cold-email/references/reply-handling.md) — The sequence's job ends when someone replies. What happens in the next hour decides whether that reply becomes a meeting. This reference covers classifying replies, responding to each type, speed, booking, and what an agent can handle on its own.
- [Subject Line Optimization](https://github.com/coreyhaines31/marketingskills/blob/main/skills/cold-email/references/subject-lines.md) — The subject line determines whether the email gets read. The data is counterintuitive: short, boring, internal-looking subject lines win decisively.

## Example

**Prompt:** Write a cold email to VP of Marketing at mid-size B2B SaaS companies. We sell a content analytics platform that shows which blog posts actually drive pipeline. Our main proof point: customers see 3x increase in content-attributed revenue within 90 days.

**Email 1 · Subject: pipeline from content**

Hi Jordan,

Most marketing teams can show which posts get traffic. Far fewer can show which ones show up in closed deals, so the budget keeps going to whatever looks busy.

We built a platform that maps each blog post to pipeline instead of sessions. Customers typically see 3x more content-attributed revenue within 90 days, mostly by putting more behind the handful of posts that actually close.

Worth seeing which of yours those are?

**Follow-up 1 (day 3) · Subject: re: pipeline from content**

One thing we see a lot: a team's top-traffic posts and its top-pipeline posts are rarely the same list.

If useful, I can pull last quarter's posts for your blog and send back the ones that touched deals. It usually takes us a day, and you can keep the list either way. No call needed.

Want me to?

(Sent as a reply in the same thread, still plain text, still no link.)

**Notes on the sequence**

Follow-up 2 (day 7–8): case study from a similar mid-size SaaS team.
Follow-up 3 (day 14): one insight on how pipeline-driving posts differ from traffic posts.
Follow-up 4 (day 21–28): breakup, offering the audit one last time.

To make email 1 stronger: a real signal per account (a new content hire, a recent board-reporting change) and a named customer you're cleared to cite. Send plain text from a warmed secondary domain, with open tracking off and an opt-out line in the signature.

## More things to ask

- Help me write a cold email to CTOs at enterprise companies. I sell cybersecurity training. My current email has a 2% open rate and 0% reply rate.
- write me a follow-up sequence. prospect didn't reply to my first email about our HR software. how many should I send and how far apart?
- What are the best subject lines for cold emails? I want to maximize open rates.
- Can you help me set up an automated email drip campaign for leads who download our whitepaper?
- Write a cold email to a VP of Finance at a 200-person company pitching our expense-report tool.

## FAQ

### What makes a good cold email subject line?

Short, boring, and internal-looking: 2 to 4 lowercase words that could have come from a colleague, like "reply rates" or "hiring ops." The skill's subject-line data shows 2-word lines getting about 60% more opens than 5-word ones (Lavender), and the prospect's first name in the subject costing about 12% of replies. Skip numbers, urgency words, emojis, product pitches, and fake "Re:" or "Fwd:" prefixes.

### How long should a cold email be?

Under about 80 words for the first email, and the skill's benchmarks put the sweet spot at 25–75 words, with emails under 75 words getting 83% more replies. Aim for a 3rd–5th grade reading level. Every sentence has to move the reader toward replying, and "you" should outnumber "I" and "we." If the email feels like it could be shorter, it should be.

### How many follow-ups should I send, and how far apart?

Three to five emails in total, capped at four follow-ups, on increasing gaps: day 0, day 3, day 7–8, day 14, and a breakup email on day 21–28. Response rates drop sharply and spam complaints climb by the fourth follow-up, so each email must add one new thing, such as an insight, a case study, or a resource. If you send a breakup email, don't contact that person again.

### What is a good cold email reply rate?

Instantly's 2025 send data puts the average around 3.4%, top quartile at 5.5% or more, and top decile above 10%. Signal-based, small segments often reach 2–3 times your static-list rate. Benchmarks come from vendors that measure differently, so judge yourself against your own baseline. Ignore open rate: Apple Mail Privacy Protection inflates it and the tracking pixel can hurt deliverability. Track replies, positive replies, and meetings.

### Why are my cold emails going to spam?

Common causes: sending from your primary domain, missing SPF, DKIM, or DMARC, skipping the 3–4 week warmup, bounces above 2%, open tracking pixels, links or images in the first email, and identical copy across thousands of sends. Check placement with seed tests and reputation in Google Postmaster Tools weekly. Pull any mailbox below about 70% inbox placement back into warmup, and never scale volume to fix a low reply rate.

### Can an AI agent run cold outreach on its own?

Partly. The skill lets an agent handle out-of-office replies, unsubscribes, bounces, and not-now replies on its own, and draft positive replies, objections, referrals, and information requests for a human to send until accuracy is proven over 100+ replies. Before any autonomous send, every claim should match an approved source, nothing from the banned-phrase list should appear, and a daily cap should apply. Personalization research should cite a source for every fact.

## Related skills

- **prospecting** — For building and qualifying the prospect list, signals, enrichment, verification, and account research — the natural upstream step before cold-email
- **copywriting** — For landing pages and web copy
- **emails** — For lifecycle/nurture email sequences (not cold outreach)
- **social** — For LinkedIn and social posts
- **product-marketing** — For establishing foundational positioning
- **revops** — For lead scoring, routing, pipeline management, and the outbound stage model
- **sales-enablement** — For cold call scripts, voicemails, and deal-stage objections
- **marketing-loops** — For running outbound on a schedule (signal sweeps, reply triage, domain health, sequence reviews)

---
Canonical: https://marketing-skills.com/skills/cold-email