Free playbook · No email required
Marketing's Outbound Prompt Playbook
For the marketing lead whose SDR team sends 200 emails a week that all sound the same.
Six AI prompts for personalized cold outreach. Marketing writes the context block once. Sales runs the prompts per prospect. Research, messaging, and call prep for one account drops from an hour to a few minutes. Every prompt is on this page. Copy what you need.
Works with Claude or ChatGPT. No form, no gate, no signup.
By Dorian Kendal, Kendal Solutions. Last updated August 2026.
Layer in AI is one of the ways we help teams move faster.
The short version
Marketing writes one context block describing what you sell. Sales runs six prompts against it, one prospect at a time. Prompt 1 researches the account. Prompt 2 finds a reason to reach out now. Prompts 3 to 5 draft the email, the follow-up, and the LinkedIn note. Prompt 6 preps the call. A person checks every output before it goes out. Free, no form, works with Claude or ChatGPT.
The argument
Should Marketing or Sales own cold outreach messaging?
Marketing owns the context block. Sales runs the prompts.
Outbound messaging is a Marketing job that got handed to Sales. When reply rates drop, the reps get blamed. That is the wrong address. A rep writing positioning from scratch, forty times a week, will drift. Anyone would.
Marketing owns the message on the site, in the deck, and in every campaign. Outbound is the one place it got dropped. This playbook puts it back where it belongs.
| What | Marketing owns | Sales owns |
|---|---|---|
| The context block | Writes it, updates it when positioning changes | Never has to paste it into a chat |
| The prompt set | Keeps it current in one place | Runs it per prospect |
| Research | Carried by prompts 1 and 2 | Verifies what comes back |
| The message | Drafted by prompts 3 to 5 | Edits it, then sends it |
That split is the whole idea. The prompts below are just how it gets done.
Set it once
The context block
This teaches the AI what you sell, so every prompt after it produces on-message output. Marketing fills it in once and keeps it current.
In Claude: create a Project and paste the block into the project instructions. Every chat started inside it already knows the context.
In ChatGPT: put the block in a Project's instructions, or build a custom GPT with it baked in and share it with the team.
The context block
WhenOnce, before anything else. Update it when positioning changes.
Context: I sell [PRODUCT/SERVICE], a [ONE-LINE DESCRIPTION OF WHAT IT IS AND THE MAIN OUTCOME IT DELIVERS]. How it works: [HOW IT'S DEPLOYED, INTEGRATED, OR ADOPTED: setup, time to value, anything notable]. What it includes: [KEY CAPABILITIES OR MODULES, EACH WITH THE PROBLEM IT SOLVES]. How we're different: [MAIN ALTERNATIVES OR INCUMBENTS AND WHAT THEY MISS]. Proof: [CERTIFICATIONS, COMPLIANCE, NOTABLE CUSTOMERS, OR RESULTS THAT BUILD TRUST]. Who buys: [TARGET ROLES AND SEGMENTS]. Keep this in mind for every prompt that follows.
The playbook
Six AI prompts for personalized cold outreach
They run in sequence. Each one feeds the next. Research, then messaging, then call prep. Fill in the brackets and go.
Jump to a prompt
Paste the lot into a Claude Project or a ChatGPT Project once, and the whole team has them.
The prospect brief
WhenStart here. One prompt covers the company, the person, their situation, and their likely pain.
Research [NAME], [TITLE] at [COMPANY] ([WEBSITE]) and give me one brief I can read in five minutes. The company: what they do, who they sell to, rough size, and which [REGULATIONS, STANDARDS, OR MARKET FORCES] likely apply. Their situation: [WHAT TOOLS, SYSTEMS, OR CONDITIONS ARE RELEVANT TO WHAT I SELL]. Do they use [RELEVANT PLATFORM] or a [RELEVANT INCUMBENT]? Check job postings, LinkedIn profiles, and [OTHER PUBLIC SIGNALS]. Label each finding confirmed or inferred. The person: career history, time in role, what they post about, and anything public I could reference naturally. Their likely pain: the top three [PROBLEM AREA YOUR PRODUCT ADDRESSES] problems someone in this role deals with, in their words. Cite sources so I can verify.
Reason to reach out now
WhenRight after the brief. A recent event beats a cold opener every time.
Search for news about [COMPANY] from the last 90 days that gives me a reason to reach out now: [RELEVANT TRIGGER EVENTS: a leadership change, funding, M&A, rapid hiring, a public incident, an audit finding, a new initiative, or a technology change tied to what I sell]. For each: one sentence on what happened, one on why it makes [YOUR TOPIC] a live issue for them. If you find nothing, say so. Do not stretch.
First cold email
WhenFeed it your findings from prompts 1 and 2. That is what makes it personal.
Write a first cold email to [NAME], [TITLE] at [COMPANY]. Use this research: [PASTE 2-3 FINDINGS FROM PROMPTS 1-2]. Rules: under 75 words, plain text. Open with something specific to them or their industry, not "I hope this finds you well." One idea per email. No buzzwords. End with a real question, not a demo ask. Write three versions with different openers so I can pick.
Follow-up
WhenNo reply after 3 to 5 business days. It has to add something new.
Write a follow-up to this email I sent [X] days ago with no reply: [PASTE ORIGINAL EMAIL]. It must add something new. A relevant stat, a peer example, or a different angle on the same problem. Never "just circling back." Under 60 words. Two versions: one lighter, one more direct.
LinkedIn connection note
WhenSecond channel alongside email, same day or the day after.
Write a LinkedIn connection request note to [NAME], [TITLE] at [COMPANY]. Under 40 words (the limit is 300 characters). Reference: [SOMETHING SPECIFIC: their industry, a post they wrote, a shared interest]. No pitch. The goal is an accepted connection, nothing more. Three options.
Call prep
WhenA call is booked. Run this an hour before.
I booked a call with [NAME], [TITLE] at [COMPANY]. Here is my research: [PASTE OUTPUTS FROM PROMPTS 1-2]. Current tool, if known: [INCUMBENT]. Give me a one-page brief: what matters most to this person, the three best discovery questions, the objections they are most likely to raise (including where they have a point, and how to respond honestly), and the one thing I should not do on this call. Keep it scannable.
What it looks like in practice
A worked example, start to finish
Say you sell a data observability tool. Here is one account run through prompts 1 to 3. The company is invented, the shape is real.
What prompt 1 came back with
- Meridian Freight, 340 people, sells logistics software to regional carriers. Confirmed
- Hiring two analytics engineers and a data platform lead. Confirmed, from job postings.
- Runs Snowflake and dbt. Inferred, from those job descriptions.
- Priya Raman, VP Data Engineering, 14 months in role, posts about pipeline reliability.
What prompt 2 found
- Shipped a carrier-facing analytics product six weeks ago. That puts new pipelines in front of paying customers, where breakage gets noticed.
What prompt 3 wrote
"Priya, you shipped carrier-facing analytics last month. That usually means pipeline problems stop being an internal annoyance and start being a customer email."
"We help data teams catch broken pipelines before the customer does. Two of your three open roles are platform hires, so I am guessing this is already on your list."
"How are you catching bad data today, before or after someone downstream notices?"
Why it works.
Nothing in that email could be sent to another company. The opener came from prompt 2. The guess about the open roles came from prompt 1, and it is framed as a guess rather than stated as fact. That is the standard. If your draft would survive a copy-paste to a different account, it is not personalized.
Read this before you send anything
What AI gets wrong here, and what you have to check
Treat every output as a draft. These prompts assume a person reads the result and edits it. Five things go wrong often enough to name:
It invents titles and tenure. Check every name and title against LinkedIn before you send.
It states guesses as facts. Prompt 1 asks it to label findings confirmed or inferred. Read those labels.
It reaches for a hook that is not there. Prompt 2 tells it to say so when there is no news. When it says so, believe it.
It sounds like AI when you give it nothing. That is the context block's job. Skip the block and you get brochure copy.
It cannot judge whether the person is worth emailing. That call stays yours.
Want the PDF version?
Same six prompts, formatted to print or pass around. No form.
Download the playbook (PDF)Questions
Common questions about these prompts
Marketing should own the context block and the prompt set. Sales should run the prompts per prospect and edit every message before sending. A rep writing positioning from scratch dozens of times a week will drift from the message. Marketing already owns the message on the site, in the deck, and in every campaign. Outbound is the one place it usually gets dropped.
It is a free set of six AI prompts for personalized cold outreach. Marketing sets up a context block once. Sales runs the prompts per prospect. The sequence covers account research, the reason to reach out, the first email, the follow-up, a LinkedIn note, and call prep.
No. Every prompt is published on this page and the PDF downloads without a form.
Both. Put the context block in a Claude Project's instructions, a ChatGPT Project, or a custom GPT. Any assistant that can search the web and hold context will run these prompts.
Marketing writes the context block and updates it when positioning changes, then keeps the prompt set current in one place. Sales runs the prompts per prospect, verifies what comes back, and edits every message before sending. Nobody has to paste positioning into a chat and nobody sends unreviewed AI output.
Minutes instead of the hour or more it takes by hand. Prompt 1 produces a brief you can read in five minutes. Prompts 3 through 5 draft the messages. You still verify facts and edit before sending.
Yes. Run prompts 1 and 2 across a target account list to build the ABM plan, then run prompts 3 through 6 for each contact.
Not if you fill in the context block and feed real research into prompt 3. Generic AI email comes from generic input. The word limits and the ban on opener cliches handle the rest.
A short description of what you sell, how it works, how you are different, your proof, and who buys. You write it once. Every prompt reads it, so output stays on message without anyone pasting positioning into each chat.
There is no new software and no per-seat cost. You keep control of the messaging and every message gets human review. AI SDR tools automate sending. This speeds up the thinking and the drafting.
Names, titles, and any claim about the company. Prompt 1 labels each finding confirmed or inferred for that reason. Treat every output as a draft.
Before you hit send
You just personalized outreach to 20 accounts. Three of them already visited your pricing page twice.
Do you know which three? Most in-market demand never raises a hand. Accounts research you, compare you, and come back, without filling in a form. Personalization is worth more when you aim it at people already looking.
Invisible Demand Diagnostic
Find the accounts already looking.
The Invisible Demand Diagnostic surfaces the demand already forming in your market but not yet in your pipeline, so you meet it early.
See the Invisible Demand DiagnosticWork with us
Want it set up for you?
Kendal Solutions helps teams put this playbook to work. Setup, customization, and automation, plus go-to-market foundations and any other marketing help you need.
Talk to us. Email dorian@kendalsolutions.com