Draft Outreach in the Conversation
Outcome: a message drafted from the enrichment already in your session — no new calls, no template, and nothing sent that you have not read.
- Surface
- MCP server
- Level
- Beginner
- Uses
- No new calls — uses what you pulled
- Credits
- 0
- Prerequisite
- A session with enriched data in it
The data is already there
By the end of any chain in this course, the assistant has the evidence in context: a job post, a comment, a stack detection, a recent hire. Drafting from it costs nothing extra.
That is the advantage of working this way. In a table you assemble evidence into a column and then generate; in a conversation the evidence is already in the conversation.
The drafting prompt
Using only the data already in this conversation — no new tool calls —
draft the opening line of an email to <person>.
Rules:
- One sentence, under 25 words.
- Reference a specific fact from what we pulled. Name it.
- Do not open with "I noticed", "I came across", "I saw that",
or "I was impressed by".
- Do not compliment the company.
- Do not mention our product or ask for a meeting.
- If we do not have a specific fact about them, say SKIP.
Show me the fact you used alongside the sentence.“Show me the fact you used” is the line that makes this reviewable. Without it you cannot tell whether the sentence came from the enrichment or from the model’s general knowledge — and those are very different things, one of which is checkable.
The test
Could this sentence have been sent to a different company?
If yes, it is decoration. Reject it and rewrite the constraint, not the sentence.
| Weak | Strong |
|---|---|
| ”I saw you’re hiring in RevOps." | "Two RevOps roles in a month usually means someone is about to own three systems that disagree." |
| "Congrats on the funding!" | "Post-Series-A is normally when the spreadsheet the team ran on stops holding." |
| "I noticed you use Acme." | "You added Acme in Q1 — most teams hit its reporting ceiling around month four.” |
The formula: a fact you did not state plus an implication you did. They make the connection, which is what makes it read as understanding rather than monitoring.
The abstain rule
If the enrichment produced nothing specific, the honest answer is SKIP.
A model asked to personalize with nothing to work from will invent — a compliment about a project that does not exist, growth inferred from a headcount number. The recipient notices, and it costs more than sending nothing.
Check that SKIP fires. Run the prompt on a row where you know there is no evidence. If it produces a sentence anyway, the constraint is not working.
What to draft and what not to
Draft the opening line
The one sentence that must differ per person.
Do not draft the pitch
It is the same for everyone. Write it once, properly, yourself.
Do not draft the ask
Same reason, and a generated call-to-action tends toward the generic.
Do not draft a whole sequence in one go
You get four variations of the same sentence. See AI-Powered GTM lesson 15.
Read before sending
Every message, on this surface, because the volumes are small enough that you can.
- Read it aloud. Bad AI copy reads fine on screen and sounds wrong spoken.
- Check the fact is actually true — the assistant showed you which one it used.
- Check the tone against the person’s seniority and the channel.
- Ask whether you would send it to someone you know.
Do this now
Take a session from lesson 02, 05 or 06
With enrichment already in context.
Run the drafting prompt
With the “no new tool calls” constraint.
Check the fact it used
Against the enrichment output.
Apply the interchangeability test
Test the abstain path
On a row with no evidence.
Read three drafts aloud
Send one
Then compare the reply against your usual baseline.
Check your work
- No new tool calls were made during drafting
- Every draft names the fact it used
- No draft could have been sent to another company
SKIPfires on evidence-free rows- You read the messages aloud before sending
Where this breaks
Allowing new tool calls during drafting quietly turns a free step into an expensive one — the assistant decides it needs more context, runs three more lookups per person, and the drafting stage costs more than the sourcing did. “No new tool calls” is one line and it keeps this step at zero.
Course complete
You can now connect an MCP client, build a contact card, brief a company before a call, backfill a bare email list, source a warm list from post engagement, turn a job post into a buying committee, prompt without wasting credits, and draft outreach from what you pulled.
Where to go next:
| You want | Course |
|---|---|
| The same work as repeatable tables | SyncGTM 101 |
| Agentic builds and MCP chains | AI-Powered GTM |
| Every MCP tool, clustered by job | SyncGTM MCP Mini Course |
| To roll this out to a team | AI for RevOps |
Reference for this lesson: MCP prompting guide, MCP tools, AI-Powered GTM, Automated Outbound.