Outreach with Claude

Here is what I know about {contact} at {company}:
{paste the job post, the podcast transcript, the
support doc, the pricing page, the earnings call
section, whatever you actually have}

I sell {product} to {buyer type}.

Pick the ONE fact in there that a competitor of mine
would not have bothered to read. Build the whole email
on that fact.

Rules:
- First sentence is about them, not me, not my company
- 70 words maximum
- No em dashes, no en dashes
- No "I hope this finds you well", no "I noticed"
- One ask, at the end, and make it small
- Do not describe my product. Describe the problem the
  fact implies they already have.

Then tell me which fact you picked and why you picked
it over the others.

The line doing the work is “a fact a competitor would not have bothered to read”. It pushes the model past the funding announcement and the homepage headline, which every other rep in the inbox already used that week. Asking it to justify the choice at the end is your check on whether it actually read anything.

Here are the five emails in my current sequence:
{paste all five}

Rewrite each one so the subject of every sentence is
the prospect or their situation, never my product.

Constraints per email:
- One idea only
- Under 90 words
- One ask, at the end
- No em dashes or en dashes anywhere
- Email 3 must be readable on a phone in under
  eight seconds

Then, before the rewrites, answer this: which two of
these five would you delete entirely, and what is the
sequence trying to say that it is not saying?

“Which two would you delete” is the load-bearing line. Ask for improvements and you get five improved emails. Ask what to cut and the model has to form a view on what the sequence is for. Most five-email sequences are three emails with padding.

Here is the email you just wrote:
{paste it back}

Now do three things.

1. Name the single weakest sentence in it and say
   exactly why it is weak.
2. Tell me which sentence a busy {buyer title} would
   skip on a phone screen.
3. Tell me what a sceptical version of this person
   would think when they hit the ask.

Do not rewrite anything yet. Just answer.

“Do not rewrite anything yet” is the whole trick. If you let it rewrite in the same turn it will paper over the weak line instead of naming it, and you learn nothing. Getting the critique first means the second draft fixes a diagnosed problem rather than reshuffling adjectives.

Here is the cold email I am sending {contact}:
{paste}

Write the LinkedIn version. This is a different
channel, so treat it as one:
- It sits next to their friends and their old
  colleagues, so a formal opening reads as a sales
  robot
- Under 400 characters
- No subject line thinking, no sign-off, no title
- Lowercase is fine. Sentence fragments are fine.
- It should read like something a human typed on a
  phone between meetings
- No em dashes or en dashes

Keep the same specific fact. Lose everything else.

The line about sitting next to their friends and old colleagues is what changes the output. Without it you get the email with the greeting removed, which is the single most obvious tell that a message was recycled. The character cap does the rest.

I emailed {contact} twice about {topic}. No reply.
Both emails:
{paste}

Here is what has happened since:
{paste anything new, or write "nothing"}

Write a third email that:
- Does not reference the earlier two
- Does not ask if they saw my last email
- Does not invent urgency or a fake deadline
- Gives them a reason to reply that did not exist
  when I sent email two

If there is genuinely no new reason in what I gave
you, do not write the email. Say so, and tell me what
single piece of information would make a third touch
worth sending.

The last paragraph is the one people delete, and it is the only one that matters. Left to itself the model will always produce the email, because producing the email is what it was asked to do. Giving it an explicit exit means the times it refuses are real signal about the account.