Fix sheet: Using pure templates on outbound messages

This sheet appended when your report flagged the `templates` weakness.

Why this is now worse than it was 3 years ago

LinkedIn's bot classifier got dramatically better at "this exact string was sent to 200 people last week" in the 2024-25 model generation. Templates that worked in 2022 — Hi {{first_name}}, I noticed you... — are now a clean fingerprint. Templated outbound is fighting an algorithm built specifically to detect it.

Templates correlate with BOTH lower acceptance AND faster account flags. The first symptom is acceptance — you push the same template to a new audience and it lands at 8% acceptance instead of 25%. The second symptom is a 24-72h restriction.

The 3 things to fix this week

AI-rewrite touch 1 per profile. Use a per-prospect AI rewrite (Claude, GPT, whatever you have access to) to generate the connection note. Feed it the profile's recent post titles, role, company. Output: one sentence that is unique per recipient, even if the underlying structure is the same.

NEVER reuse touch 5 verbatim. Touch 5 is the "I am stopping here" close. If you send the same touch 5 to 200 people in a week, the algorithm will fingerprint it AND prospects who got it once and gave you nothing will see it again from another teammate using the same template. Both bad outcomes.

Audit your current sequence for duplicate strings. Open the tool you use to send sequences. Count how many of your last 100 messages contained the exact phrase "I help" or "Hope you are well" or "Wanted to reach out" — these are the canonical bot tells. Rewrite the ones that recur.

What "fixed" looks like

Acceptance rises 5-15 percentage points over 2-3 weeks even with no other changes. Account-level risk markers drop. The qualitative test is: read three of your outbound messages back-to-back; if they sound like the same person could not have written all three to a friend, they are still templated.

How BIGM solves it

We generate the per-profile opener in <200ms using public profile data + your ICP brief. No template is visible anywhere — every message has the same structure but completely different surface text. LinkedIn's classifier cannot fingerprint what does not repeat.