> ## Content Index
> Fetch the complete content index at: https://www.liamdarmody.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# AI Experiment 011: To Note or Not to Note
- URL: https://www.liamdarmody.ai/ai-experiment-011/
- Published: 2026-10-27T16:27:39.000Z
- Updated: 2026-10-28T02:44:50.000Z
- Description: Do LinkedIn connection notes work? My data export says yes: invites with a note were accepted at 50%, without at 28%.
- Author: Liam Darmody
- Tags: AI Lab

The short answer: The most-asked question in LinkedIn nerd circles is whether to attach a note to a connection request. My modus operandi was always bare: I wanted people to look at my profile before deciding. Then I fed my full LinkedIn data export to my Claude Social Media Analyst, and it surfaced a stat I didn't think I had data for. Invites with a note: accepted at 50% (31 of 62). Without: 28% (77 of 272). Nearly 2x. My assumption was wrong, and the method that proved it is the real story here.

![](https://storage.ghost.io/c/4c/2e/4c2ec722-2bd5-4a69-ad93-121c6c1c3e73/content/images/2026/10/Screenshot-2026-10-27-at-12.23.29---PM.png)

## The question

"To note or not to note" comes up constantly, and my answer was always the same: no note. The reasoning felt solid. If someone's deciding whether to connect with me, I'd rather they see the profile first. A note felt like a sales pitch before the handshake.

But "felt solid" is not the same as "data-backed." The only way to answer it, I believed, was to manually log every invite: note or no note, accepted or ignored. Who has time for that? Not I.

## The build

I pointed my Social Media Analyst at my LinkedIn data export. If you've never pulled yours: 

- Me dropdown
- Settings & Privacy
- Data Privacy,
- Download your data.

You get a pile of CSVs, including Invitations.csv, which has a Message column. Send an invite with a note, the note text lands there. Send the bare "Connect" with no note, the field stays empty.

So "has a note" equals Message is non-null.

Brilliant! How did I not know about this sooner? 

334 outgoing invites in my export: 62 with a note, 272 without. Then the matching step, because the export has no accept/decline field: Claude matched invitee profile URLs against my Connections.csv. If the URL appears in my connections list, the invite got accepted.

The result: 31 of 62 with notes accepted (50.0%), 77 of 272 without (28.3%). Only 19% of my outgoing invites carried a note. The thing I almost never did was working nearly twice as well.

Oops.

## What I'm testing

This one's already run, since the data is historical. What I'm testing now is whether the finding changes my behavior, and whether the method generalizes.

First, the behavior change. Will I actually start writing notes? The 50% number is compelling, but 62 invites is a small sample, and there's a selection-bias smell I can't ignore: I probably wrote notes to people I already had a warmer reason to contact. Warmer prospects accept more often. The note might be riding shotgun on intent that was already there.

Second, the method. The invitations trick is one file in a much bigger export. Connections, messages, posts, profile views: there's a whole quantitative self-portrait sitting in that download that would take serious manual effort to replicate. The real experiment is what else is in there.

## Caveats

My assumption was wrong, or at least incomplete. "Let them look at my profile first" sounded principled. The numbers say the note doubles acceptance. I don't love being wrong in public, but that's the lab.

The caveats, because they matter:

- 62 is a small n.
- Acceptance is derived, not reported: URL matching is clever, but it's inference, not a field.
- And the selection bias problem is real and unmeasured.

Still: 50% versus 28% is a big enough gap that "no note" is no longer my default. The burden of proof flipped.

## Worth borrowing

1. Download your LinkedIn data. The export is the most underused analytics tool in networking. It holds answers to questions you'd never track manually.
2. The Message column in Invitations.csv is the whole trick. Note versus no-note is one filter away.
3. Upload it to your favorite LLM and ask for a deep analysis. You don't need a data team. You need a good prompt and a skeptical eye.
4. Distrust your own defaults. I had a principled-sounding reason for no-note, and it was worth about half the acceptance rate.

Sources: LinkedIn data export (Invitations.csv, Connections.csv), analyzed October 8, 2026\. 334 outgoing invites; acceptance derived by URL matching against Connections.csv.