Social Media
As of Sep 18, 2026, 4:14 AM ET - scorecard week ending Sep 18
The cold-open scene with a named person and real dialogue took the top of both channels that move, for the second time in the corpus. Against that, the week's two widest-distributed posts - 795 LinkedIn impressions and 696 X views on the same Stripe master - returned three engagements between them, because Matthew's own stake sits in paragraph four.
3 masters, 5 platforms
35 of it on Substack
LinkedIn, Sep 14
both the Sep 17 Stripe post
Scorecard
| Date | Platform | Hook | Eng | Rate | C/L | Read |
|---|---|---|---|---|---|---|
| Sep 14 | Substack | I asked someone who bootstrapped an agency to $15M and sold it whether I had already missed my window. | 19 | n/a | 0.29 | Best post of the week on any platform. Cold-open scene, a named person, real dialogue, and Matthew asking the embarrassing question instead of the flattering one. Four comments, the highest count in the window. |
| Sep 17 | Substack | Stripe says the hard part of their internal AI agent was governance, not model choice. | 9 | n/a | 0.50 | Three comments on six likes. The Substack framing puts Matthew's own scale-of-one experiment up front; the LinkedIn version of the same idea buries it and drew 1. |
| Sep 14 | I asked a man who built and sold a fifteen million dollar agency whether I had already missed my window. | 7 | 0.45% | 0.00 | Best engagement rate of the week on 1,539 impressions and the strongest LinkedIn post since Sep 11. Same master as the top Substack note. The Kaufer shape, and this is its second appearance at the top. | |
| Sep 15 | Substack | John Cutler is right that the consensus you see in a company is mostly people adapting to what the powerful already believe. | 7 | n/a | 0.17 | Opens on Cutler and still cleared 7 here against 1 on LinkedIn. The note gets to Matthew's two-week window faster than the post does. |
| Sep 17 | X trap | Stripe built an internal agent 10,000 employees use weekly. Their read: governance, not model choice. | 2 | 0.29% | n/a | 696 views against an X median of 22 this quarter - the widest X distribution on record for the account - and it converted one reply and one quote. |
| Sep 14 | Bluesky | I asked a guy who bootstrapped an agency to $15M and sold it whether I'd missed the AI window. | 2 | n/a | 1.00 | One like and one reply. The only Bluesky post to draw a reply in the pulled history. |
| Sep 17 | LinkedIn trap | Stripe's Sharadh Krishnamurthy walked through Kai this week, the internal AI agent more than ten thousand Stripe employees now use weekly. | 1 | 0.13% | 0.00 | 795 impressions, the second-widest LinkedIn distribution of the week, and one reaction. Matthew's own exposure - ten agents, one person, governance built from nothing - sits in paragraph four. |
| Sep 15 | John Cutler wrote this week that the consensus you see inside a company is mostly people adapting to what the powerful in the room already believe. | 1 | 0.38% | 0.00 | Only 266 impressions, the narrowest LinkedIn distribution of the week. The unresolved question in the last paragraph is rule 17 done properly; almost nobody saw it. | |
| Sep 17 | Bluesky | Stripe's internal AI agent is used weekly by 10,000+ employees. | 0 | n/a | 0.00 | No engagement. |
| Sep 15 | Bluesky | John Cutler: the consensus you see in a company is mostly people adapting. | 0 | n/a | 0.00 | No engagement. |
| Sep 17 | Mastodon | Stripe's engineering team says the hard part of their internal AI agent was governance. | 0 | n/a | 0.00 | No engagement. |
| Sep 15 | Mastodon | John Cutler wrote this week that the consensus you see inside a company is mostly adaptation. | 0 | n/a | 0.00 | No engagement. |
| Sep 14 | Mastodon | I asked someone who bootstrapped an agency to about $15M and sold it whether I had missed the window. | 0 | n/a | 0.00 | No engagement. |
| Sep 15 | X | John Cutler: the consensus in a company is mostly people adapting to what the powerful already believe. | 0 | 0.00% | n/a | 37 views, 0 engagements. |
| Sep 14 | X | I asked a guy who bootstrapped an agency to $15M and sold it whether I'd missed the AI window. | 0 | 0.00% | n/a | 11 views, 0 engagements. The week's best master, and on X it reached eleven people. |
Sorted by total engagement. C/L is the comment-to-like ratio. Substack, Bluesky and Mastodon expose no impressions, so their engagement rate is null throughout and is never inferred. A distribution trap is 500 or more impressions at an engagement rate under 0.30 percent.
Recent standouts, rolling 10 days
| Date | Platform | Hook | Impressions | Engagement | Rate |
|---|---|---|---|---|---|
| Sep 11 | I once walked into the CEO's office to recommend that we shut down the product I had spent two years building. | 1,871 | 18 | 0.96% | |
| Sep 10 | Substack | A founder I worked with had one constraint that reorganized her whole growth plan: nine to three. | n/a | 8 | n/a |
| Sep 9 | Substack | John Cutler named the recontextualization tax: every handoff costs work to rebuild lost context. | n/a | 8 | n/a |
| Sep 10 | A founder I worked with had one constraint that reorganized her entire growth plan. | 327 | 2 | 0.61% | |
| Sep 11 | Substack | I once walked into the CEO's office to recommend we shut down the product I'd spent two years building. | n/a | 2 | n/a |
What is working and what is not
The cold-open scene with a named person and real dialogue, for the second time in the corpus. The Sep 14 Greg Raiz post opens inside a call Matthew had rescheduled twice, names the person, quotes the exchange, and has Matthew asking the embarrassing question instead of the flattering one. Best engagement rate of the week on LinkedIn at 0.45 percent on 1,539 impressions, and the best raw engagement anywhere at 19 on Substack. The voice profile flags this shape as promising on a single data point, Kaufer at 118. This is the second.
Substack again, and by a wider margin. Three notes drew 35 of the week's 48 engagements, 73 percent, at 11.7 per note, against LinkedIn's 9 off the same three masters. Third consecutive week Substack has out-earned LinkedIn on identical material.
Other-company analysis with Matthew's stake buried. The Sep 17 Stripe post took the week's widest distribution on two platforms, 795 LinkedIn impressions and 696 X views, and returned three engagements combined - 0.13 percent and 0.29 percent, both distribution traps. The exposure is in the post: ten agents, one person, a governance layer built from nothing, and an honest refusal to say which lesson generalizes. All of it sits below four paragraphs about Stripe. Same failure pattern as Sep 9 and Sep 10 last week, and this time it cost real distribution rather than a quiet post.
Cadence. Three masters this week against five last week, with nothing published Sep 12, 13, 16 or 18. LinkedIn engagement fell from 21 to 9. The channel that most needs frequency got 40 percent less of it.
Topic balance
Three masters, each fanned to all five platforms. Three of the six themes are empty. Marketplace mechanics is absent for the first time in the pulled record, and the voice profile names it as Matthew's sharpest and most differentiated angle - which is why it carries the attention flag rather than the other two zeros.
What works, by cohort
Format and topic tagging has started - 20 staged rows carry both tags - but no tagged post has metrics yet, because all 20 are scheduled for Sep 21 to 25. Every row that currently has an engagement rate predates tagging and carries neither, so the by-format and by-topic cuts cannot be computed this run. They fill once the tagged posts publish and are collected. Nothing here is inferred from post text.
Engagement rate by posting day
| Day | Posts | Mean engagement rate | Note |
|---|---|---|---|
| Friday | 1 | 0.96% | thin - not ranked |
| Thursday | 3 | 0.34% | thin - not ranked |
| Monday | 2 | 0.23% | thin - not ranked |
| Tuesday | 2 | 0.19% | thin - not ranked |
| Wed, Sat, Sun | 0 | n/a | no rated posts |
Every cohort is under five posts, so none is ranked as a finding. The whole cut rests on 8 rated rows spanning Sep 10 to Sep 17. Read it as a placeholder that fills, not as a result.
Underperformers
The like-for-like format-and-topic comparison cannot be formed yet for the reason above, so this falls back to a platform-and-weekday cohort and is labelled as the substitute it is.
| Date | Platform | Hook | Rate | Cohort median |
|---|---|---|---|---|
| Sep 17 | Stripe's Sharadh Krishnamurthy walked through Kai this week, the internal AI agent more than ten thousand Stripe employees now use weekly. | 0.13% | 0.37% |
One post qualifies, not five. Six of the seven platform-and-weekday cohorts hold a single post, so no member can sit below its own median, and the one that does not holds two. A median over two posts is not a benchmark - treat this as near-noise that happens to agree with the distribution-trap finding above. The second underperformer list, LinkedIn posts absent from the export top-50 across three or more consecutive exports, cannot be computed: the ledger has one collection pass on file, so there is no export history to compare.
Audience, LinkedIn
+636 over the trailing year
Aug 21 to Sep 17
1.15% over the last 28 days
By seniority
By location
Account figures as of Sep 17, 2026, from the LinkedIn export. Top industries: Technology and Internet 25%, Software Development 18%, IT Services and Consulting 8%. 209,384 impressions and 88,363 members reached over the trailing year; 38,480 impressions and 21,622 members reached in the last 28 days.
Voice profile
Two items are flagged for Matthew and deliberately not written. First, the Sep 14 Greg Raiz post is the second instance of the cold-open scene with named dialogue at the top of the corpus, after Kaufer; the profile calls that shape promising on one data point, and two is still short of the threshold for a rewrite. Second, for the second week running, the profile's cross-platform line still reads "Substack 0.64 engagements per note" against an actual 11.7 this week - a distribution finding, which the profile itself separates from a voice finding.
Recommendations
2. Write more cold-open scenes with a named person and real dialogue. Greg Raiz is the second time that shape has produced the week's best post, after Kaufer. Two is not proof, and a scene is never manufactured - but when the raw material contains a real exchange, that is the shape to reach for.
3. Restore cadence and put marketplace mechanics back into it. Three masters against five, four publishing days empty, and three of the six themes at zero including the one the profile calls his sharpest angle.
Two posts published before this window are still drawing engagement: the Sep 7 founder-interview post at 21 and the Sep 3 Etsy post at 6. Ledger coverage: 110 posts on file, 20 carrying a format and topic tag, 8 carrying an engagement rate, and those two sets do not yet overlap. The one decision Matthew owns this week is whether the cross-platform Substack line in the voice profile gets corrected, which sits outside this run's write authority. Rendered from the stored weekly scorecard, week ending Sep 18, generated 2:20 AM ET. Account figures in the Audience panel are export-dated Sep 17 and are not from this week's post pull.