The Curated Peer Recommendation Loop
Grow supply-side discovery by letting creators recommend each other, not by recommending for them.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 95%
Instead of an algorithm suggesting content, Substack lets each writer hand-pick a handful of other writers, and surfaces those picks at the moment a reader subscribes. Because the recommendation is curated by a person the reader just chose to trust, intent stays high; because being recommended is flattering and notified, a goodwill viral loop forms between creators. This is a repeatable growth architecture for any platform where supply already drives demand.
Origin
Built at Substack in 2022 by a team spearheaded by PM Dane Rathbone with designer Gabriel, described here by Head of Product Sachin Monga. It was conceived as the deliberate anti-pattern to Facebook's People You May Know (PYMK) unit, which Monga worked alongside at Facebook from 2011.
Core principles
- 01Look for the discovery behaviour users are already doing organically, and productize that — don't invent a new one.
- 02Curation by a trusted person beats an algorithm on intent, even if it loses on volume.
- 03Place the recommendation at the moment of maximum trust — the instant someone commits to a creator.
- 04Notify the recommended creator: the flattery is the viral loop.
- 05Recommendations are not a step in a flow, they are a graph of goodwill and influence that can be mined later.
How to run it
- 1
Find the organic cross-discovery already happening
Audit how users currently find new supply on your platform without you helping. At Substack it was guest posts, and readers clicking a commenter's profile to see what else that person subscribes to — always through the lens of a writer the reader already trusted.
Pro tip The common shape of the organic behaviour tells you what the productized version must preserve. Here, every organic path was writer-mediated, so the built version had to be writer-mediated too.
- 2
Confirm you have supply density
The loop only works when there are enough creators and enough collective audience that cross-pollination has somewhere to go. Substack shipped this only after years of supply-side growth had accumulated both writers and a large collective reader base.
Watch out Ship this too early and creators recommend a thin pool, readers see irrelevant picks, and the feature discredits itself.
- 3
Ask creators one simple question
Build the minimum viable version: ask each creator who they recommend, and let them pick freely. No algorithm, no ranking, no platform override.
Pro tip Keep it opt-in. The whole value is that a human deliberately chose these names.
- 4
Surface picks at the moment of commitment
Show the recommendations inside the subscribe flow — right after a reader has chosen to trust a creator. That trust transfers to the picks.
Watch out Recommendations shown at first contact convert lower-intent readers than those shown to long-standing subscribers. Monga flags this as the known limitation and the next surface to expand into.
- 5
Close the goodwill loop with a notification
Email the recommended creator: here is who is recommending you, and here are the readers they are sending you. This makes recommending feel like giving a gift, and being recommended feel like receiving one, driving reciprocal adoption.
Pro tip This notification is what turns a one-off feature into a viral loop. Do not treat it as a nice-to-have.
- 6
Treat the output as a graph, not a flow step
The accumulated recommendations form a social graph of influence and goodwill. Plan the next phase of surfaces (long-standing subscribers, reader app, cross-network placements) against that graph rather than against the single subscribe-flow placement.
Pro tip Surfacing a creator's picks to their loyal, years-long subscribers should produce far higher-intent conversions than first-contact placement.
In the wild
Lenny Rachitsky picked roughly ten newsletters he genuinely rated. Around 500 other newsletters recommended him back. The day the feature launched, his growth chart turned into a hockey stick.
→ 70% of Lenny's subscriber growth came from this single feature, with open rates dropping only slightly — evidence the recommended readers were still meaningfully high intent.
Recommendations rolled out across the writer base, driving discovery entirely through human curation rather than an algorithmic feed unit.
→ Millions of new subscriptions across tens of thousands of unique writers; more than one in three new subscriptions and around one in ten paid subscriptions across Substack now come from the network.
Common mistakes
Killing the idea because too many things have to be true
The founder's initial objection was that the feature required writers to opt in, pick good people, and generate enough surface area — a long chain of dependencies. That chain is a reason to run a pilot, not a reason to default to the algorithmic version, which in fact took off faster than expected.
Assuming curated referrals bring junk traffic
Recommended subscribers are lower intent than direct ones, but only slightly. Judging the channel by a fear of low quality rather than measuring open rates and conversion leaves most of the growth on the table.
Letting the platform pick the recommendations
An algorithmic unit inserted into a creator's space breaks the trust chain that makes the recommendation convert at all — you get volume with no transferred trust, and you annoy the creator whose space you took.
Is it for you?
Best for
Marketplace or creator-platform growth teams with dense supply, where creators already have direct audience relationships and supply naturally drives demand.
Not ideal for
Thin-supply platforms, anonymous consumption products with no trusted creator relationship, or categories where creators view each other as pure zero-sum competitors.
From the transcript
“most obvious maximal way to just put writers in control? What is like the simplest version of this? What if we just ask writers, who…”
“But we tried it and it took off really quickly and it there's this like virality at play now where like when you recommend a…”
“more than one in three new subscriptions across Substack are coming from the Substack network, and around one in 10 paid subscriptions now, too.”
“the fact that 70% of my user growth comes from this feature, and my open rates have only come down a little bit”
From the episode
Building Substack
Sachin Monga (Substack, Facebook)