How to Measure Substack Articles, Notes and Recommendations Without Mixing the Signals

To compare Substack articles, Notes and recommendations, give each a separate job, measure that job with the appropriate native metric, and then compare what the subscribers acquired through each source do over an equal period. Article views, Note likes and recommendation signups are different units. Adding them together produces a larger number without producing a clearer decision.

My view is that a writer should protect the quality of the newsletter before optimizing the channel that makes the audience counter move fastest. Discovery matters, but its value depends on whether the people it brings want the publication you intend to keep writing. A useful weekly review therefore has two layers: how each channel performed, and whether newly acquired readers subsequently engaged with the work. Narrareach’s Substack analytics can support that review and help turn the findings into the next article and supporting Notes; Substack remains the authority for native subscriptions and billing.

Give each channel a job before choosing its score

An article can introduce a stranger to your work, satisfy an existing subscriber, persuade a free reader to upgrade, or do several of these things. A Note can start a conversation, send someone to an article, or explain what the publication offers. A recommendation introduces readers through another publication’s endorsement. These activities can contribute to one reader’s decision at different moments.

Start with the intended job rather than whichever number is easiest to find. If a Note exists to explain a useful argument, a thoughtful reply may tell you more about its immediate success than a new subscription. If an article introduces a paid series, attributed upgrades are relevant alongside evidence that subscribers opened it. Changing the job after seeing the results makes almost any outcome look successful.

This table is a way to choose evidence, not a requirement to pursue every outcome. A writer building a free educational newsletter and a writer selling specialist analysis have different reasons to distribute their work. Their reporting should reflect that difference.

Understand the units before comparing the totals

Substack’s metrics guide defines article views across web, email and app, including repeated views. Recipients are deduplicated across email and push delivery; open rate concerns recipients who viewed the post, with repeated views by one person counted once. Post statistics also report free and paid subscriptions attributed to the piece.

These definitions create a practical boundary. A post with 2,000 views has not necessarily reached 2,000 separate people. Dividing its 20 attributed subscriptions by 2,000 gives one subscription per 100 recorded views, or 1%. It does not establish that 1% of previously unsubscribed visitors joined. Existing subscribers and repeated viewing can be in the denominator.

For Notes, use the native detail rather than treating likes as the outcome. Substack’s Notes documentation lists attributed revenue, free and paid subscribers, followers, clicks and shares in the individual Note’s statistics. Keep followers separate from publication subscriptions. A reader choosing to see someone’s activity is making a different commitment from subscribing to the newsletter.

Recommendations need their own source record. Preserve the platform’s exact category and referring publication where available. Do not silently combine a named writer recommendation with platform onboarding simply because both happen inside Substack. If you cannot identify the origin, retain the uncertainty rather than assigning the signup to the relationship you most recently worked on.

Why an engagement leaderboard cannot rank acquisition

There is a useful distinction in Narrareach’s recorded Note performance. Across 2,660 posted Substack Notes from 56 writers, published August 17–September 15, 2026, 2,249 had at least one recorded like, while 304 had at least one recorded comment. That is 84.5% and 11.4% respectively, using the latest available observations retrieved September 16.

Those are two different descriptions of the same set. Neither percentage tells us how many subscribers the Notes acquired. The observations have unequal ages, recorded zeros can reflect incomplete collection, and the aggregate has no click or subscription fields. It is a scoped product observation, not a benchmark for every newsletter.

The decision it supports is simple: a ranking built from likes and comments answers an interaction question. It cannot become a subscriber ranking merely because acquisition is the goal you care about. A Note with many likes deserves examination for the idea or presentation that resonated. You still need the native subscription evidence before claiming that the format grows the newsletter.

An engagement score can be useful for finding candidates to study. Keep its ingredients visible and avoid calling it “growth” or “reader quality.” Weighting a comment five times more heavily than a like expresses your preference; it does not measure five times as much commercial value.

Use two clocks: content age and subscriber age

A fair review needs more than a shared date filter. A Note published yesterday and an article published three weeks ago have had different opportunities to accumulate activity. Meanwhile, a subscriber acquired yesterday has had fewer chances to receive your newsletter than someone who joined at the beginning of the month.

For content, choose a repeatable observation age. Seven days after publication is a reasonable starting convention for a weekly review, not a platform rule or an ideal growth window. Compare seven-day observations with other seven-day observations. Add a later review for articles that continue attracting search traffic, rather than judging an evergreen guide entirely by launch week.

For subscribers, group people by acquisition source and joining period. Give each group the same elapsed time before judging later activity. If you publish weekly, also note how many issues each group could have received. A group that saw one issue and a group that saw four do not have equivalent opportunities, even if both appear in a calendar-month report.

Substack’s subscriber dashboard guide describes subscriber source, joining information, activity and filtering. Its activity measure includes email opens or web views in the last month. That rolling measure is useful, but it is not automatically a fixed 30-day follow-up from each person’s signup. If the available report cannot produce the comparison you need, describe the result as a rolling snapshot and avoid claiming a precise acquisition-cohort retention rate.

A worked comparison: volume and later activity can disagree

Consider a constructed reporting exercise. Over one acquisition period, Notes introduce 120 subscribers, named recommendations introduce 80, and a group of articles introduces 40. Assume you can observe all three groups over the same subsequent period and define activity consistently. Of those subscribers, 36, 40 and 24 respectively show the selected activity.

Notes win on acquisition volume. Articles have the largest active share. Recommendations have the largest absolute count of subsequently active subscribers. Those statements can all be true. They answer different questions, so a single winner would hide information you need.

Now add effort. Suppose the distribution work took six hours for Notes, five for recommendations, and eight for the articles. The observed active-subscriber counts per hour would be six, eight and three. That calculation can help allocate the next small block of distribution time, but it is not a return-on-investment verdict. An article can serve existing subscribers and attract readers later; relationship-building can benefit future recommendations. Costs and benefits do not all land in the same reporting window.

Record the effort you actually controlled. Do not charge the full cost of an essential newsletter issue to acquisition if you would have written it anyway. Conversely, do not call recommendations free if you spent several hours finding suitable publications and building a relationship. The purpose of the calculation is to reveal a tradeoff, not to manufacture precision.

Diagnose a weak cohort before abandoning its source

Suppose Notes produce many subscriptions and limited later activity. One possibility is that the Notes promise a different experience from the newsletter. A humorous observation may attract people who do not want a detailed weekly analysis. Another possibility is that the welcome message does not make the publication’s subject or frequency clear. Or the new group may simply have received fewer issues.

Inspect the path before blaming the entire channel. Read the acquisition Note, the publication description, the welcome email and the next issue in order. Would a new reader recognize the same subject and benefit throughout? If the Note offers a practical budgeting tip and the next issue is a long essay about unrelated business news, the discontinuity is a useful explanation to test.

Apply the same scrutiny to recommendations. An endorsement from a large publication can bring readers with little interest in your specific promise. A smaller publication with closely aligned readers may be more useful. Audience size is an input, not proof of fit. Ask whether the recommendation describes the work accurately and whether the first issue delivers that description.

Low public interaction is not proof that nobody reads. Email readers can consume the issue without visiting the app or leaving a like. Use the activity and subscription evidence appropriate to your goal, while remembering that an open still does not establish careful reading. Replies can explain a difficulty that aggregate counts leave invisible.

Keep overlapping influences out of the sum

A reader might see a Note, open an article, encounter a recommendation and subscribe later. Native attribution gives you a reported outcome according to the platform’s system; it does not reconstruct every influence on that person. Therefore, do not add all content-level and source-level subscription counts together unless the platform explicitly establishes they are mutually exclusive.

Keep one publication-level acquisition total as the reference, then examine content and source breakdowns as separate views. If the breakdowns differ, check their periods, free versus paid status, and whether the report concerns an original subscription or an upgrade. A paid upgrade can happen long after the free subscription, so the upgrade source need not answer the original acquisition question.

For unresolved categories such as Direct, preserve an “unknown origin” portion. The traffic-source guide explains why missing referral information is different from evidence of deliberate navigation. No analytics tool can confidently recover an origin it never collected.

Turn the comparison into the next week’s work

Use Narrareach’s Substack analytics workspace to review the available article, Note and audience evidence with the appropriate publication scope selected. Choose one pattern that warrants editorial work: an article that attracts subscriptions, a Note topic that prompts useful questions, or a source whose new readers seem poorly matched to the continuing publication.

Write a decision before drafting. For example: “The next issue will answer the unresolved question in this article, and its supporting Notes will clearly describe the weekly publication.” Draft that issue and two or three Notes addressing different aspects of the same subject. Review each Note for a complete useful idea, an accurate link and a promise the article actually fulfills. Then publish or schedule the approved material and record when you will review it again.

Narrareach can support the move from measurement into drafting and scheduling; it does not make an interaction total equivalent to subscriptions or replace Substack’s subscriber and billing records. Check native detail whenever the imported data lacks the field your decision requires. For individual Note attribution, use the dedicated Notes measurement method rather than estimating conversion from nearby list growth.

At the next review, keep the same definitions and observation ages. Change one part of the acquisition promise or welcome path, and retain the raw counts beside the percentages. With small groups, one reader can move a rate substantially: three active people out of ten is 30%, while four is 40%. That ten-point difference is only one person, so wait for more evidence before reorganizing the whole publication.

For your next report, create three rows—articles, Notes and recommendations—and give each one intended job, one matching native measure and one later reader outcome. The row that helps you choose a specific piece of work is the useful one. Leave any field you cannot observe empty instead of filling it with a guess.