What YouTube Studio Won't Tell You: Which Topics Retain Subscribers?

Tuesday morning. You open YouTube Studio and scroll the last 30 days. The dashboard congratulates you. Watch time is up, views are up, your reactions playlist did 124K views on one upload. You screenshot the bar chart and post it in your group chat.

Then you check the subscriber graph. Net subs for the month: a wobble. Up some days, down others. You closed where you started.

The dashboard told you the show is healthy. The subscriber count told you it isn't. Both are right — they're measuring different things, and YouTube Studio refuses to put them in the same view.

The hidden question

Every creator past 10K subscribers eventually asks the same thing: which of my topics actually builds the channel, and which one just rents traffic from the algorithm and gives it back?

YouTube Studio cannot answer this. It shows you views and watch time per video. It shows you subs gained vs lost per day. It will not join those two together at the topic level. There is no view in Studio that says "the people who first found you via topic X — how many are still watching twelve weeks later?"

That is a cohort question, and the standard answer in any other industry — SaaS, ecommerce, podcasts — is a cohort retention table. For creators, it doesn't exist as a feature. So creators optimise for the metric the dashboard shows them: views.

Views are a renting metric. A trending take, a reaction video, a Shorts compilation — those bring traffic that the algorithm hands you and then takes back when the next thing trends. A deep-dive tutorial earns slower but the audience stays. The headline number does not distinguish between the two. The subscriber retention curve does.

What Anna does

Anna takes the question seriously. The work is a five-step joining-and-tagging job — boring in concept, impossible in practice without the right tool.

Step 1. Export from YouTube Studio. Studio exports CSVs for video performance, traffic sources, and subscriber activity. Connect the YouTube data integration, or upload the CSVs directly. Anna sees them as three related tables.

Step 2. Tag every video by topic. This is the step that breaks most spreadsheet attempts. You have 160 videos over twelve months. Tagging them by hand takes an afternoon, the taxonomy drifts halfway through, and you never want to update it. Anna adds a topic column with an =AI() formula on the title and description: =AI("Classify this video into one of: deep-dive tutorial, case study, news & reactions, interview clip, trending opinion, Shorts compilation", title, description). The formula persists in the data. The column lives in your dataset. Re-run it after you publish a new video and the tag is there.

Step 3. Build the subscriber cohort table. YouTube Studio doesn't expose viewer-level events to most creators — what it does expose is aggregate cohort data: returning-viewer counts, retention curves, audience demographics, and comment activity per video. Anna works from that. For each video, she records the topic, the subscribes-from-video number Studio reports, and the cohort retention pattern Studio shows over the following weeks. Stitched across the channel, this gives her a topic-level cohort view — coarser than per-viewer, but honest about what the platform actually shares.

Step 4. Compute net subs by topic. A viewer subscribed via a Shorts compilation and unsubscribed three weeks later contributes +1 then -1. The net is zero. Anna sums it honestly per topic, not the gross-subscribes vanity number Studio shows you.

Step 5. Rank topics by retention curve shape. The shape matters more than the height. A topic that drops to 80% in the first two weeks and plateaus is healthy. A topic that drops to 50% by week eight and keeps falling is rented traffic. Anna labels each topic Build, Rent, or Mixed based on the curve, and produces the report.

What the report looks like

Anna delivers a single URL. Three sections, written in plain English with the data underneath.

The first chart is the cohort retention curve by topic. Each line is a cohort of subscribers grouped by the topic that first earned them. The Y-axis is the percentage of that cohort still watching the channel weeks later.

The shape difference is the entire story. Deep-dive tutorial subscribers drop a normal amount in the first two weeks — some people subscribe, watch one more video, and decide it's not for them — and then plateau around 80%. That plateau is the audience you actually have. Shorts compilation subscribers drop fast, never plateau, and are below 40% by week twelve. Those people did not subscribe to a channel. They subscribed to a feed, and they leave when the feed changes.

The second view is the topic ranking table. Anna shows views (the rented number), net subs over 90 days (the earned number), and the eight-week cohort retention.

Topic	Videos (12mo)	Avg views	Net subs (90d)	W8 retention	Verdict
Deep-dive tutorials	28	42K	+4,820	82%	Build
Case study breakdowns	14	38K	+2,640	79%	Build
News & reactions	41	67K	+1,210	68%	Mixed
Interview clips	22	29K	+940	74%	Build
Trending opinion	19	88K	-310	52%	Rent
Shorts compilations	36	124K	-1,470	43%	Rent

Trending opinion and Shorts compilations are the two highest-viewed topic groups. They are also the only two with negative net subs over 90 days. The audience they bring in unsubscribes within two months. The deep-dive tutorial cohort is the smallest by views and the largest by net subscribers earned. The interview clips line is a quiet winner — third-lowest views, fourth-highest retention.

What an operator does with this

The report does one useful thing: it tells you what to make next.

A creator without this analysis publishes in the order the algorithm rewards. A reaction does big numbers, so reactions get more slots in the schedule. That is the loop that ends in burnout and a stalled subscriber graph.

A creator with this analysis publishes against the cohort retention curve. The decision is no longer "what got views?" but "what topic, when watched, produces a retained subscriber four weeks later?" The schedule starts to look different. Deep-dives go from one a month to two. Reactions go from weekly to fortnightly, and only when there's something to actually say. Shorts compilations come off the schedule entirely or move to a separate channel where the renting-is-fine.

The report URL is the deliverable. You paste it in the team chat. You open it on the third Monday of the month when you're planning the next quarter's slate. You stop arguing about whether the channel is healthy and start arguing about which topics to commission. The data has already done the part you were dreading.

Frequently asked questions

What data do I need to run a topic-level cohort analysis on my YouTube channel?

You need three things from YouTube Studio: the video performance export (title, description, views, watch time per video), the subscriber activity export (subscribes and unsubscribes per video), and the audience report (returning viewers, retention curve per video, basic demographics). Studio gives every creator aggregate cohort data — not per-viewer events, but per-video retention shapes you can stitch into topic-level cohorts. That's enough to run this analysis. The report is more granular the more videos you've published.

How is this different from the "audience retention" graph YouTube already shows me?

YouTube's audience retention graph shows you how long people watch a single video before clicking away. It is within-video retention. A cohort analysis tracks across-video retention — whether a subscriber who first found you on video A is still watching anything you publish twelve weeks later. The two metrics answer different questions and you need both.

My channel is under 10,000 subscribers. Is the cohort analysis still useful?

The shape of the curves is meaningful at any size, but the per-topic confidence intervals widen as cohorts shrink. Under 10K subs, Anna will typically recommend grouping by quarter rather than month, and labelling topics Build or Rent based on the directional pattern rather than precise percentages. The qualitative finding ("Shorts are renting") usually shows up clearly even at small scale.

Can Anna update the analysis automatically as I publish new videos?

Yes. Once the dataset and the =AI() topic-tagging column are set up, refreshing the YouTube export or reconnecting the integration re-runs the tagging on the new rows and re-computes the cohort curves. The report URL doesn't change — anyone you shared it with sees the updated charts on their next visit.