Streaming · Registered before retrieval
Netflix's 2026 slate opens at twice the demand of its 2023 slate. Its audience is still shrinking.
Bloomberg reported falling engagement. We registered that the catalogue was thinning, pulled the data, and the prediction missed by a factor of two.
The slate is not thinning. That was our prediction and it missed.
Bloomberg reported on October 25 that Netflix subscribers are watching less and opening the app less often. Jefferies, via Reuters, puts average monthly viewing at 66 hours in the first half of 2023 and 49 in the first half of 2026. We registered a question before pulling anything: is the top of the slate thinning, and does search demand show it first?
It is not, and the data says so by a factor of two. Taking the top ten series by hours from Netflix's own engagement report in each half, and the highest search month each title reached in its release window:
| Cohort | Median opening month | Against 2023 |
|---|---|---|
| H1 2023 (10 titles) | 368,000 | — |
| H1 2026 (10 titles) | 748,000 | 2.03× |
| H1 2026, three broken terms removed (7) | 823,000 | 2.24× |
The result is stronger than it looks. Seven of the ten 2026 terms are title + netflix constructions, while the 2023 cohort leans on bare franchise terms like bridgerton and outer banks that collect every season at once. The handicap runs against the 2026 cohort and it still came out at double.
What did change, and it is not only Netflix
| Platform | 2023 average | Last 12 months | Change |
|---|---|---|---|
| youtube | 165,166,667 | 223,416,667 | +35.3% |
| netflix | 13,391,667 | 10,863,333 | -18.9% |
| disney plus | 3,720,000 | 4,132,500 | +11.1% |
| hulu | 5,953,333 | 3,882,500 | -34.8% |
| prime video | 3,146,667 | 3,447,500 | +9.6% |
| hbo max | 3,380,833 | 3,126,667 | -7.5% |
Netflix's share of this six-platform set fell from 6.88% to 4.37%. Most of that is the denominator growing. Netflix is not uniquely declining — it is declining less than Hulu. The one service holding its own peak is Prime Video, and nobody is writing about that.
A limit on every percentage above. Google bands these figures. The Netflix series only ever reads 9.14M, 11.1M, 13.6M, 16.6M, 20.4M, 24.9M. YouTube's apparent step in April 2025 is exactly one band. The directions are sound; the decimal places are not.
The first run used the bare word max, which is a common English word. It reported −43.7% and is not in the table above; the row shown is a re-run on hbo max at −7.5%. The wrong number is named here rather than quietly replaced.
The control, and the half of it that worked
Ten non-Netflix originals from the same half of 2023 — HBO, Apple TV+, Prime Video, Disney+ — selected by a rule written before any title was looked up: a title enters only if its bare search term has no large pre-existing referent.
| Cohort | Median | n |
|---|---|---|
| Netflix | 12.2% | 10 |
| Non-Netflix | 13.4% | 10 |
The 2023 baseline is a category number, not a Netflix number. That is the control doing its job. Checking its spike shapes against the Netflix cohorts is also how the confound named in the limits below surfaced.
The 2026 half of the control is not closed. No publisher releases a non-Netflix new-originals ranking by half, so only three titles could be named with confidence, and all three are low-spike titles that do not test the band where the question lives. Three is below our own five-respondent floor and it settles nothing.
Every prediction, registered before the pull
| # | Prediction | Band | Basis | Outcome |
|---|---|---|---|---|
| N1 | The 2026 cohort's median opening demand is below the 2023 cohort's. | likely · 55–80% | declared belief | Missed. The 2026 median opening is 748,000 against 368,000 in 2023 — 2.03×, and 2.24× once three broken search terms are removed. The registered falsifier fired on its own terms. |
| N2 | The 2026 cohort's spread is wider than the 2023 cohort's. | likely · 55–80% | reasoned inference | Holds, and the hold is worthless. It holds because one term, ms rachel netflix, returned 1,300 searches at peak. Nobody searches that phrase. A single floor value set a ratio. Scored as registered and logged as a defect in the registration. |
| N3 | Netflix's share of a six-platform set is lower than in 2023. | likely · 55–80% | reasoned inference | Holds. 6.88% of the six-platform set in 2023, 4.37% across the last twelve months. |
| N4 | YouTube exceeds Netflix in every month measured. | very likely · 80–95% | measured | Holds. The calibration row. Had it failed, nothing else on this page would be worth reading. |
| N5 | A 2026 title exceeds the 2023 cohort's highest opening. | roughly even chance · 45–55% | declared guess | Missed. No 2026 title cleared the 2023 maximum. |
| N6 | Two independent endpoints agree on Netflix's rank. | very likely · 80–95% | reasoned inference | Holds. Historical series and a live share-of-search snapshot both place Netflix second of six. |
Four of six. By basis: measured 1 of 1, reasoned 3 of 3, belief and declared guess 0 of 2 — the pattern the calibration ledger reports across 57 priors, reproduced inside one pass.
What this cannot see
- Search demand is not viewing. A title watched off the recommendation rail without a search is invisible here, and that surface is exactly what Netflix built.
- No impression, no CPM, no rate card was observed. Anything this implies about what inventory is worth is reasoning, not measurement.
- United States and English only. Netflix is a global service.
- Netflix's engagement reports are self-published, and the company has said it is moving from twice-yearly to annual reporting.
- No 2026 retention figure is given, because the measure does not support one. Retention divides a later month by the opening peak, so a title that spikes higher retains less as arithmetic rather than as behavior (r = −0.53 across 28 titles). The 2026 cohort spikes 89× above its baseline against 11× in 2023, so the two are not comparable on it, and a measure that nets the baseline out shows no difference at all. The 2023 figure above is a like-for-like comparison against the control and stands on its own.
- The two cohorts are not symmetrically specified. Bare franchise terms in 2023, mostly qualified terms in 2026. We tried to fix this with season-qualified terms throughout and the screen rejected the entire 2023 cohort — a 2023 title's own opening sits inside the pre-2025 window the screen treats as contamination.
This page is re-run when Netflix next reports, and the decay question re-opens if a spike-independent measure is built.
Sound Decisions carries each re-run and each correction, with the registered question written before the data is pulled.
Subscribe to Sound Decisions