Can You Compare Streaming Views With TV Ratings in a Pitch?

By pitch.dogAdvanced11 min read20 September 2026

Television

Sometimes. Less often than the shared word "audience" suggests, and never by declaration.

The two figures usually arrive in a producer's hands looking compatible. One is a number of views from a platform report. The other is a rating from a measurement service. Both are audience numbers. Both are countable. Neither of those facts creates a shared basis, because "audience" is a category that both measurements belong to — it is not a unit either of them is measured in.

The answer to whether you can compare them is settled by arithmetic, not by rhetoric. Reconstruct each figure from the report that published it: what went into the numerator, what went into the denominator, which population, which window, which unit. Then ask whether the two reconstructions share a basis. Sometimes they do. Often they don't, and the honest pitch contains two claims instead of one comparison. That is a weaker sentence and a stronger pitch.

Build a source card before you build a comparison

A source card is a small list of fields, filled in from the document itself rather than from memory of the brand:

Publisher and release date. Who published the number, and when.
Observation period. The clock time the figure covers.
Geography. Which territory.
Basis. Households, people, accounts, plays, hours.
Unit and universe. The precise unit, and the population it is a share or multiple of, if any.
Viewing window. What counts as watched, and over how long.
Title scope. One episode's first telecast, a season, a whole catalogue.
Exclusions. What the report says it left out.
Measured or modeled. Whether the figure was observed or estimated.

Two fields on that list do most of the work, and neither is the platform's name. A familiar logo tells you who published something. It does not tell you that two releases from the same publisher share a definition, a period, or a title scope — and it certainly does not tell you that a platform report and a measurement service are counting the same thing.

The card is not bureaucracy. It is the only place the mismatch becomes visible before it reaches a slide.

What hours divided by runtime actually reports

Netflix's What We Watched the Second Half of 2025, published January 20, 2026, defines views for the July–December 2025 period as total hours viewed divided by runtime. That is a real, usable definition — inside that report, for that period.

It is not the definition of a streaming view. Other publishers use other rules, and this definition is dated; a later report from the same publisher would need checking rather than assuming. Nor should the definition be extended to all streaming data, or to all of a platform's titles, or backward to earlier releases.

Here is what the quotient actually reports: runtime-equivalent viewing. Twelve thousand hours divided by a two-hour runtime is 6,000. Read that as 6,000 times two hours' worth of watching. Not 6,000 people. Not 6,000 completed plays. The division works because it collapses everything into time, and in collapsing it, it destroys the information that would let you tell those readings apart.

Rewatches sit inside the numerator. Partial viewing sits inside the numerator. Neither leaves a mark on the quotient. The report gives you a real measurement of viewing time — which is a lot — and it gives you a calculation. It does not give you an audience count, and it is not hiding one.

One more dependency: the division needs a single runtime to divide by. A film with a stated two-hour runtime behaves differently in this formula than a season, an anthology, or a bundle whose runtime the publisher expresses as a total.

What a rating's percentage is a percentage of

Nielsen's November 2025 explainer, What are Nielsen ratings?, describes ratings as percentages against a defined household or person population, and identifies several viewing windows. Both halves of that sentence matter.

The denominator. A rating is a share of a named universe. Two percent of 100,000 households is 2,000 households; two percent of forty million is 800,000. Same percentage, different audience. Quoting "it did a 2" without the universe is quoting an unfinished sentence — and a producer who quotes it that way in a pitch has handed a commissioner a number they cannot check.

The basis. Household and person ratings are different metrics even when the digits look identical. A household rating counts households on its own terms. It doesn't hand you the people inside those households, and it doesn't establish how many were watching.

The window. The explainer presents more than one viewing window, which is itself the useful information: the same program measured over a longer window accumulates more viewing. A figure without its window is comparable to nothing — including another figure from the same provider.

The list that has to line up

To build one defensible comparison you need five things aligned:

Unit. Hours, runtime-equivalent views, households, people, accounts.
Population basis. Which universe, and whether it is households or people.
Geography. The same territory.
Period and window. The same stretch of clock time, measured over comparable windows.
Aggregation. One episode's first telecast, or a title's cumulative viewing across six months.

Converting a percentage into a count addresses the first item and nothing else. You have changed a rate into a number of households. Population, geography, period, and aggregation are all still standing where they were. This is why "but I converted it" doesn't finish the argument.

And two numbers don't become comparable because both can be written as whole numbers. Six thousand and two thousand look like the same species of thing once the units are stripped off. That is the work a bar chart performs: it removes the units and asks the reader to assume the axis means one thing. If the axis means two different things, the chart has not illustrated a comparison. It has asserted one.

Two invented cards

The arithmetic below is invented — invented publishers, invented figures, no real dataset behind it. Take it as a rehearsal, not a case.

Card A: streaming. A streaming service's public engagement report. Period: January 1–June 30. Title: a film with a stated runtime of two hours. Reported: 12,000 total viewing hours, under a definition of views = hours ÷ runtime. Derived: 12,000 ÷ 2 = 6,000 runtime-equivalent views.

Card B: linear. An audience-measurement service. Geography: one country. Period: the same January 1–June 30 stretch, covered by a single episode's first telecast. Viewing window: live plus same day. Basis: households. Universe: 100,000 TV households. Metric: a 2% household rating for that first telecast. Derived: 2% × 100,000 = 2,000 households.

Side by side: 6,000 and 2,000. Three to one on a chart. But Card A's unit is two hours of viewing time accumulated across six months, and Card B's unit is one household on a metric's own terms, with its own threshold, in a window that closes the same day. Placed on one axis, those two bars compare nothing to nothing. They merely share a font.

Now watch what Card A's number will not tell you. Those 12,000 hours — and therefore the derived 6,000 — can be produced by:

6,000 people watching once (6,000 × 2 hours = 12,000 hours)
1,000 people watching six times (1,000 × 6 × 2 = 12,000 hours)
24,000 people watching thirty minutes each (24,000 × 0.5 = 12,000 hours)

All three allocations produce 12,000 hours, and all three therefore produce 6,000 views under Card A's definition. The first implies 6,000 viewers, the second 1,000, the third 24,000 — and in the third, nobody finished the film, yet the view count is identical. The report cannot distinguish these cases, because the quotient discarded what would distinguish them. None of these populations should be assumed to have existed; they are the range of possibilities the measurement leaves open.

So the pair cannot be compared, and it cannot be repaired by one adjustment either. To put the two on one axis you would need both in the same unit — two counts of the same kind of entity — for the same geography and period, aggregated the same way. Getting Card A to a person count requires a rewatch assumption the report doesn't supply. Getting Card B to a person count requires a household-size assumption its metric doesn't supply. Two invented multipliers, stacked, and you are no longer quoting either source.

The route that sometimes does work: if both sources publish total viewing time for the same defined population over the same period, hours against hours is a real comparison, and you check the populations and periods before you trust it. That route is available only when the reports actually supply time on both sides. Often they don't.

The claims you can make

When reconciliation fails, write the two claims at the level each source supports.

Streaming: Across the half-year, the title's viewing totaled 12,000 hours, which the publisher reports as 6,000 views under its hours-divided-by-runtime definition.

Linear: In the window the card states — the episode's first telecast, live plus same day — the episode drew a 2% household rating in a universe of 100,000 households, or 2,000 households on that metric's basis.

What has disappeared is "reached more households," "bigger audience," "outperformed." Those sentences require the two figures to share a basis, and they don't. Keeping the words and dropping the comparison is the failure mode: the chart goes away, the conclusion stays, and nobody notices the conclusion was the part that depended on the chart.

Two practical tests are worth applying before a deck leaves the room. First, the one-axis test: if you cannot label a single shared axis with one unit that is true of both bars, you do not have one chart. Two labeled panels with different axes are honest and completely presentable. First, the repeat test: could a commissioner check either claim against the named source without assistance? If the claim includes a conversion you performed, the answer is no, and the commissioner will eventually find that out.

When not comparing is the result

An explicit non-comparison is a legitimate pitch outcome. It is not an absence of evidence; it is a statement about what the evidence is for.

If you compare anyway, the missing denominator has to come from somewhere. It comes from an invented conversion, an assumed average household size, or a completion rate nobody measured. In the deck, that number is indistinguishable from a number that was observed. That is how a measurement problem turns into a different problem — the kind where a commissioner remembers, six months later, that the audience figure they commissioned against didn't exist.

Say what each source establishes. Say plainly that the two figures sit on different bases — that one is a quotient of viewing time accumulated over six months and the other is a share of a defined universe, measured live plus same day. If a commissioner wants a single ranking more than they want a defensible one, they are still entitled to know which of the two they're getting. Handing them the ranking without that sentence is not a favor.

Frequently asked questions

Can a streaming view count and a TV rating ever be compared directly?

Only sometimes, and the decision is settled by arithmetic rather than by how the two numbers are described. Reconstruct each figure from the report that published it — numerator, denominator, population, window, unit — and ask whether the two reconstructions share a basis. When both sources publish total viewing time for the same defined population over the same period, hours against hours is a real comparison. Often they don't, and the honest pitch carries two claims instead of one comparison.

What does "views" mean in a report that defines views as total hours divided by runtime?

Runtime-equivalent viewing. Netflix's report for the second half of 2025, published January 20, 2026, defines views for July–December 2025 that way: 12,000 hours divided by a two-hour runtime gives 6,000. Read that as 6,000 times two hours' worth of watching — not 6,000 people and not 6,000 completed plays. Rewatches and partial viewing both sit inside the numerator and leave no mark on the quotient.

Why is a rating without its universe an unfinished sentence?

A rating is a percentage of a named population. Two percent of 100,000 households is 2,000 households; two percent of forty million is 800,000. The body also separates household from person ratings — different metrics even when the digits match — and notes that the same program measured over a longer window accumulates more viewing, so a figure without its window is comparable to nothing, including another figure from the same provider.

I converted the rating into households, so isn't the comparison valid now?

No. Converting a percentage into a count addresses the unit and nothing else. Population basis, geography, period and window, and aggregation are all still standing where they were. Placing both on one axis would require both in the same unit for the same geography and period, aggregated the same way — which means a rewatch assumption the streaming report doesn't supply and a household-size assumption the rating's metric doesn't supply. Two invented multipliers stacked is no longer quoting either source.

What should a pitch say when the two figures can't be reconciled?

Write the two claims at the level each source supports, and say plainly that they sit on different bases — one is a quotient of viewing time accumulated over six months, the other a share of a defined universe measured live plus same day. Dropping the chart while keeping the conclusion is the failure mode, since the conclusion was the part that depended on the chart. Two tests help: can you label a single shared axis with one unit true of both bars, and could a commissioner check either claim against the named source without assistance? An explicit non-comparison is a legitimate outcome.