Co-Viewing: Why Streaming Reach Is Bigger Than the Impression Count

What co-viewing actually is

Co-viewing is the industry term for more than one person watching the same screen at the same time — a household watching a show together rather than each person on a separate device. It matters for streaming and connected-TV specifically because the ad-serving unit is the device or the authenticated household, not an individual tracked person the way a mobile app impression often is. One ad serve to a living-room TV during a co-viewed program can reach several people at once, and none of that additional reach shows up as additional impressions in a standard delivery report.

Why this is a live measurement question right now

Streaming has become too large a share of total TV viewing for this to stay a footnote — streaming captured a record 47.5% of total TV viewing in December 2025 and 46.6% of ad-supported household viewing in the first 90 days of 2026, according to Nielsen's own published data. Nielsen has separately been rolling out updated co-viewing measurement, layering a "Big Data + Panel" approach across tens of millions of households and folding further updates into its methodology from September 2026 onward. The industry-wide effort to measure this more precisely is itself the strongest evidence that co-viewing isn't a rounding error — it's large enough that better measurement of it has become a genuine methodology project for the company that measures television for a living.

What this means for reading your own numbers

An impression count from Amazon DSP tells you how many times an ad was served to a device or household — it does not, and structurally cannot, tell you how many individual people were in the room. That means the reach and frequency figures in a standard DSP report are conservative by construction: real audience exposure on shared-screen streaming inventory is very likely higher than the reported impression count implies, not lower. This cuts against a common instinct to treat CTV reach numbers with suspicion because they seem smaller than a brand's mental model of "everyone watching TV" — the reported number is probably understating things, not overstating them.

It's also part of why streaming's new-to-brand rate reads as strong as it does in our own accounts. Across 27 advertisers in our own Amazon DSP book over a 31-day window this summer, streaming and connected-TV inventory carried the highest new-to-brand rate of any format on the report, 56.3%, against a 0.77x last-click return. A co-viewer who never generated their own tracked impression can still be the person who searches the brand later — invisible to the delivery report, present in the new-to-brand number the following week. reMKTR runs 109 live Amazon DSP advertiser seats and reads streaming's new-to-brand pattern as one of the places co-viewing's invisible reach most plausibly shows up in the data that does exist.

What to do with this, practically

Don't try to apply a specific multiplier to your own impression counts — there is no single, reliable, publicly verified co-viewing ratio to apply per household, and inventing one would be exactly the kind of confident-sounding but unsupported figure worth avoiding. What's actionable is qualitative: when comparing a streaming reach number against a single-viewer digital format like mobile in-app video, remember the comparison isn't apples to apples, and factor that structural undercount into how conservatively you read a streaming reach report relative to other channels in the same plan.

The same caution applies to any secondary source that does offer a specific co-viewing multiplier with confidence. Precise per-household ratios circulate in industry commentary, but Nielsen's own measurement of this is still actively being overhauled as of this writing, which is a strong signal that the industry itself doesn't yet consider the number settled — treat a confidently-stated multiplier from a source other than Nielsen's own current methodology with real skepticism.

The common mistake

The mistake is treating a streaming impression count as a precise, literal count of individual people reached, then comparing it unfavorably against a channel that tracks individuals directly, like a logged-in mobile app. The two numbers aren't measuring the same thing, and judging streaming reach as weak because its impression count looks smaller than a social platform's user-level reach figure ignores that one number is a conservative floor and the other is closer to a precise count. reMKTR presents streaming reach data with that context attached rather than as a bare number sitting next to channels measured on entirely different terms.

The opposite mistake is also worth naming: using co-viewing as a rhetorical escape hatch to explain away every weak result. It supports a real, structural argument about how impression counts should be read, not a blanket excuse for a campaign that also shows poor completion rate and no branded search movement — those two signals aren't affected by co-viewing at all, and they should still be trusted as the honest read of whether the creative and targeting are actually working.