Samsung Says Its ACR Captures Every Pixel. Inside Netflix and YouTube, ACR Mostly Goes Dark.
Samsung, LG and Vizio sell glass-level data, but their own filings, LG's statement and lab tests say ACR mostly goes dark inside the apps that carry a fifth of US TV viewing.
By Pesach Lattin · November 4, 2026
Smart TV makers have a terrific line. Samsung tells advertisers its ACR "captures every pixel being passed back through the glass." Vizio sells "glass-level" data from 28M+ devices. It is a lovely pitch, and like most lovely pitches it gets less lovely when you read the footnotes. This page pulls together what Unmeasured and ADOTAT have reported on what automatic content recognition can see, what it cannot, and who is suing whom over it.
The pitch is every pixel. The filings say otherwise.
Start with the companies' own words, because they are the most useful ones. Vizio's last 10-K says contracts with Netflix and Disney+ restrict it from using viewing data from their content. LG said on Sept. 12, 2026 that its ACR does not operate inside Netflix or YouTube apps. In a 2024 lab study, LG's ACR traffic inside Netflix and YouTube fell to idle levels, and Samsung's US fingerprint server showed none. Those two native apps carry more than a fifth of US TV viewing. So the glass sees a lot. It just blinks at some of the most watched things on it.
EDO CEO Kevin Krim put the condition this way: "If you want our native app on your TVs, then you have to agree to zero data collection when it's active." The full reporting is in Samsung, LG and Vizio Claim to Sell Every Pixel. Inside Netflix and YouTube, They're Blind., with a short version in the ADOTAT brief and a buyer's read in Veritad's note on ACR coverage gaps.
Even when ACR is on, the library has holes
ACR matches what is on screen against a reference library, and a library only knows what it monitors. Nielsen found two ACR partners missed 23% of viewed minutes on stations they did not monitor. That is not a rounding error. That is nearly a quarter of viewed minutes slipping past because nobody was watching those stations, as the ACR investigation lays out.
And when Samsung, LG Ad Solutions, Vizio, Roku, Samba TV and TVision got written questions about where their coverage goes dark, none of them responded. KnotSimpler's Josh Chasin has a question he puts to every ACR vendor, and it is still the right one: "Where are the gaps in your coverage, for both content and advertising, and why?"
Samsung sells the fix for 41 percent duplication
Samsung Ads sells Optimal Reach on a 41 percent duplication figure it says can be fixed. Unmeasured found that same 41 percent appearing as household overlap, repeat impressions and duplicated impressions in three Samsung documents. Those are three different quantities wearing one number. The $141 billion headline turns out to be $9.3 billion in duplicated time multiplied by an assumed $15.16 return per ad dollar. Meanwhile ARF DASH data says in 15 percent of TV homes the bedroom set, often a different brand, is the most watched TV, which is a hard thing for any one maker's glass to see. The issue: Samsung Sells The Fix For 41% Duplication. Nobody Owns The Stop Button.
Roku says 87 percent. It does not say plus or minus what.
Roku and Amazon Ads say brands' other Amazon DSP channels earn 87 percent higher ROAS on days their Roku CTV campaigns run. There is no paper: two blog posts, a one-line source note and a chart pasted in as a screenshot, with no sample size and no ROAS definition. Both case studies compared Roku against paths with audience matching deliberately switched off. Measured's own go-dark geo tests on another CTV platform found an effect of about 3.4 percent. Measured CEO Trevor Testwuide was blunt: "In short, no, that design cannot support the claim." Read Roku Says 87 Percent. It Doesn't Say Plus Or Minus What.
The patent fight: a $92 million judgment and a clean Roku win
Anonymous Media Research Holdings says a Texas jury validated its ACR patents, and Judge Gilstrap entered an amended judgment near $92.3 million against Samsung: $78,512,999 plus about $13.8 million in prejudgment interest. Eleven days after Texas upheld that patent family, Judge Chhabria in California invalidated claims across seven of its patents, including the one carrying 52 percent of the Texas damages, and Roku won outright. Same family, two courts, opposite results. And no firmware change was ordered: Samsung, LG, Vizio, Roku, Hisense and TCL still run ACR. The issue: Samsung Owes $92 Million For ACR Patents Roku Beat Outright.
Then there is the theory that ACR's patents are expiring and the data is about to become a commodity. Three companies push it, one of them an ACR patent holder, and each of them profits if it is true. History is less excited. Nielsen's oldest patent, US 6,647,548, lapsed in September 2016 and no rival currency followed. Its watermarking patents run to at least October 2028, and later patents run to 2033 and 2037. The details: Three Companies Say ACR's Patents Are Expiring. The Oldest One Expired in 2016 and Nothing Happened.
What a buyer should ask before paying for ACR data
ACR feeds targeting, reach estimates and attribution, so a coverage gap does not stay in the TV. It travels into every report built on top of it. The Bondstark diligence desk on ACR and TV data has the full question list. The short version:
- Inside native Netflix and YouTube apps, what does your ACR capture, if anything?
- Which app or content contracts limit your use of viewing data, and which partners do they cover?
- Which stations does your reference library monitor, and what share of viewed minutes falls outside it?
- Can an independent lab test what your sets send to fingerprint servers while an app is playing?
- How do you define a reached household? Netflix's 250 million monthly active viewers counts households that watched at least one minute of ads in a month, per ADOTAT's Netflix yield story and its brief.
The glass sees a lot. The trick is getting the sellers to tell you, in writing, where it doesn't. Reporting by Pesach Lattin; how we work is on the ADOTAT standards page.