Frank would like it noted that he predicted this in 2019, to a man in a Rosemont hotel bar who did not ask.

Four weeks before the NFL season opened, Nielsen reminded the networks of a rule it says was always there. Combining Nielsen figures with a network's internal data, or with a third party's, to make comparative claims in the press is, in Nielsen's words, "not permissible."

The reminder followed a summer argument at the FIFA World Cup. Fox issued viewership releases using Nielsen-only figures. NBCUniversal's Telemundo issued releases built on Total Audience Delivery, a blend of Nielsen Big Data + Panel for television and Adobe Analytics for streaming. Telemundo also promoted its growing audience among viewers who don't speak Spanish. The two sets of numbers then circulated side by side, as though they described the same thing.

NBC had long defended the blend. Its argument was that Nielsen did not measure Peacock streaming, so a Nielsen television number plus an Adobe streaming number was the closest available equivalent to what other networks reported. That argument had an exception. For the Super Bowl, Nielsen did measure the streaming audience, and NBC used no Adobe blend.

Fox welcomed the reminder publicly. Mike Mulvihill, Fox Sports' president of insights and analytics, wrote that he hoped the policy would be "rigorously enforced during football season."

The policy covers comparative claims in press releases. It does not cover what a network measures internally, what it shows advertisers in private, or how two Nielsen numbers built from different inputs get compared to each other.

"Not permissible." Not "wrong." Not "misleading." Not permissible. That's the word they use for cargo shorts at a country club.

NBC Sports stopped reporting Total Audience Delivery for NFL games beginning with the Patriots-Seahawks kickoff on September 9, a rematch of February's Super Bowl. It now reports a Nielsen-only figure that includes NBC's first-party streaming data. Adobe Analytics still measures Peacock internally, with Nielsen auditing layered on top. FreeWheel continues to measure the streaming audience for monetization.

NBC is treating the change as a trial running through January. It applies to NFL games and MLB wild-card coverage. College football keeps TAD internally, without public comparisons to other broadcasters, and NBA measurement for the new season has not been decided.

The kickoff averaged 25.1 million viewers. Last year's kickoff, Cowboys-Eagles, averaged 28.3 million. That 28.3 million was two numbers added together: 23.3 million Nielsen-measured linear viewers and 4.9 million streaming viewers tracked by Adobe.

The headline comparison is a decline of about 11 percent, and trade coverage called it the least-watched kickoff since 2022. ADOTAT's arithmetic shows why neither comparison available holds up. This year's 25.1 million is 3.2 million below last year's blend. It is also 1.8 million above last year's Nielsen-only linear figure. The first comparison sets a single measurement against a sum of two vendors. The second sets a number that includes streaming against one that does not. No like-for-like comparison has been published.

Sunday Night Football shows the same pattern. The Cowboys-Giants premiere averaged 25.7 million, reported as up 4 percent on last year's 24.7 million, and last year's figure was also Nielsen plus Adobe.

NBC officials expect the Nielsen-only method to produce higher NFL numbers, and argue they will also be more accurate. Both of the season's first year-over-year comparisons on NBC set a Nielsen-only number against a blended one.

Down 11 percent, or up 8, depending which ruler you like that morning. I watched a station group change ruler vendors in May once and hold a sales meeting in June titled "Momentum." The coffee was good. I'll give them the coffee.

On August 31, Nielsen expanded its co-viewing methodology, the part of measurement that estimates how many people are watching a given screen, and updated the weighting formulas inside Big Data + Panel. Trade coverage counted it as the third major pivot in Nielsen's methodology in less than two years. It went live nine days before the NFL season opened, at the start of a fall calendar that also carries college football, the MLB and WNBA playoffs, and the NBA and NHL openers.

The NFL objected before the change took effect. Paul Ballew, the league's senior vice president of data and analytics, told Front Office Sports the changes were "overly rushed." He said they would leave this season's measurement muddled, called the relationship with Nielsen a real struggle, and said the early results looked negative for the NFL and for sports generally. He put properly measured co-viewing at least two years away. The league said it is expanding its work with rival measurement companies.

The complaint is not new. A year ago, Ballew told The Wall Street Journal that Nielsen was systematically undercounting millions of NFL viewers. Nielsen replied that Big Data + Panel made that season the most accurate yet. The NFL then finished the season averaging 18.7 million viewers per game, the second-highest average since 1988. ESPN's coverage noted that part of that increase came from the change in how viewers were counted.

NASCAR resumed reporting Nielsen Big Data + Panel figures in September, timed to the modified methodology Nielsen rolled out at the end of August.

Every Week 1 audience this season was measured under a method that had been live for less than two weeks.

Third pivot in two years. My mother moved the living room couch less often than that, and she moved it every time my uncle came over, because of the stain. The stain isn't relevant. The boat isn't either. Put a second coat of varnish on the transom Sunday. It ran.

The 2026 NFL Melbourne Game, 49ers versus Rams, was the first regular-season NFL game played in Australia. Netflix, the league's U.S. rights holder for the game, streamed it on a Thursday night. The NFL announced on September 14 that it averaged 18.5 million viewers in the U.S. according to Nielsen Big Data + Panel.

The release carried the supporting figures. Viewership peaked at 21.3 million between 9:30 and 9:45 p.m. ET. The game ranked among the top three most-streamed Thursday prime-time NFL games. It averaged 4.0 million viewers ages 18 to 34, which the league called the highest non-Christmas streaming delivery for a regular-season game. Globally, the league reported 21.7 million viewers across more than 200 countries and territories. By ADOTAT's arithmetic, that puts the international audience at about 3.2 million, against 1.2 million outside the U.S. for last year's Brazil game.

The source line at the bottom of the NFL's release reads: "Nielsen Big Data + Panel inclusive of 1st party data."

That line changes what the number is. The Netflix figure is not a panel estimate alone, and it is not Nielsen's big data set alone. It is Nielsen's product with Netflix's own delivery data folded into it. Trade coverage attributed the 18-to-34 figure specifically to Nielsen data combined with first-party data from Netflix.

Folding platform data into Nielsen measurement is now standard, and Nielsen sanctions it. Amazon's Prime Video has reported NFL audiences on that basis. It is the arrangement NBC just adopted for Peacock. It is also what the NFL demanded. Ballew's complaint a year ago was specifically that Big Data + Panel lacked first-party data from the streamers carrying NFL games, and that the gap would produce undercounts.

Sports Media Watch put the result in context. Melbourne was the least-watched of the five NFL games Netflix has streamed. The other four all aired on Christmas Day, when viewing levels run far higher. None of that makes the number less accurate. It makes it a different number from the ones it gets compared against.

The question here is not whether first-party integration is good measurement. It may well be better measurement. The question is narrower. "According to Nielsen" now covers a family of measurements. The members differ in which inputs each platform supplied, and in how Nielsen treated those inputs. The release names the ingredient. It does not say how much of the audience the ingredient accounts for.

"Inclusive of." Beautiful phrase. Inclusive of what, how much, checked by who? At comScore we had a guy in legal whose entire job was the word "inclusive." Nice guy. Drove a Buick. Never once told me what was in it.

On Friday, September 5, 2025, YouTube streamed Chiefs-Chargers from São Paulo, free and worldwide, as its first exclusive NFL broadcast. It was the league's Week 1 international game, the slot Melbourne filled this year. YouTube also used the game to promote Sunday Ticket, its paid package of out-of-market games.

The first figure YouTube published was a 17.3 million global average minute audience. It broke down as 16.2 million in the U.S., attributed to Nielsen, and 1.1 million outside the U.S., attributed to YouTube. The footnote on the U.S. figure identified the source as "Nielsen. YouTube Custom Analysis," and noted that the data was inclusive of YouTube first-party data.

Rival networks had objected before the numbers came out. On game night, ESPN's senior vice president of research, Flora Kelly, said the approach differed from the one every other rights holder used and lacked MRC accreditation. Her conclusion was that the rating was "not a fair comp." When the figures were released, she said that without transparency into the methodology, the numbers could not be interpreted. Executives at Fox raised the same objection. Trade coverage added a second caveat: Nielsen had been counting expanded out-of-home viewing since February 2025, which lifts audiences relative to earlier years.

Four days later, the number changed. YouTube published an editor's note blaming an internal technical issue. It said "a subset of legitimate views were not categorized as actual views," and that it had revalidated the figures with Nielsen after supplying updated first-party data. YouTube said it would do better next time.

The revised figures were 18.5 million in the U.S., 1.2 million outside it, and 19.7 million globally. By ADOTAT's arithmetic, the U.S. audience rose by 2.3 million, about 14 percent, and the entire change came from data the platform itself resubmitted. The corrected U.S. figure beat the 14.2 million that watched the 2024 Brazil game on Peacock by about 30 percent.

The revision did not end the dispute. Fox and ESPN executives kept objecting to the use of a custom measurement that did not conform to the Big Data + Panel method every other rights holder used. YouTube and Nielsen did not fully explain why the game could not be measured the standard way. That explanation has still not been published.

Two point three million people showed up four days late to a Friday night game and got counted. My brother-in-law tried that at my daughter's wedding. They gave him cake. They did not put him in the headcount.

Set the two releases next to each other and the match falls apart.

YouTube, September 2025:

Netflix, September 2026:

Neither release says what share of its U.S. audience came from first-party data. Neither says how that data was weighted, or what the other game's number would have been under its method. Neither can be restated on the other's basis from anything published.

A reader who sets the two 18.5 millions side by side is comparing a custom analysis with a standard one, across a co-viewing methodology change, with the platform-supplied share undisclosed on both sides. That describes most of the coverage that noticed the match.

The equality holds as arithmetic. As a measurement, it has not been established.

Nielsen's method was not the only variable. Between September 2025 and September 2026:

Every one of those changes may have improved measurement. Every one of them also means a number from September 2026 cannot be read against a number from September 2025 without a restatement. No network, streamer or measurement company has published one.

Five things changed in twelve months and the only thing anybody printed was the number. At Rentrak we changed one input and put out a memo the size of a phone book. Nobody read it either, but it existed, and it was in the lobby, and the lobby had chairs.

This issue draws on the NFL's and Netflix's September 14 releases on the Melbourne game, including the NFL's source line, and on YouTube's September 2025 post with its footnote and editor's note. It also uses public statements by ESPN, Fox and NFL executives, and reporting in Front Office Sports, Sports Media Watch, Sports Business Journal, TV Tech, Awful Announcing, ESPN and The Wall Street Journal. Streaming history in the paid section comes from Dan Rayburn's published work at StreamingMediaBlog.com, plus one statement from an August 26 call that he cleared for attribution. Rayburn says he has access to a lot of data from many companies covered by NDAs. The kickoff comparisons, the international audience figures, the YouTube revision percentage and the Peacock comparison are ADOTAT's arithmetic. The first-party share of either 18.5 million figure is not public, and this issue does not estimate it. Nielsen, Netflix, YouTube and NBC Sports were sent questions before publication on how the figures were constructed and whether they are comparable, and their responses will be added as they arrive. ADOTAT has no financial relationship with Netflix, YouTube, NBCUniversal, Fox or Nielsen. Pesach Lattin consults for clients on ad tech analysis as a standing engagement.

Above the wall, two numbers match. Below it, nobody in the chain is obligated to say what either one is made of.