A household of three sits on a sofa and watches one video on one television. Until last week, YouTube's analytics recorded that as a single view. As of September 5 it can record it as three, but only in a report the creator sees and nobody else can check.
Luis Rijo
That change arrived in the same week that Nielsen cut the qualifying threshold for local television tuning by 80 percent, that Google described a conversion metric which keeps counting sales for 180 days after a click, and that FreeWheel published survey data showing media buyers overestimating viewer hostility to artificial-intelligence-generated advertising by 38 percentage points. Four separate announcements, four separate companies, one shared subject: what counts as a countable event, and who is permitted to see the count.
The week also produced a policy change at Google that bans an entire class of third-party integration from the Ads API, and a class action in San Francisco that treats a beauty retailer's influencer programme as a component of an unfair business practice rather than as marketing that happens to sit near one.
The view that multiplies, and the audit that cannot follow it
YouTube introduced views co-viewed on September 5, 2026, and PPC Land covered it the following day. The metric estimates the total number of people watching on televisions rather than counting each device as one view, using statistical modelling built on demographic patterns, video genre and viewing times to predict when people are watching together.
The mechanics are modelled, not observed. No sensor counts the people on the sofa. The system infers group viewing from the characteristics of the content and the moment it is played, which is a reasonable inference for a Sunday evening cooking video and a much weaker one for a niche technical explainer watched at two in the afternoon. YouTube has not published the model, its inputs beyond those three categories, or any confidence interval around the estimate.
Three constraints define what the number can and cannot be used for.
The first is timing. The metric requires up to 48 hours to process, which systematically deflates any reporting window that runs to the present day. A creator opening the default last-28-days view, or this-month, sees a figure that is structurally incomplete at its most recent edge. That is not a bug so much as a property, but it is a property that produces a wrong number every time somebody reads the default.
The second is money. Revenue sharing, AdSense payouts and Partner Programme eligibility continue to run on engaged views and qualified views. Co-viewed counts feed none of them. A creator whose co-viewed figure triples does not earn more, and an advertiser buying YouTube inventory through Google Ads does not reach more people because the estimate rose.
The third is visibility, and it is the one that matters commercially. The figure appears only in the creator's own YouTube Analytics. It is private, unaudited, and covers organic views of the video itself rather than any advertisement served against it. What it can do is appear in a sponsorship deck. A creator negotiating a direct brand deal can now quote a number three times larger than the public view count on the same video, sourced to the platform, with no route for the brand to confirm it beyond asking for a screenshot.
That distinction matters, because media buyers are not short of co-viewing data on the inventory they actually purchase. Google rolled co-viewing metrics into the planning and measurement tools of both Google Ads and Display and Video 360 by the end of the second quarter of 2022, citing Nielsen research that multiple adult viewers watch YouTube together on the television screen 26 percent of the time. Four years later the definition tightened: on June 2, 2026, seven campaign-level fields in Google Ads moved to a Total Co-view basis, among them unique users, average impression frequency per user, and the five unique-users-two-plus through ten-plus threshold variants. Those now count every individual who viewed the advertisement on a connected television rather than device-level accounts, and they appear in the Google Ads API, the interface and Editor. Historical data was not restated. A DV360 buyer running YouTube connected television inventory has had a co-view number for years, and a Google Ads buyer now has a redefined one.
So the gap is not between buyers and co-viewing data. It is between two different transactions. Programmatic and biddable purchases of YouTube inventory come with a co-view figure the buyer can pull from a report, alongside Nielsen's wearable-fed calculation from August 31, 2026 and TVision's sensor-based measurement of presence and eye contact, neither of which agrees with Google's or with each other. Direct sponsorship deals, negotiated between a brand and a creator outside any buying platform, now come with a fourth figure that exists only inside the creator's account. Four estimates of the same living room, three of them auditable in some form, one of them quotable and not checkable. Co-viewing has been a measurement problem in television for as long as television has had more than one chair in front of it, and the industry's answer has generally been to argue about methodology in public. This one moves the argument behind a login, in the corner of the market that has the least measurement infrastructure to begin with.
Two smaller YouTube changes shipped alongside it. Amazon product tagging expanded, still limited to creators in the United States, and the branded content controls updated again, with automated detection of undeclared sponsorship promised in the coming months. The tagging change puts commerce closer to the video. The disclosure change puts enforcement closer to the creator. The co-viewing metric puts the audience number further from the brand paying for the sponsorship. All three landed in the same release notes.
Nielsen lowers the bar for what counts as watching
Nielsen made a smaller change on August 31, 2026 that will move more money.
The company cut its local television tuning threshold from five minutes to one minute, an 80 percent reduction in the time a set must stay on a station before that tuning is credited to a quarter hour. Nielsen's own framing is that the five-minute standard left upwards of 24 percent of tuning events unreported. A quarter of detected local viewing was being discarded because it did not last long enough to qualify.
The distinction between undetected and unqualified is the whole story here. Nielsen's meters were already seeing those events. The methodology threw them away. Lowering the qualifier does not add a single new household, a single new device or a single new streaming source to the panel. It changes the arithmetic applied to data the company already held, and it changes it across all designated market areas at once, in the middle of negotiations for the 2026/2027 broadcast season.
Short tuning events are not distributed evenly. Channel surfing, news headline checks, sports score glances and the first minute of a programme somebody abandons all cluster in particular dayparts and particular genres. Stations whose audiences behave that way stand to gain more than stations whose audiences settle in. Nielsen published no per-market lift estimates, which leaves every local buyer and every station group to discover the shape of the change in their own numbers.
Three further updates activated on the same date. A revised household demographic assignment model, intended to remove bias in how demographics are attributed within a home, rolls into both the local and national systems. Portable People Meter data, the wearable technology Nielsen uses for radio and cross-platform work, now feeds co-viewing calculations. And a Spanish language universe estimate combining American Community Survey data with the National Hispanic Television Survey takes effect on September 26, 2026.
That last item is a universe estimate change, which is the most consequential kind of measurement adjustment and the least discussed. Universe estimates set the denominator. Changing one moves every rating and every impression figure derived from it, for every advertiser buying that population, without anybody's viewing behaviour changing at all.
The local threshold change also arrived in the same window as Nielsen's national currency cutover. The company changed its television currency on August 31 as ad-supported viewing fell to 71.5 percent of total consumption, having altered seven national currency metrics at once and forced buyers to re-check August delivery. The 71.5 percent figure itself came from Nielsen's own Gauge measurement of ad-supported share. Local and national methodology moved on the same day, which means a year-over-year comparison spanning August 31 is comparing two different measurement systems in two different layers of the same buy.
The gap Nielsen closed is also narrower than the gap that exists. The Coalition for Innovative Media Measurement estimated in November 2025 that 20 to 30 percent of local viewing goes unmeasured altogether, spread across streaming applications, mobile viewing and free ad-supported streaming television channels that the local panel does not reach. CIMM and the Television Bureau of Advertising have since set out requirements for local currency measurement. Recovering a quarter of detected-but-discarded tuning is a real improvement inside the measured universe. It does nothing about the viewing that never reaches the meter.
Six months to count a sale
Google's contribution to the week's counting problem extends in the opposite direction: not who is watching, but how long a conversion remains attributable.
Qualified Future Conversions was described on September 2, 2026 in an Ads Decoded episode featuring John Chen, senior director of product management for ads measurement, and Ginny Marvin, ads product liaison, and PPC Land covered the mechanics on September 5. The metric was first announced at Google Marketing Live in May 2026, one of the launches from that event.
It works in two stages. Within seven days of clicking an advertisement, a user must take a follow-up action that indicates the advertisement influenced them: an attributed branded search, a visit to the advertiser's website, or an add-to-cart event. Chen's illustration was a click on a Nike advertisement followed within the week by a search for Nike or for Jordan 1s. Once that qualifying signal fires, conversions are then counted within the following 30, 90 or 180 days.
Six months from a single click, gated behind a seven-day behavioural check. The gate is what distinguishes this from simply lengthening the lookback window, and it is also where the assumption sits. A branded search within seven days is treated as evidence of advertising influence. It is also, for a large brand, a thing people do constantly for reasons that have nothing to do with any advertisement. The signal is cheap to trigger for advertisers with high organic brand demand and expensive to trigger for advertisers without it, which means the metric will flatter exactly the accounts that need flattering least.
Chen was careful not to present attribution as authoritative. He divided measurement into three instruments separated by operational cadence: attribution, which updates daily and enables real-time optimisation; incrementality testing, which runs quarterly or annually, answers causal questions and forfeits sales through holdback groups; and mix modelling, which updates yearly or quarterly at an aggregate level. The three sometimes agree and sometimes do not, he said.
The candour is warranted by the evidence. A Zalando researcher found in August 2026 that standard mix modelling practice overstates paid search returns by roughly 2.5 times when checked against experimental ground truth. The IAB State of Data 2026 report, published on February 7, 2026 from a survey of more than 400 senior planning and analytics decision-makers at United States brands and agencies, found 75 percent of advertisers saying advanced measurement approaches fail to deliver the rigour, timeliness, trust and efficiency needed to justify spending. All three instruments, in other words, are already failing the people who depend on them.
Marvin introduced data strength as the factor determining whether the new metric can function at all. Chen enumerated the layers: web transactions as a mandatory baseline, application sales data where applicable, offline and partner transaction data from retailers, upper-funnel signals such as add-to-cart and site visits, and media cost data to permit a return-on-ad-spend calculation. That is a demanding list, and it excludes most advertisers below enterprise scale.
One structural constraint cuts against the whole design. Google's pipeline maintains a seven-day gate on offline conversion uploads, which prevents late-arriving records from feeding data-driven attribution models. A metric that counts sales for 180 days after a click sits on top of an ingestion path that ignores offline records arriving after seven. For any business whose sales close in a shop or over a telephone weeks after the click, the extended window measures what the pipeline was already willing to accept.
Around the metric, Google has been steadily lowering the technical barrier to its mix model. Meridian was released open source in March 2024, when Google opened the model to public use. Meridian Studio arrived in May 2026 to run on BigQuery data inside Google Cloud. Meridian landed inside Analytics 360 the same month with a turnkey setup requiring only existing data connections. And the Scenario Planner beta opened in February 2026 with per-channel spend constraints and no code requirement. Chen noted that the platform versions automatically absorb updates such as reach and frequency innovations and Google query volume features, removing manual model maintenance.
The convenience is real. So is the arrangement underneath it. Google builds the measurement instruments, hosts the model, supplies the query volume features that feed it, and sells the media those instruments evaluate. Chen's own guidance for practitioners facing a client who does not believe the attributed return was to reach for incrementality studies as causal validation, or to adjust targets manually. One advertiser, he said, reduced June targets by a third on the understanding that the spend pays off in November and December. His framing was agricultural: an advertiser that only harvests will eventually run out of things to harvest, which makes planting for the next cycle necessary.
Meta's competing framework places randomised experiments above modelled attribution, which is a different ordering of the same three instruments and an acknowledgement of the same fragmentation.
Google shuts the door on API wrappers, including MCP servers
The most operationally disruptive item of the week was a policy update, published on August 31, 2026 and covered on September 5.
Google's revised Ads Developer Policies now prohibit programmatic proxies. The definition is deliberately wide: a third-party hosted interface, secondary API, wrapper service, MCP server, proxy endpoint, or any similar service that solely replicates, wraps, or re-exposes Google Ads programmatic capabilities. The explicit naming of Model Context Protocol servers is the part with the longest reach, because MCP servers are precisely how the current generation of agentic tooling connects a language model to an advertising platform.
Two requirements follow. Each integration must connect through its own dedicated Google Cloud project with direct authentication, rather than routing through shared infrastructure. And users must sign in themselves to make account changes, instead of scripts authenticating on their behalf with vendor-controlled credentials.
Two exceptions survive. Developers automating their own agency's accounts remain compliant. So do open-source tools where the end user downloads the software locally and authenticates with their own credentials. The distinction is architectural rather than commercial: software running on the user's machine with the user's credentials is permitted, software running on a vendor's server with the vendor's credentials is not.
Google's stated rationale runs on two tracks. Unaudited proxies can expose unauthorised access and enable cross-tenant data leaks when multiple customers route through shared infrastructure. And shared proxies can throttle throughput and force rate-limiting that affects every downstream user when a single client generates excessive traffic. Both are ordinary multi-tenancy risks, and both are more acute when the client generating traffic is an autonomous agent rather than a person clicking a button.
A review path exists. The Developer Secondary Interface Review process handles legitimate cases requiring an intermediary layer, and requires adherence to Required Minimum Functionality, measurement transparency standards and SOC 2 Type II auditing. That last requirement is a meaningful filter. SOC 2 Type II is an audited control assessment conducted over a period of months, not a form, and small tool vendors do not typically hold one.
Enforcement is already underway. The Ads API Compliance team is reviewing existing integrations, developers receive notice at their registered email address, and a remediation period precedes penalties. The penalties themselves are graduated: access downgrades, rate-limiting restrictions, and termination of API access.
Read against Google's other recent access changes, the direction is consistent. Passkeys became required for sensitive Google Ads actions from July 15, 2026, with a phased rollout continuing through September. Customer Match API access consolidated on April 1, 2026. Data retention was cut to 37 months in June 2026. Each step narrows what a third party can do on an advertiser's behalf while the platform's own reporting surface expands.
The competitive context is worth stating plainly. Amazon, Meta and Microsoft have all shipped advertising MCP servers. Google's policy does not prohibit an MCP server as such; it prohibits one that sits between the advertiser and Google's systems as a hosted intermediary. For agencies and tool vendors that built agentic workflows on exactly that pattern over the past year, the remediation is a rebuild rather than a configuration change, and the clock is running from whenever the compliance email arrives.
A skincare lawsuit that puts the influencer programme in the complaint
On September 4, 2026, a class action was filed in the United States District Court for the Northern District of California, San Francisco Division, as case 3:26-cv-09653-LJC. PPC Land covered the filing on September 5.
The plaintiffs are Joseph Vettel, a parent in San Diego County, and L.K., a 13-year-old in Alameda County represented by her guardian Lauren Kremer. The defendant is Sephora USA, Inc., the LVMH subsidiary headquartered in San Francisco, which operates 1,813 retail locations in the United States. Counsel are Kirby McInerney LLP and Wolf, Haldenstein, Adler, Freeman and Herz LLP.
The complaint contains a single count under California Business and Professions Code section 17200, the Unfair Competition Law. It alleges that Sephora marketed more than 1,600 skincare products containing active ingredients to minors without age-appropriate warnings, while separately publishing material describing those same ingredients as potentially unsuitable for younger skin. Two classes are proposed, both running from September 4, 2022: a tween class of purchasers under 12, and a teen class aged 13 to 17. The relief sought is restitution and an injunction rather than damages, with the amount in controversy pleaded above 5 million dollars under the Class Action Fairness Act.
The ingredient categories named are alpha-hydroxy acids including glycolic, lactic, malic, citric and tartaric acid, appearing in more than 1,200 products; beta-hydroxy acids, principally salicylic acid; vitamin C derivatives across more than 400 products; retinoids; peptides; and chemical sunscreen filters including avobenzone and oxybenzone. The complaint asserts that 62 percent of the retailer's skincare products contain at least one ingredient it characterises as unsuitable.
What makes the filing relevant beyond cosmetics is the section describing how the products reached the audience. The complaint enumerates marketing architecture rather than product labels: character collaborations beginning with Hello Kitty in January 2011 and Disney Reigning Beauties across 2012 and 2013; the Sephora Squad influencer programme running since 2019; the Sephora and TikTok Incubator Programme established in March 2023; product gifting to youth creators; and teen-founded brands including Sincerely Yours, co-founded in September 2025 by the then-15-year-old Salish Matter, and Evereden. Both of those brands, the complaint notes, contain flagged ingredients despite being marketed for developing skin.
A disclosure allegation runs through it. Plaintiffs contend that creators received undisclosed compensation including free products and promotional merchandise in violation of California law, and argue that the undisclosed pecuniary relationship materially affected the credibility of the endorsement. That is a familiar theory in regulatory enforcement, where it produces fines measured in tens of thousands of dollars. Attached to a restitution claim across two classes covering four years of purchases at 1,813 stores, the exposure is of a different order.
The scientific support cited is recent. The American Academy of Dermatology issued a press release on October 28, 2025 advising that tweens and teens avoid retinol, vitamin C and exfoliating acids. A study published in Pediatrics in July 2025 by Molly Hales and colleagues found that youth skincare routines demonstrated on TikTok averaged six products, with alpha-hydroxy acids appearing up to seven times in a single routine through layering.
There is regulatory history too, and the plaintiffs use it against the settlement rather than for it. The Connecticut Attorney General opened an investigation into youth-targeted marketing in November 2024. Sephora settled on April 20, 2026 with four commitments: brands must provide warnings, the website must carry a conspicuous disclosure, employees must be trained for customers under 13, and an online resource must identify unsuitable products. The complaint points at what the settlement leaves out, namely shelf labels, physical point-of-sale material, dedicated tween and teen sections in stores, and any training specific to 13-to-17-year-olds.
Two details from the timeline appear in the filing as evidence of intent. In December 2022, North West posted an extensive tween skincare routine at age nine. In October 2024, chief executive Artemis Patrick stated that TikTok trends at Sephora were not by accident. The Tweens and Teens Skincare Guide appeared on the website on August 4, 2025 and has since been archived.
The novel compliance question the complaint raises concerns reach rather than targeting. It does not principally attack algorithmic targeting decisions, which is where most youth-marketing enforcement has landed. It attacks organic reach: followers watching creator content that was seeded by a brand and that reaches minors incidentally through a social platform. That is a harder thing to control with an audience exclusion, because there is no audience setting to change.
Sephora had not filed a response as of publication, and every allegation remains unproven. The retailer's commercial calendar continues regardless: it is bringing its Drop Shop format to TikTok Shop starting September 19, 2026.
Buyers think viewers hate AI advertising. Viewers mostly do not.
FreeWheel published research in September 2026 that measures a gap between what the buy side believes about audiences and what audiences report, and PPC Land covered it on September 5.
Forty-eight percent of media buyers and 39 percent of media sellers believe viewers dislike advertisements generated with artificial intelligence. Ten percent of viewers reported disliking the artificial-intelligence advertisements they actually encountered. The buy-side gap is 38 percentage points. The sell-side gap is 29.
The methodology combines three surveys, which is worth stating because it affects how the comparison should be read. Buyers and sellers were sampled through AdExchanger in April 2026, 226 buyers and 50 media sellers, with an earlier November 2025 survey of 216 marketers and agencies feeding the trend lines. Viewers were sampled separately through Dynata in April 2026, 2,496 United States adults. Fifty sellers is a small base, and the two populations answered different instruments. The direction of the finding is clear; the precision implied by two-decimal comparisons is not.
The viewer responses are more permissive than the headline suggests. Eighty-nine percent supported artificial intelligence being used to reduce advertisement repetition. Eighty-nine percent favoured using it to time advertisements so they interrupt less. Seventy-six percent accepted artificial-intelligence-personalised advertising. Those are approvals of applications, not of the technology in the abstract, and each one describes a use that improves the viewing experience rather than the production cost.
Buyers and sellers disagree with each other almost as much as they disagree with viewers. Fifty-three percent of sellers believe artificial intelligence improves television advertising against 41 percent of buyers. Forty-eight percent of buyers report that it creates job friction, against 42 percent of sellers. And 57 percent of sellers think the benefits are distributed equally between the two sides, against 32 percent of buyers, a 25-point spread on the question of who is actually gaining.
Their automation priorities diverge accordingly. Buyers want campaign performance monitoring at 68 percent, relevant inventory targeting at 53, publisher performance tracking at 48 and help drafting requests for proposals at 47. Direct publisher negotiation sits last at 19 percent, which is a clear statement that buyers want the machine to do the paperwork and not the deal. Sellers want inventory allocation at 60 percent, inventory classification at 55 and assisted pricing at 53.
Both sides converge on one thing, and it returns to the theme of the week. Forty-seven percent of buyers named proof of return on investment as the top barrier to increasing connected television spend. Sixty-nine percent of sellers named advanced attribution as the most important artificial intelligence capability they need. The disagreement about creative is loud. The agreement about counting is quiet and much larger. Comparative work in the category, such as the VAB and TVision study contrasting premium video against YouTube on connected television metrics, keeps landing on the same measurement questions from the opposite direction.
Deadlines: an auto-archive, a help desk, and a seven-day commission
Three smaller items carry dates that matter.
Display and Video 360 has one hard cutoff in its third-quarter roadmap. On October 26, 2026, all remaining active and paused legacy line items will be auto-archived and stop serving, covering legacy maximum cost-per-view line items and Video reach 1.0 line items. There is no automatic conversion path to Video reach 2.0 or Video view line items; migration is manual, which leaves roughly seven weeks from early September to rebuild affected campaigns. Archiving paused inventory as well removes historical reference material that buyers use to rebuild. Two further changes sit in the same roadmap: a new Interactive Advertising Bureau-aligned audio taxonomy arriving in the third quarter, under which line items that do not explicitly target unknown content type may see a reduction in inventory, a reach loss that occurs silently with no setting changed; and a February 2027 billing shift from on-download to begin-to-render for display impressions, which Google expects to produce a slight decrease in overall billed impression volume and therefore a year-over-year discontinuity in every report spanning it. Brand suitability controls consolidate on October 1, 2026 into three inventory modes and ten content themes, deprecating sensitive categories and digital content labels for programmatic inventory and removing category-level exclusions that some advertisers hold as contractual obligations. Connatix becomes JWX and Seedtag becomes Seedtag NeuroX on September 21, with exchange identifiers unchanged. Structured Data Files v9, v9.1 and v9.2 sunset on January 28, 2027.
Google completed a separate rollout on September 3, 2026, putting an artificial intelligence assistant in front of 100 percent of English-language AdSense Help Centre traffic. The Help guide answers onboarding, troubleshooting and feature questions through a floating widget, displays response sources behind expandable controls, retains conversation history and offers four feedback channels. It routes escalations to Help Centre articles, community forums or human support depending on account tier, and account suspension queries route to policy articles rather than to a person. Google states that responses may vary in accuracy and that decisions based on the guide are the publisher's responsibility, which places the consequence of a wrong answer on the publisher receiving it. The commercial backdrop is a Network segment, comprising AdSense, AdMob and Ad Manager, that fell 4 percent to 6.97 billion dollars in the first quarter of 2026 and 1 percent to 7.3 billion in the second, while Search advertising grew 17 percent.
And Apple's updated United Kingdom Developer Programme License Agreement, running to 114 pages, contains a clause on page 102 that claims commission on purchases made on a developer's own website within seven days of a link-out from the application, including on a return visit involving no further Apple platform interaction. The clause was flagged publicly by xigxag co-founder Kelli F. in a LinkedIn post. The practical consequence for lifecycle marketing is direct: a retention email sent within seven days of an initial link-driven sale could trigger commission liability on the resulting purchase, extending the fee from the steering transaction into organic re-engagement. Equivalent European Union terms were published on August 18, 2026 with a parallel seven-day attribution window and take effect on October 1, 2026. The regulatory frame around it is active. The United Kingdom Competition Appeal Tribunal ruled on October 23, 2025 that Apple abused a dominant position in app markets; the Competition and Markets Authority opened steering conduct consultations on June 30, 2026 and Apple objected to the proposed fair and reasonable fee framework on July 29. In the European Union, Apple replaced per-install fees with a 5 percent commission on non-App Store sales. In the United States, courts eliminated external purchase commissions after a civil contempt finding on April 30, 2025.
Also noted
- September 3: Press Gazette identified four more finance and cryptocurrency writers, Sadie Ann Williamson, Ralph Tkatchuk, Victor G Snyder and Daan Pepijn, whose existence it could not verify despite bylines in Forbes and Entrepreneur, bringing the running total to eight writers across more than 1,000 articles linked to the blockchain public relations firms Inbound Junction and MarketAcross. PPC Land
- September 1: Pinterest reported that teenagers on the platform re-pin 64 percent more often and search 44 percent more than other age groups while avoiding posting selfies, with accounts for 13 to 15-year-olds defaulting to private and comments disabled, as Generation Z passes half of the platform's 640 million global users. PPC Land
- September 5: OpenX said its infrastructure processes 3 million queries per second without a cloud tax, publishing throughput figures at a moment when supply-path economics are being renegotiated across the exchange layer. PPC Land
- September 7: Anthropologie is adding nine Nike styles by September 21 and expanding footwear from eight stores to 200, after customer interest in sneakers rose nearly 30 percent over the past year, with footwear general merchandise manager Jessica Irick Peek describing the shoe customer as the retailer's most valuable. Digiday
- September 6: Disney passed 4 billion dollars at the 2026 global box office ahead of a seven-film autumn slate, a figure that sets the promotional inventory available to the studio's marketing partners through the end of the year. PPC Land
Discussion