United bans AI agents from bookings as Brussels targets paid AI picks

PPC Land is now also on WhatsApp, with a new PPC Land channel carrying its stories.

On Wednesday, November 11, 2026, one American airline and one European institution both drew lines around the AI shopping agent. United Airlines changed its Contract of Carriage to forbid AI agents from buying its tickets, seats and upgrades, while the European Commission adopted a Digital Fairness Act proposal that would let Brussels fine companies up to 6% of worldwide turnover and would require AI shopping agents to disclose when a suggestion has been paid for. One company wants agents kept away from the checkout. The other wants them to say who is paying. Around those two decisions sat a day of smaller moves on the same question of control: Google expanding its AI answers and testing where it puts their sources, brands buying software to correct what chatbots say about them, and an ad industry still unable to say whether an agent's conversation ever turned into a sale.

The Landline newsletter can also be heard. Marketing Is Calling reads each edition aloud every day, on Spotify, Apple Podcasts and YouTube. Marketing Rodeo picks the most controversial advertising story of each day, weekends included, and has some fun with it, on Spotify, Apple Podcasts and YouTube.

United closes the checkout to agents

The airline's revision is short but sweeping. A new Rule 6.J.6 lists "the use of AI Agents to make, modify or service a Ticket, and/or to purchase any service or product" among prohibited practices. A new definition in Rule 1 covers autonomous and semi-autonomous software, including AI assistants built into consumer platforms, AI browsers and third-party assistants reached through travel agency tools, "whether or not Passenger or another individual supervises or approves" the action. Rule 5.H lets United cancel "reservations made, modified or serviced by AI Agents" without notice, and the remedies in Rule 6.K run from invalidating tickets and banning passengers to deleting miles, revoking Elite status and ending MileagePlus membership.

Agents are not banned from the airline's fares altogether. View from the Wing and InsideFlyer reported that agents may still search and display prices, as long as the customer completes the purchase. View from the Wing, citing The Wall Street Journal, attributed to chief executive Scott Kirby a wish to avoid "another middleman." United itself pointed to privacy, unintended purchases and regulation. PPC Land noted that the contract leaves key terms undefined, describes no way of detecting an agent and makes no exception for partners, a contrast with the terms Walmart published for agents on October 21, 2026.

Brussels writes rules for paid suggestions

The Commission's proposal is a 72-page regulation that would replace the Unfair Commercial Practices Directive of 2005. It was adopted together with a Simplification Directive and a revised Consumer Protection Cooperation Regulation, and it would give the Commission its own power to investigate and fine. Ten dark patterns would be banned outright. Subscriptions would need renewal reminders and an online cancellation button. Article 26 would require AI shopping agents to disclose paid suggestions, and Article 11(6) would make brands and agencies liable for the compliance of the influencers they pay.

For advertising technology, Article 12 is the clause to read. It would extend the Digital Services Act's ban on profiling-based ads aimed at minors, and on targeting with sensitive data, from platforms to all traders, ad tech included. The Commission's own Q&A put it plainly: "The DFA will extend these bans to all websites and apps." Coordinated national actions would carry maximum fines of at least 4% of turnover or EUR 2 million. The impact assessment put consumer detriment at no less than EUR 10.3 billion in 2026 and per-company compliance costs at EUR 15 to EUR 41 a year. Christoph Busch of the University of Osnabrueck said "The DFA addresses the right issues. But there is still room for improvement in the details." The text would apply 12 months after entering into force, and the European Parliament and the Council still have to agree on it.

Hims & Hers faces seven states

A day earlier, on Tuesday, November 10, the Federal Trade Commission and seven states filed a first amended complaint against Hims & Hers Health in the Northern District of California. Arizona, Florida, Nebraska, Ohio and Wisconsin joined California and Utah, and the number of counts rose from nine to 16. The 68-page filing alleges that "free" consultations masked paid prescription subscriptions, that cancelling was made hard, and that health data went to ad platforms despite privacy promises.

The advertising detail is specific. The complaint says Hims uploaded about 14 million consumer email addresses to Meta between September 2020 and June 2023, of which about 10 million matched user IDs, and sent about 8 million pixel and server events. Five Snap lists held more than 1 million addresses. The complaint says the company "was only able to create audiences with such specificity because it flouted the promises it made to its users." It also quotes the Hims chief executive: "a near 10% D0 Cancellation is wild." Florida can seek up to $50,000 per violation for selling sensitive data without consent. The Commission voted 2 to 0. Hims denies the claims, and no ruling on the merits or schedule has been set.

Google's AI answers grow while their sources move

Google's own pages kept changing shape. On November 11, Glenn Gabe noticed that AI Overviews for definition queries now appear fully expanded by default, filling the first screen so that web results sit below the fold. Gabe called it a big heads-up for sites that publish definitions. Barry Schwartz recalled that Google's Robby Stein said in August that AI Overviews "only dynamically expand for topics where our systems determine it to be most useful."

Late the same evening, Brodie Clark of SERP Alert shared a test in which Google moves the source icons to the bottom of AI Overviews. In an earlier version, a count of referenced sites sat at the top, directly above the citations. "This version is different: hovering displays the full list of citations," Clark wrote. Schwartz could not reproduce it and argued that the placement could push publisher clicks lower still.

Who appears inside those answers is its own contest. Semrush's Peak Season Positioning report, published on November 9 and based on 10,514 AI answers to US holiday shopping questions, found that Ashley Furniture appeared in 66.5% of Black Friday furniture prompts against Walmart's 13.4%. Amazon, Walmart, Best Buy and Target led only 23.6% of the 89 AI Overviews topics with enough prompts to name a leader. Gemini named Amazon in 40.5% of prompts but never cited amazon.com, and Reddit was cited in 53.2% of Christmas gift prompts in AI Overviews. "Winning citations or mentions offers no guarantee of earning the other," Semrush wrote. Black Friday falls on November 27.

Google has its own reading of the shift. PPC Land examined a Think with Google essay by Brendon Kraham claiming AI search shoppers visit 12 touchpoints, nearly three times as many as other shoppers. The figure comes from Google-commissioned Ipsos research fielded in December 2025 across 25 markets, comparing 14,305 AI-using shoppers with 21,881 non-users. "The reality is the new era of Search doesn't shrink the consumer journey; it expands it," Kraham wrote. PPC Land noted that Search Console's AI reports count impressions rather than clicks and that the survey does not show causation.

Survey data from the other side pointed the same way and the opposite way at once. Hostinger asked 1,014 US adults in October and found that 65% had turned to Reddit or TikTok when Google did not help, rising to 80% among Gen Z. Yet 88% still used Google while choosing their latest business, and 94% said they double-check AI recommendations elsewhere. The report itself cautioned that "All figures are self-reported and reflect stated attitudes and recalled behavior, not observed behavior."

The money question got a model. Sydney consultancy Marketix Digital published a scenario on November 11 in which a firm ends up A$43,400 a month short despite gaining AI visits. Its business loses 20% of 100,000 monthly organic visits worth A$4 each and gains 5,000 AI referrals worth A$7.32 each, which replaces only 45.75% of the A$80,000 displaced. Closing the gap would take about 10,929 AI visits. Managing director Shoaib Mughal said "Businesses need to put a revenue target against their AI search investment." The numbers are illustrative, not forecasts, and Marketix sells advisory services.

Smaller engines claimed the attention that Google is said to be losing. Brave said on November 10 that Brave Search handled 2.7 billion queries in October, up 56% from a year earlier, with users rising from 101 million to more than 130 million and ad revenue growing 2.85 times, without a baseline. "We have heard a lot about search being dead during 2026, but our numbers tell a different story," said Jean-Paul Schmetz, chief of ads. PPC Land flagged two different user figures in the release and noted that it left out the lawsuits The New York Times and three other papers filed against Brave on November 2.

On the technical side, Gary Illyes and Chery Prommawin closed Google's video series with a warning that moving to a content delivery network can accidentally overwhelm a site. A faster server can prompt Googlebot to raise its crawl rate, and a fresh CDN cache starts empty, so a crawl of 100,000 uncached URLs still lands on the origin. Engineers who mistake the spike for an attack and block bots can end up blocking Googlebot too, a problem Google's Search Central post describes in detail.

Brands try to correct the chatbot

Software makers are now selling the repair kit. Sitecore launched MarketingIQ, which drafts fixes when ChatGPT or Gemini gets a brand wrong, on November 11 at its Symposium in Orlando. It pairs AI-answer monitoring from Scrunch, which Sitecore acquired in June in a deal Adweek put at about $225 million, with content and campaign tools. When a chatbot omits a brand or quotes outdated pricing, it drafts a correction from approved content, and Sitecore says "nothing leaves a staging area without human involvement." It is included with SitecoreAI licences rather than sold separately, and no customer performance figures were given.

Mars is approaching the same problem from the grocery aisle. Neil Reynolds, its global chief customer officer, told Digiday that Mars wants to future-proof impulse buys as AI takes over more grocery runs. Drawing on a future of shopping report produced with Kantar and Synthesis, he described a two-part roadmap: making Mars brands discoverable by language models and making sure discovery converts, and building experiential in-store destinations with retailers. "how do we make sure our brands are discoverable by the LLMs," he asked, and on retailer relationships: "Why is our data team not talking to the data team at Walmart?"

Measuring what agents deliver remains the weak link. Invoca's Broken Funnel Report, based on 451 US marketing leaders surveyed between September 14 and October 5, found that 74% of those running AI agents cannot fully track whether an agent-handled lead becomes a sale. Only 19% can reliably send outcomes back to ad platforms, and just 7% use revenue itself as the main measure for their agents. "Too many consumer brands measure their AI agents like a help desk and expect them to perform like a sales team," said chief executive Gregg Johnson. Invoca sells tools that address the gap.

Agents and controls inside the ad accounts

AppsFlyer moved the agent into bidding. On November 11 it put an Apple Ads Share of Voice agent into closed beta that estimates share of voice by keyword and geography, recommends bid changes and applies them only after a marketer approves. It also released Create an Agent, for custom agents built from a written workflow, and four additions to its MCP server. "Marketing teams are moving from manual campaign orchestration to agentic execution," said Inna Weiner, vice president of product. The release left out pricing, results and how approval is enforced. Apple's Campaign Management API sunsets on January 26, 2027, and Google, Meta, Moloco and Unity invested in AppsFlyer at a $2.7 billion valuation in June.

Google, meanwhile, gave advertisers more ways to see what its own automation is doing. A new ValueTrack parameter, {search_term_matchtype}, separates AI Max traffic from keyword clicks, using "e" for exact, "p" for phrase, "b" for broad and "a" for AI Max expansions and keywordless matches. Hana Kobzova spotted it. A pilot reporting card called Customer Journey Insights now sits on the Overview page for eligible Performance Max, Search and Shopping campaigns, showing unique users across four stages labelled Unaware, Aware, Consider and Act.

Several help pages changed on November 11 too. Google clarified that brand lists draw on a single global library of canonical brands rather than localised variants, so a regional brand stuck "under review" should be replaced with the primary name. It added explanatory text to Performance Max page feeds, spotted by Arpan Banerjee, setting out how they interact with Final URL expansion. And it published eligibility rules for its merchant loyalty program: 14 supported countries, one country per account for now, and active listings through free listings, Shopping ads, free local listings or local inventory ads.

The ads themselves are also changing. Google is adding a "Visit site" link to mobile search ad descriptions, which Sachin Patel reported and which Adrian Dekker described as part of a label test already running on sponsored results. Target frequency campaigns on YouTube can now mix non-skippable ads with Shorts and In-feed ads, with non-skippable ads allowed up to 20 seconds instead of 15 and a single frequency range of two to seven a week. Google also shipped Google Ads API version 26.2 on November 11, a minor release that follows versions 26 and 26.1 in October.

Microsoft Advertising published a guide to its "Made with AI" watermark. Ads show the label when they use a display, image or video asset tagged "AI Content + Watermark" and are served in the EU or New York State. Premium Streaming video campaigns and HTML5 ads are excluded, and the platform maps AI disclosures imported from Google Ads and Pinterest.

Measurement, supply and the long wait for standards

The industry body that audits measurement set its own timetable. The Media Rating Council said on November 10 that its first draft standards for AI in media measurement will arrive for public comment in the first quarter of 2027, with a second phase by mid-2027. Its July guidance had promised completion "throughout 2026 and early 2027," so the schedule has slipped without explanation. Nothing yet covers agentic AI traffic, zero-click measurement, synthetic audiences or ads inside chatbots.

Verification firms kept extending their reach. Zefr announced on November 11 four ad quality measurements for CapCut and TikTok's TopView format: brand safety, suitability, viewability and invalid traffic, along with coverage for image ads and search inventory. "Social media today extends far beyond a platform's core feed," said co-chief executive Richard Raddon. IAS announced coverage the same morning and named seven surfaces to Zefr's four.

On the sell side, PubMatic reported a swing to a $5.5 million profit as third-quarter revenue rose 17% to $79.6 million, above guidance, with adjusted EBITDA up 133% to $26.0 million. Fourth-quarter guidance is $91 million to $93 million. Extending the useful life of its servers from three to five years added $1.5 million to net income, and the board added $150 million to its buyback. "And this is not a one-quarter event," said retiring chief financial officer Steve Pantelick, while chief executive Rajeev Goel noted that "Safari is a small part of our business."

Yahoo argued that the default way of buying connected TV is the wrong one. PPC Land examined a Yahoo blog post claiming one CTV advertiser reached 60% more homes for 32% less spend after moving away from programmatic guaranteed deals. "Our objection isn't to PG. It's to PG as a default, because a default is a decision nobody made," Yahoo wrote. It cited eMarketer figures showing that 93.8% of CTV spend is programmatic. PPC Land noted that the post gives no spend, prices, publishers or methodology and promotes Yahoo DSP.

Data and screens changed hands as well. AudienceMix announced on November 10 that it gained Alliant consumer data spanning more than 6,000 audiences, adding to more than 20,000 segments it sells through five buying platforms. VIOOH, the JCDecaux programmatic marketplace, added GSTV's video screens at more than 29,000 US gas stations, which it says reach 115 million adults a month through a full-sound loop of about four minutes. "Full-sound video, a captive audience and coast-to-coast reach is a rare combination," said Gavin Wilson, its global chief commercial officer. And Bannerflow said it signed more than 20 UK and US clients ahead of opening offices in London and New York, naming SEGA, bet365, Ooni, SailGP and Aldi Sud, and reported more than EUR 25 million in annual recurring revenue. iGaming accounted for 76.1% of its impressions in September.

Commerce, creators and the betting boom

TikTok tested a membership. Modern Retail reported, in a story Digiday carried, that TikTok Shop launched a paid premium tier for 1 million of its top US shoppers. TikTok Shop Plus costs $9.99 a month after a 30-day free trial and offers expanded free shipping, an extra 10% off selected products, three monthly coupons, instant refunds as store credit and exclusive Black Friday coupons. Amazon Prime costs $14.99 a month and Walmart+ $12.95. "TikTok Shop is no longer experimental," Patrick Nommensen, president of strategic initiatives, said at an October summit. Sales passed $500 million over the 2025 Black Friday weekend, and this is the first holiday season with Best Buy, Ulta Beauty and Sephora selling there.

Creators showed how much production and distribution now sit with one person. Digiday profiled Zach Hirsch, known as Mystic Zach, a sports betting analyst whose DraftKings-sponsored podcast gained more than 200 million short-form views since the NFL season began. He cited a 2024 Edison Research finding: "Eighty-five percent of Gen Z sports podcast listeners are introduced through social media, which is clips." Mat Micheli of Viral Nation said "Betting is now deeply ingrained in the culture of the brands that deem it important." The article gives two different TikTok view counts, 151 million and 43.7 million. Creator Grace Wells told Digiday's Creator Strategies event that each branded video she makes takes about 100 hours, with about 30 hours of storyboarding before filming and about 20 in post-production. "The bulk of my time is spent on set filming, just because I am one person with two hands," she said.

Betting brands are spending at the top of the funnel too. Adweek examined how prediction markets used the midterm elections as a branding moment, with the Super Bowl as their next test. In mid-October a Times Square billboard showing the odds of Democrats winning the Senate turned out to be an ad for Kalshi. Kalshi and Polymarket together took in nearly $1.9 billion in wagers on the midterm vote, a figure Adweek drew from a Washington Post analysis. Polymarket, Kalshi and Novig still face legal fights and NFL restrictions that limit where their ads can run.

Music discovery remained a major-label business. Vevo released its 2027 DSCVR Artists to Watch list on November 10, and PPC Land found that major labels hold 16 of the 18 places: nine Universal, six Sony and one Warner, chosen from more than 600 submissions. Universal and Sony founded Vevo in 2009.

Brands make the case for human craft

Two campaigns sold the hand behind the work. Adobe made "Made to Create" the centrepiece of Adobe Max in Miami, and chief marketing officer Lara Balazs told Adweek it is Adobe's biggest brand campaign in five years, the first in that time to bring its whole product range into one execution. It features more than 30 artists from 18 countries, alongside the tools behind their work, at a moment when, in Adweek's words, "AI is challenging the economics of the creative software business."

Johnnie Walker released a holiday spot in which Sabrina Carpenter swaps espresso for whisky, directed by Damien Chazelle as a French New Wave take on the long-running "Keep Walking" campaign. Carpenter walks through a jazz bar to order a Go Go Highball, one of her signature cocktails, as part of a partnership aimed at the next generation of whisky drinkers.

The Washington Post nears breakeven

Publishing offered a rare recovery story. Executive editor Matt Murray told an audience at Web Summit in Lisbon that The Washington Post is close to breaking even after "hundreds and hundreds of millions" in losses. "I expect next year will be our best year in seven years," he said. Direct ad sales had their strongest year-on-year growth since 2021 in the third quarter, and a spokesperson said programmatic revenue was up 48% as of September. "We went from something like 35,000 subscriptions before Trump to 3 million," Murray said. The newsroom is hiring again, and the Post plans to revive a version of its daily podcast, with possible ad sales through Acast.

The same Adweek column carried other publishing news. City Cast co-founder David Plotz is stepping away after six years, with Andi McDaniel becoming chief executive. Slate named Hillary Frey chief executive and made small layoffs. Le Monde chief executive Louis Dreyfus said the paper has 720,000 subscribers, 620,000 of them digital-only, and about 600 journalists, up from 310. "Seventy-five percent of my revenue comes from readers," he said.

Readers who pay and agents that are kept at the door point in the same direction. United wants the customer to finish the booking. Brussels wants the agent to declare its sponsor. Publishers and brands want the answer engines to get them right. Each is trying to keep a human relationship from being absorbed by software that sits in the middle, and none of them can yet measure how much that software is taking.