Meta told investors on July 29 that its advertising business had never grown faster, and then spent the rest of its earnings call explaining why net income fell anyway. The numbers, laid out in a report from PPC Land, capture a tension that recurred across the marketing industry on Wednesday and Thursday: platforms are getting better at making money from advertising even as costs, regulatory pressure and structural questions about how people find things online complicate the picture underneath.

Meta's second-quarter advertising revenue reached 59.36 billion dollars, up 27 percent from a year earlier. Total company revenue hit 60.80 billion dollars, an increase of 28 percent. Yet net income dropped 8 percent to 15.8 billion dollars, and diluted earnings per share fell 13 percent to 6.18 dollars. The culprit was not the advertising engine itself but two one-off items sitting on top of it: a 2.4 billion dollar charge tied to legal proceedings and 1.18 billion dollars in severance costs connected to a headcount reduction that began in May. Operating margin, which had held between 40 and 48 percent for eight consecutive quarters, dropped sharply to 31 percent. Total costs and expenses rose 55 percent to 42.03 billion dollars.

Susan Li, Meta's chief financial officer, walked through the mechanics on the earnings call. Ad impressions across the Family of Apps rose 14 percent year over year, while the average price per ad climbed 12 percent. Family daily active people averaged 3.60 billion for June, a gain of just 3 percent, a growth rate that has been essentially flat since the fourth quarter of 2025. Regional detail told a more interesting story than the headline figures suggested: price per ad rose 20 percent in the US and Canada, but only 1 percent in Asia-Pacific, even though impression growth in Asia-Pacific outpaced every other region at 17 percent. Two regions, in other words, are monetizing new volume in almost opposite ways.

The quarter's most consequential disclosure for anyone who buys or sells advertising through Meta's systems was a new ranking architecture the company calls the Meta Generative Recommender. Rather than scoring each potential ad individually, Li explained, the system now uses large language models to reason jointly about ad content and user preferences before predicting the single best match for each person. Early pilots using the technology to understand user preferences produced a 1 percent lift in app event conversions on Instagram. A separate combination of Meta's user-understanding models and what the company calls its GEM model for ranking generated an 8.3 percent increase in ad clicks and a 15.7 percent uplift in conversions on Facebook. Those figures sit on top of a trajectory PPC Land has tracked for several quarters: the GEM model doubled its training cluster size in the fourth quarter of 2025, when it produced a 3.5 percent lift in Facebook ad clicks, giving the new numbers a clear point of comparison. Meta's automated Advantage+ suite, meanwhile, crossed an annual revenue run rate above 75 billion dollars, up from 60 billion dollars disclosed in the third quarter of last year.

None of that engineering progress changed the fact that Meta is now absorbing a legal bill large enough to move its full-year guidance. The company raised the lower end of its 2026 expense outlook specifically to accommodate the 2.4 billion dollar charge, pushing the range to between 165 billion and 169 billion dollars. Meta was unusually direct about what remains unresolved, telling investors it continues to face youth-related trials scheduled for later this year in the United States that "may ultimately result in a material loss." Chief executive Mark Zuckerberg was pressed separately about an apparent contradiction in the company's compute strategy: Meta is fielding offers to sell its own spare compute capacity at a premium while simultaneously buying more from other providers. His answer, that there is simply nowhere near enough compute to meet demand, doubled as an explanation for why capital expenditure reached 31.08 billion dollars for the quarter, up sharply from 17.01 billion dollars a year earlier. Free cash flow, weighed down by that spending and a 24.91 billion dollar increase in long-term debt issuance, fell to just 784 million dollars from 8.55 billion dollars in the same quarter last year.

Just hours after Meta's numbers landed, Microsoft published its own fiscal fourth-quarter results, and the contrast between the two reports illustrated how differently platforms are currently trading advertising growth against everything else they build. Microsoft's fourth-quarter results, released July 29 for the period ended June 30, showed total revenue of 90.0 billion dollars, up 18 percent, and net income climbing 31 percent to 35.8 billion dollars. A 3.2 billion dollar gain from Microsoft's investment in Anthropic, whose valuation rose after a combined 15 billion dollar commitment from Microsoft and Nvidia in November, padded that net income figure considerably. Search advertising revenue, the line that matters most to anyone buying inventory through Bing, Copilot or Microsoft's advertising partnerships, told a less flattering story: growth decelerated to 10 percent on a reported basis, or 9 percent in constant currency, down from 12 percent in the prior quarter and well below the 21 percent pace Microsoft posted throughout fiscal 2025.

Amy Hood, Microsoft's chief financial officer, attributed the growth to higher revenue per search and volume across Edge and Bing, while blaming third-party partnerships for weighing on the total. Satya Nadella, Microsoft's chief executive, offered a different emphasis in the same call, noting that Bing and Edge have gained audience share in search and advertising for five consecutive years. Those two framings are not contradictory so much as incomplete on their own: Microsoft's owned search surfaces appear to be gaining ground even as the revenue line PPC Land tracks each quarter reflects a mix increasingly shaped by publisher and content-partner deals whose economics shift from one period to the next. Full fiscal year search advertising revenue reached 15.176 billion dollars, up from 13.878 billion dollars in fiscal 2025, so the business keeps expanding in absolute terms even while its quarterly growth rate swings considerably. Guidance for the first quarter of fiscal 2027 pointed to further softness, with Hood telling analysts to expect mid-single-digit growth.

While search advertising slowed, Microsoft's cloud business crossed a threshold years in the making. Azure revenue surpassed 100 billion dollars for the full fiscal year for the first time, with quarterly Azure growth accelerating to 43 percent. Microsoft Cloud revenue, the broader measure spanning Microsoft 365 Commercial cloud, Azure, the commercial share of LinkedIn and Dynamics 365, reached 59.3 billion dollars for the quarter and 214.4 billion dollars for the full year. Commercial remaining performance obligation, a forward-looking measure of contracted revenue not yet recognized, hit 678 billion dollars, up 84 percent year over year. Capital expenditures rose 70 percent to 41.0 billion dollars for the quarter, with roughly two-thirds directed toward the short-lived chips that power Azure's AI workloads. Read together, Meta and Microsoft's results on the same day suggest that among the largest advertising platforms, revenue from ads is increasingly one line among several competing for capital, attention and management's explanatory energy, rather than the central story it once was.

Away from the earnings calendar, a different kind of reckoning arrived for the crawler traffic that increasingly determines whether content shows up anywhere at all. IAB Australia published a Bots and Crawler Guidance and Decision Matrix on July 30, unveiled two days earlier at the trade body's Discovery: AI and Search Summit in Sydney. The document asks publishers to stop treating every automated visitor as an identical threat and instead sort each one into one of five job categories, then apply one of four verdicts: allow, allow with conditions, require licensing, or block. The timing matters because the underlying traffic composition has already shifted past a threshold that would have seemed remarkable even a year ago. Cloudflare data cited in the guidance puts automated requests at 57.5 percent of web-page traffic as of June 2026, the first time bot traffic has outpaced human traffic on record. Of that automated share, roughly 52 percent goes toward training artificial intelligence models, while only about 2.6 percent represents real-time fetches triggered by an actual person asking a question.

The guidance's central technical argument is that volume and commercial value have become separate questions, and that publishers who conflate the two risk making the wrong call in both directions. Discovery and search indexers such as Googlebot and bingbot get an unconditional allow recommendation. AI training crawlers, including GPTBot and ClaudeBot, get a block-or-license recommendation for publishers, though brands are advised toward a more selective stance that lets basic product facts through while withholding anything resembling proprietary content. Live AI agents such as ChatGPT-User and Claude-User, which fetch pages in real time because a person asked a direct question, receive a conditional recommendation restricted on ad-funded pages. One mechanic recurs throughout the document: a single crawler operator often runs more than one bot for more than one purpose, and blocking a training-only signal does not necessarily remove a publisher's content from AI answer features built on the same underlying index. Anthropic's own documentation, updated in February, separates ClaudeBot for training, Claude-User for live queries and Claude-SearchBot for search quality, illustrating exactly the kind of granularity the guidance is asking every operator to build toward.

The guidance also carries an operational deadline that traces back to an announcement Cloudflare made on July 1: before September 15, new domains and free-tier zones sitting behind Cloudflare's network will have Training and Agent crawlers blocked by default on any page carrying advertising. DataDome figures cited in the IAB Australia document show AI-agent requests climbing 45 percent quarter on quarter to 17.7 billion in the second quarter, even as 80 to 88 percent of AI referral traffic to sites now traces back to ChatGPT specifically. The document frames licensing, built on the Content Monetization Protocol that IAB Tech Lab finalized on April 28, as an emerging third path between blanket allowance and blanket blocking, describing it not as a new category of crawler but as a new kind of commercial relationship that can attach to any of the five job categories once an agreement exists. Whether Australian publishers, or publishers anywhere else reading the same document, actually adopt that framework before the Cloudflare deadline arrives remains an open question the guidance itself cannot answer; what it does establish is a shared vocabulary for a decision that, until now, most sites were making crawler by crawler, without much documentation and often without noticing which bot was actually doing what.

A parallel but distinct question, this time about human consent rather than machine access, surfaced the same day in Vienna. The privacy group noyb filed a complaint against dict.cc, a free multilingual dictionary site, arguing that a single click on the site's consent banner asks visitors to accept data sharing with 1,741 named technology partners at once. The complaint, filed with the Austrian Data Protection Authority on July 30 under case number C107, centers on a visit the group says took place on July 7 at 14:45, when a complainant clicked "Accept all and visit website" simply to reach the translation search that had brought them to the site in the first place.

What distinguishes this complaint from the routine flow of consent-banner disputes is its arithmetic. noyb calculated how long it would actually take to read the privacy policies named in dict.cc's banner: roughly 12 minutes for the first partner listed, 32 minutes for the second, 36 minutes for the third. Extrapolating across all 1,721 partners named in the legal filing itself, at a minimal pace of one minute per partner just to locate basic information, the exercise would take 28 hours and 41 minutes. At a more realistic six minutes per partner to actually understand how each company processes data, the total climbs to 172 hours, more than seven full days. Felix Mikolasch, a data protection lawyer at noyb, said in a statement that reading and understanding "1,741 companies" worth of data protection policies could take "days or even weeks," calling it ridiculous to assume that reading exercise would allow for an informed decision. The complaint pushes the logic one step further, noting that named partners commonly redistribute data to further recipients never disclosed in the original banner; assuming each of the 1,721 partners shares onward with just 100 further parties, the hypothetical chain reaches 172,100 potential recipients from a single click.

The legal grounds cited are specific rather than sweeping. noyb alleges dict.cc breached Article 5(1)(a) of the GDPR, the requirement that personal data be processed lawfully, fairly and transparently, and Article 6(1), which requires a valid legal basis before any processing occurs at all. The filing leans on the Court of Justice of the European Union's Planet49 ruling, which held that consent must be clear and comprehensible enough for a user to understand what they are agreeing to, and on European Data Protection Board guidance stating that where a controller fails to make information genuinely accessible, a user's control becomes illusory. The complaint asks the Austrian authority to declare the processing unlawful, order deletion of the data collected, and impose a fine under Article 83.

The dict.cc case matters beyond one Vienna-based dictionary site because the underlying structure it targets, listing hundreds or thousands of vendors behind a single consent click on the theory that disclosure alone satisfies transparency requirements, is common practice across programmatic advertising more broadly. noyb's own media statement names other sites it says follow a similar pattern, including www.repubblica.it, www.bergfex.de and www.fifa.com. IAB Europe's Transparency and Consent Framework has already drawn scrutiny at similar scale: the group's own 2025 compliance report recorded enforcement procedures against registered vendors rising 118 percent year over year to 587 cases, against a backdrop of 953 registered vendors. If the Austrian regulator accepts noyb's reasoning that disclosure at a scale no realistic visitor will ever read cannot function as informed consent, the implications would not stop at one dictionary site; any consent banner listing partner counts in the hundreds or thousands would face the identical underlying question.

Two smaller but related threads ran through the same period, both concerning how conversational AI surfaces are reshaping the paths people take to find and buy things. Productrise, a platform that tracks organic product visibility inside Google Shopping, published a study on July 29 finding that Google's AI Mode surfaced roughly 95 percent fewer product listings than standard search across more than 100,000 identical shopping queries tracked over three weeks. Standard search returned at least one product on about 88 percent of shopping queries, compared with just 23 percent for AI Mode; when products did appear on both surfaces, standard search displayed an average of 22.5 listings per page against 4.3 for AI Mode. Only about 0.8 percent of products visible in standard search also appeared in AI Mode for the identical query on the identical day, a finding that suggests the two surfaces draw from largely separate slices of the same underlying catalog rather than simply showing a trimmed version of one result set. Jason Berkowitz, founder of Break The Web, framed the shift as a widening advantage for brands that invest early in product data quality, while Hugo Huijer, Productrise's founder, offered a competing theory: AI Mode may show fewer products upfront precisely because its conversational structure defers additional options to a follow-up exchange rather than surfacing everything at once, meaning each visible slot could carry disproportionately higher value even as the total pie shrinks.

Meanwhile, over on the marketplace side of the industry, Digiday reported on July 30 that OpenAI has begun offering promotional ad credits to new advertisers testing its ChatGPT ads platform, a familiar tactic every major platform has deployed at a similar stage of building an advertising business. One promotion reviewed by Digiday offers new advertisers a 50 dollar credit after they themselves spend 50 dollars within 14 days of opening an account; another version showed a 100 dollar credit. Farhad Divecha, group chief executive of Accuracast, told Digiday the timing made sense given how early OpenAI's advertising business remains, though he added that a 500 to 1,000 dollar credit would likely do more to convince hesitant advertisers than the smaller amounts currently circulating. The move follows Digiday's earlier reporting that OpenAI is already building out a specialist advertising sales organization, a detail PPC Land had separately reported on July 29 in connection with a Head of Scaled Ads Solutions posting that names resellers and outsourced call-center channels as the company's next growth lever, while no listed role yet covers merchant feed data quality specifically. Taken together, the coupon strategy and the sales-org buildout point toward a company moving deliberately from experimentation toward the harder, less glamorous work of recruiting and retaining advertisers at scale, a process every prior platform has needed years, not months, to get right.

The Video Advertising Bureau added its own contribution to the day's identity conversation, publishing a 15-page guide on July 29 that names UID2, OpenID and RampID as the primary identifier options ad buyers should evaluate, and urging buyers to audit vendor match-rate accuracy before signing any agreement. The guide arrives at a moment when identity infrastructure is being rebuilt across several fronts simultaneously: Nexxen separately announced a self-serve demand-side platform feature that cuts data onboarding time to 24 hours, validating audiences against an identity graph whose global footprint has nearly doubled, by the company's own account ahead of its second-quarter earnings. Whether these overlapping identity efforts converge into anything resembling a shared standard, or simply add another layer of vendor-specific tooling for buyers to evaluate, is precisely the kind of question the VAB's guide seems designed to help ad buyers ask before they commit budget rather than after.

Also noted

  • July 29, 2026 - AdExchanger reported that eyewear company Blacksheep accused Google of replacing its top organic search result with a broken lookalike domain after a viral television segment drove a traffic spike, prompting the company to park 25 LED billboard trucks outside Google's New York offices. Read more
  • July 29, 2026 - Search Engine Roundtable confirmed that Google's Platform Properties feature inside Search Console, which shows content performance data for Instagram, TikTok, X and YouTube, rolled out to all users, ending a staggered availability period that had left some accounts without access. Read more
  • July 30, 2026 - MediaPost reported on a study from Reddit and Omnicom's Weber Shandwick Collective describing the platform as an emerging hub for women's health discussion, with researchers pointing to anonymity and unfiltered community candor as factors driving the shift away from traditional health information sources. Read more
  • July 29, 2026 - PPC Land reported that Onton, an AI shopping platform, published benchmarks claiming its model outperforms Google Shopping and Amazon on product trust accuracy, though the figures come from the company's own testing and have not yet been verified by an independent laboratory. Read more
  • July 29, 2026 - PPC Land reported that Google's Gemini assistant for Mac gained voice control over any open application through a long-press of the Fn key, letting users dictate, edit and generate images inside other software, with the feature launching in English only and additional languages expected in future updates. Read more