Meta informed traders on July 29 that its promoting enterprise had by no means grown sooner, after which spent the remainder of its earnings name explaining why internet earnings fell anyway. The numbers, specified by a report from PPC Land, seize a pressure that recurred throughout the advertising and marketing business on Wednesday and Thursday: platforms are getting higher at being profitable from promoting whilst prices, regulatory stress and structural questions on how individuals discover issues on-line complicate the image beneath.
Meta’s second-quarter promoting income reached 59.36 billion {dollars}, up 27 p.c from a yr earlier. Complete firm income hit 60.80 billion {dollars}, a rise of 28 p.c. But internet earnings dropped 8 p.c to fifteen.8 billion {dollars}, and diluted earnings per share fell 13 p.c to six.18 {dollars}. The offender was not the promoting engine itself however two one-off objects sitting on high of it: a 2.4 billion greenback cost tied to authorized proceedings and 1.18 billion {dollars} in severance prices linked to a headcount discount that started in Could. Working margin, which had held between 40 and 48 p.c for eight consecutive quarters, dropped sharply to 31 p.c. Complete prices and bills rose 55 p.c to 42.03 billion {dollars}.
Susan Li, Meta’s chief monetary officer, walked by means of the mechanics on the earnings name. Advert impressions throughout the Household of Apps rose 14 p.c yr over yr, whereas the common value per advert climbed 12 p.c. Household day by day lively individuals averaged 3.60 billion for June, a acquire of simply 3 p.c, a progress price that has been primarily flat for the reason that fourth quarter of 2025. Regional element informed a extra attention-grabbing story than the headline figures recommended: value per advert rose 20 p.c within the US and Canada, however just one p.c in Asia-Pacific, although impression progress in Asia-Pacific outpaced each different area at 17 p.c. Two areas, in different phrases, are monetizing new quantity in nearly reverse methods.
The quarter’s most consequential disclosure for anybody who buys or sells promoting by means of Meta’s methods was a brand new rating structure the corporate calls the Meta Generative Recommender. Fairly than scoring every potential advert individually, Li defined, the system now makes use of massive language fashions to purpose collectively about advert content material and person preferences earlier than predicting the only finest match for every particular person. Early pilots utilizing the know-how to grasp person preferences produced a 1 p.c elevate in app occasion conversions on Instagram. A separate mixture of Meta’s user-understanding fashions and what the corporate calls its GEM mannequin for rating generated an 8.3 p.c enhance in advert clicks and a 15.7 p.c uplift in conversions on Fb. These figures sit on high of a trajectory PPC Land has tracked for a number of quarters: the GEM mannequin doubled its coaching cluster dimension within the fourth quarter of 2025, when it produced a 3.5 p.c elevate in Fb advert clicks, giving the brand new numbers a transparent level of comparability. Meta’s automated Benefit+ suite, in the meantime, crossed an annual income run price above 75 billion {dollars}, up from 60 billion {dollars} disclosed within the third quarter of final yr.
None of that engineering progress modified the truth that Meta is now absorbing a authorized invoice massive sufficient to maneuver its full-year steerage. The corporate raised the decrease finish of its 2026 expense outlook particularly to accommodate the two.4 billion greenback cost, pushing the vary to between 165 billion and 169 billion {dollars}. Meta was unusually direct about what stays unresolved, telling traders it continues to face youth-related trials scheduled for later this yr in the USA that “could finally end in a cloth loss.” Chief government Mark Zuckerberg was pressed individually about an obvious contradiction within the firm’s compute technique: Meta is fielding presents to promote its personal spare compute capability at a premium whereas concurrently shopping for extra from different suppliers. His reply, that there’s merely nowhere close to sufficient compute to satisfy demand, doubled as an evidence for why capital expenditure reached 31.08 billion {dollars} for the quarter, up sharply from 17.01 billion {dollars} a yr earlier. Free money move, weighed down by that spending and a 24.91 billion greenback enhance in long-term debt issuance, fell to only 784 million {dollars} from 8.55 billion {dollars} in the identical quarter final yr.
Simply hours after Meta’s numbers landed, Microsoft revealed its personal fiscal fourth-quarter outcomes, and the distinction between the 2 studies illustrated how in a different way platforms are at present buying and selling promoting progress towards every part else they construct. Microsoft’s fourth-quarter results, launched July 29 for the interval ended June 30, confirmed whole income of 90.0 billion {dollars}, up 18 p.c, and internet earnings climbing 31 p.c to 35.8 billion {dollars}. A 3.2 billion greenback acquire from Microsoft’s funding in Anthropic, whose valuation rose after a mixed 15 billion greenback dedication from Microsoft and Nvidia in November, padded that internet earnings determine significantly. Search promoting income, the road that issues most to anybody shopping for stock by means of Bing, Copilot or Microsoft’s promoting partnerships, informed a much less flattering story: progress decelerated to 10 p.c on a reported foundation, or 9 p.c in fixed forex, down from 12 p.c within the prior quarter and nicely under the 21 p.c tempo Microsoft posted all through fiscal 2025.
Amy Hood, Microsoft’s chief monetary officer, attributed the expansion to larger income per search and quantity throughout Edge and Bing, whereas blaming third-party partnerships for weighing on the entire. Satya Nadella, Microsoft’s chief government, provided a special emphasis in the identical name, noting that Bing and Edge have gained viewers share in search and promoting for 5 consecutive years. These two framings are usually not contradictory a lot as incomplete on their very own: Microsoft’s owned search surfaces look like gaining floor even because the income line PPC Land tracks every quarter displays a combination more and more formed by writer and content-partner offers whose economics shift from one interval to the following. Full fiscal yr search promoting income reached 15.176 billion {dollars}, up from 13.878 billion {dollars} in fiscal 2025, so the enterprise retains increasing in absolute phrases even whereas its quarterly progress price swings significantly. Steering for the primary quarter of fiscal 2027 pointed to additional softness, with Hood telling analysts to count on mid-single-digit progress.
Whereas search promoting slowed, Microsoft’s cloud enterprise crossed a threshold years within the making. Azure income surpassed 100 billion {dollars} for the complete fiscal yr for the primary time, with quarterly Azure progress accelerating to 43 p.c. Microsoft Cloud income, the broader measure spanning Microsoft 365 Industrial cloud, Azure, the industrial share of LinkedIn and Dynamics 365, reached 59.3 billion {dollars} for the quarter and 214.4 billion {dollars} for the complete yr. Industrial remaining efficiency obligation, a forward-looking measure of contracted income not but acknowledged, hit 678 billion {dollars}, up 84 p.c yr over yr. Capital expenditures rose 70 p.c to 41.0 billion {dollars} for the quarter, with roughly two-thirds directed towards the short-lived chips that energy Azure’s AI workloads. Learn collectively, Meta and Microsoft’s outcomes on the identical day recommend that among the many largest promoting platforms, income from advertisements is more and more one line amongst a number of competing for capital, consideration and administration’s explanatory vitality, quite than the central story it as soon as was.
Away from the earnings calendar, a special type of reckoning arrived for the crawler site visitors that more and more determines whether or not content material exhibits up anyplace in any respect. IAB Australia revealed a Bots and Crawler Guidance and Decision Matrix on July 30, unveiled two days earlier on the commerce physique’s Discovery: AI and Search Summit in Sydney. The doc asks publishers to cease treating each automated customer as an equivalent risk and as a substitute kind each into certainly one of 5 job classes, then apply certainly one of 4 verdicts: permit, permit with situations, require licensing, or block. The timing issues as a result of the underlying site visitors composition has already shifted previous a threshold that might have appeared exceptional even a yr in the past. Cloudflare information cited within the steerage places automated requests at 57.5 p.c of web-page site visitors as of June 2026, the primary time bot site visitors has outpaced human site visitors on document. Of that automated share, roughly 52 p.c goes towards coaching synthetic intelligence fashions, whereas solely about 2.6 p.c represents real-time fetches triggered by an precise particular person asking a query.
The steerage’s central technical argument is that quantity and industrial worth have turn into separate questions, and that publishers who conflate the 2 threat making the flawed name in each instructions. Discovery and search indexers similar to Googlebot and bingbot get an unconditional permit suggestion. AI coaching crawlers, together with GPTBot and ClaudeBot, get a block-or-license suggestion for publishers, although manufacturers are suggested towards a extra selective stance that lets primary product info by means of whereas withholding something resembling proprietary content material. Dwell AI brokers similar to ChatGPT-Person and Claude-Person, which fetch pages in actual time as a result of an individual requested a direct query, obtain a conditional suggestion restricted on ad-funded pages. One mechanic recurs all through the doc: a single crawler operator usually runs a couple of bot for a couple of objective, and blocking a training-only sign doesn’t essentially take away a writer’s content material from AI reply options constructed on the identical underlying index. Anthropic’s personal documentation, up to date in February, separates ClaudeBot for coaching, Claude-Person for dwell queries and Claude-SearchBot for search high quality, illustrating precisely the type of granularity the steerage is asking each operator to construct towards.
The steerage additionally carries an operational deadline that traces again to an announcement Cloudflare made on July 1: earlier than September 15, new domains and free-tier zones sitting behind Cloudflare’s community could have Coaching and Agent crawlers blocked by default on any web page carrying promoting. DataDome figures cited within the IAB Australia doc present AI-agent requests climbing 45 p.c quarter on quarter to 17.7 billion within the second quarter, whilst 80 to 88 p.c of AI referral site visitors to websites now traces again to ChatGPT particularly. The doc frames licensing, constructed on the Content material Monetization Protocol that IAB Tech Lab finalized on April 28, as an rising third path between blanket allowance and blanket blocking, describing it not as a brand new class of crawler however as a brand new type of industrial relationship that may connect to any of the 5 job classes as soon as an settlement exists. Whether or not Australian publishers, or publishers anyplace else studying the identical doc, truly undertake that framework earlier than the Cloudflare deadline arrives stays an open query the steerage itself can’t reply; what it does set up is a shared vocabulary for a choice that, till now, most websites had been making crawler by crawler, with out a lot documentation and sometimes with out noticing which bot was truly doing what.
A parallel however distinct query, this time about human consent quite than machine entry, surfaced the identical day in Vienna. The privateness group noyb filed a complaint against dict.cc, a free multilingual dictionary web site, arguing {that a} single click on on the location’s consent banner asks guests to simply accept information sharing with 1,741 named know-how companions without delay. The criticism, filed with the Austrian Knowledge Safety Authority on July 30 beneath case quantity C107, facilities on a go to the group says befell on July 7 at 14:45, when a complainant clicked “Settle for all and go to web site” merely to achieve the interpretation search that had introduced them to the location within the first place.
What distinguishes this criticism from the routine move of consent-banner disputes is its arithmetic. noyb calculated how lengthy it will truly take to learn the privateness insurance policies named in dict.cc’s banner: roughly 12 minutes for the primary associate listed, 32 minutes for the second, 36 minutes for the third. Extrapolating throughout all 1,721 companions named within the authorized submitting itself, at a minimal tempo of 1 minute per associate simply to find primary info, the train would take 28 hours and 41 minutes. At a extra real looking six minutes per associate to truly perceive how every firm processes information, the entire climbs to 172 hours, greater than seven full days. Felix Mikolasch, an information safety lawyer at noyb, mentioned in a press release that studying and understanding “1,741 firms” price of information safety insurance policies may take “days and even weeks,” calling it ridiculous to imagine that studying train would permit for an knowledgeable choice. The criticism pushes the logic one step additional, noting that named companions generally redistribute information to additional recipients by no means disclosed within the authentic banner; assuming every of the 1,721 companions shares onward with simply 100 additional events, the hypothetical chain reaches 172,100 potential recipients from a single click on.
The authorized grounds cited are particular quite than sweeping. noyb alleges dict.cc breached Article 5(1)(a) of the GDPR, the requirement that private information be processed lawfully, pretty and transparently, and Article 6(1), which requires a sound authorized foundation earlier than any processing happens in any respect. The submitting leans on the Courtroom of Justice of the European Union’s Planet49 ruling, which held that consent should be clear and understandable sufficient for a person to grasp what they’re agreeing to, and on European Knowledge Safety Board steerage stating that the place a controller fails to make info genuinely accessible, a person’s management turns into illusory. The criticism asks the Austrian authority to declare the processing illegal, order deletion of the information collected, and impose a fantastic beneath Article 83.
The dict.cc case issues past one Vienna-based dictionary web site as a result of the underlying construction it targets, itemizing tons of or 1000’s of distributors behind a single consent click on on the speculation that disclosure alone satisfies transparency necessities, is frequent observe throughout programmatic promoting extra broadly. noyb’s personal media assertion names different websites it says comply with an analogous sample, together with www.repubblica.it, www.bergfex.de and www.fifa.com. IAB Europe’s Transparency and Consent Framework has already drawn scrutiny at related scale: the group’s personal 2025 compliance report recorded enforcement procedures towards registered distributors rising 118 p.c yr over yr to 587 instances, towards a backdrop of 953 registered distributors. If the Austrian regulator accepts noyb’s reasoning that disclosure at a scale no real looking customer will ever learn can’t operate as knowledgeable consent, the implications wouldn’t cease at one dictionary web site; any consent banner itemizing associate counts within the tons of or 1000’s would face the equivalent underlying query.
Two smaller however associated threads ran by means of the identical interval, each regarding how conversational AI surfaces are reshaping the paths individuals take to search out and purchase issues. Productrise, a platform that tracks natural product visibility inside Google Procuring, revealed a study on July 29 discovering that Google’s AI Mode surfaced roughly 95 p.c fewer product listings than commonplace search throughout greater than 100,000 equivalent procuring queries tracked over three weeks. Commonplace search returned no less than one product on about 88 p.c of procuring queries, in contrast with simply 23 p.c for AI Mode; when merchandise did seem on each surfaces, commonplace search displayed a median of twenty-two.5 listings per web page towards 4.3 for AI Mode. Solely about 0.8 p.c of merchandise seen in commonplace search additionally appeared in AI Mode for the equivalent question on the equivalent day, a discovering that implies the 2 surfaces draw from largely separate slices of the identical underlying catalog quite than merely displaying a trimmed model of 1 consequence set. Jason Berkowitz, founding father of Break The Net, framed the shift as a widening benefit for manufacturers that make investments early in product information high quality, whereas Hugo Huijer, Productrise’s founder, provided a competing concept: AI Mode could present fewer merchandise upfront exactly as a result of its conversational construction defers extra choices to a follow-up alternate quite than surfacing every part without delay, which means every seen slot may carry disproportionately larger worth whilst the entire pie shrinks.
In the meantime, over on {the marketplace} aspect of the business, Digiday reported on July 30 that OpenAI has begun offering promotional ad credits to new advertisers testing its ChatGPT advertisements platform, a well-known tactic each main platform has deployed at an analogous stage of constructing an promoting enterprise. One promotion reviewed by Digiday presents new advertisers a 50 greenback credit score after they themselves spend 50 {dollars} inside 14 days of opening an account; one other model confirmed a 100 greenback credit score. Farhad Divecha, group chief government of Accuracast, informed Digiday the timing made sense given how early OpenAI’s promoting enterprise stays, although he added {that a} 500 to 1,000 greenback credit score would possible do extra to persuade hesitant advertisers than the smaller quantities at present circulating. The transfer follows Digiday’s earlier reporting that OpenAI is already constructing out a specialist promoting gross sales group, a element PPC Land had individually reported on July 29 in reference to a Head of Scaled Ads Solutions posting that names resellers and outsourced call-center channels as the corporate’s subsequent progress lever, whereas no listed function but covers service provider feed information high quality particularly. Taken collectively, the coupon technique and the sales-org buildout level towards an organization transferring intentionally from experimentation towards the more durable, much less glamorous work of recruiting and retaining advertisers at scale, a course of each prior platform has wanted years, not months, to get proper.
The Video Promoting Bureau added its personal contribution to the day’s identification dialog, publishing a 15-page guide on July 29 that names UID2, OpenID and RampID as the first identifier choices advert consumers ought to consider, and urging consumers to audit vendor match-rate accuracy earlier than signing any settlement. The information arrives at a second when identification infrastructure is being rebuilt throughout a number of fronts concurrently: Nexxen separately announced a self-serve demand-side platform feature that cuts information onboarding time to 24 hours, validating audiences towards an identification graph whose world footprint has almost doubled, by the corporate’s personal account forward of its second-quarter earnings. Whether or not these overlapping identification efforts converge into something resembling a shared commonplace, or just add one other layer of vendor-specific tooling for consumers to guage, is exactly the type of query the VAB’s information appears designed to assist advert consumers ask earlier than they commit funds quite than after.
Additionally famous
- July 29, 2026 – AdExchanger reported that eyewear firm Blacksheep accused Google of changing its high natural search consequence with a damaged lookalike area after a viral tv section drove a site visitors spike, prompting the corporate to park 25 LED billboard vans outdoors Google’s New York workplaces. Read more
- July 29, 2026 – Search Engine Roundtable confirmed that Google’s Platform Properties characteristic inside Search Console, which exhibits content material efficiency information for Instagram, TikTok, X and YouTube, rolled out to all customers, ending a staggered availability interval that had left some accounts with out entry. Read more
- July 30, 2026 – MediaPost reported on a examine from Reddit and Omnicom’s Weber Shandwick Collective describing the platform as an rising hub for ladies’s well being dialogue, with researchers pointing to anonymity and unfiltered group candor as components driving the shift away from conventional well being info sources. Read more
- July 29, 2026 – PPC Land reported that Onton, an AI procuring platform, revealed benchmarks claiming its mannequin outperforms Google Procuring and Amazon on product belief accuracy, although the figures come from the corporate’s personal testing and haven’t but been verified by an impartial laboratory. Read more
- July 29, 2026 – PPC Land reported that Google’s Gemini assistant for Mac gained voice management over any open utility by means of a long-press of the Fn key, letting customers dictate, edit and generate photographs inside different software program, with the characteristic launching in English solely and extra languages anticipated in future updates. Read more
Source link
