00The week in one paragraph
This was the week the market priced what publishers already knew. Q2 earnings did the talking: Amazon cleared $19.8B in quarterly ad revenue, up 26%, and barely mentioned it; Meta grew ads 27% while its expenses grew 55%; and Reddit posted a 64% ad quarter and still fell nearly 13%, marked down for the one exposure most publishers carry — an audience that arrives via Google. Publishers spent the same week converting exposure into terms. Time became the first major publisher to sell ads to AI bots, on a site that sees more bot traffic than human traffic most days; the newsrooms that were threatening to block Googlebot are now seriously modelling what it would cost, against publisher ad supply down as much as 40% year over year; and PubMatic shut down its open-web wrapper, handing ~250 publishers to Playwire and conceding the wrapper is no longer a business worth owning. The buy side moved its margin around — holdcos are burying AI costs inside principal media commitments, and Omnicom’s third-party service costs, where principal earnings sit, jumped from $918M to $1.5B — while Google piloted the first real PMax opt-outs since 2021, FreeWheel switched on show-level reporting for seven streamers, Xbox began testing ad-supported game streaming, the IAB Tech Lab shipped AAMP 2.3 with Agentic Audiences finalised, and the FTC sued Hims & Hers over health data shared with Meta and Snap. The through-line is leverage: this week the parties who hold it started writing rate cards, and the parties who lost it started paying in margin.
01PubMatic shuts down its web wrapper — ~250 publishers pointed at Playwire
PubMatic is retiring OpenWrap Web and recommending Playwire to roughly 250 web publishers, while keeping OpenWrap SDK for its 750–1,000 app publishers. VP John Martin’s reasoning is blunt: wrapper tech “was very ripe five, six, seven years ago,” and PubMatic can reach web inventory regardless of whose wrapper you run. App monetisation is now the “bread and butter”; the open-web wrapper has become a commodity an SSP no longer needs to own (AdExchanger, 27 July).
Publisher takeaway
If you run OpenWrap Web, this is a Q3 migration project with a deadline someone else set. Start with the audit: export every line item, floor rule and bidder configuration the wrapper holds, because that configuration is the actual asset you’re moving. Then treat Playwire as one option, not the default — a hand-off negotiated between two vendors is not automatically the best deal for you, and the alternatives (self-hosted Prebid.js, another managed wrapper, or your GAM partner running the auction) should be priced against it before you sign anything. Check rev-share and data terms line by line; wrapper migrations are where margin quietly changes hands. If you don’t run OpenWrap, the signal still applies: PubMatic just told the market that owning a web wrapper buys an SSP nothing, because it can reach your inventory through anyone’s. That means whoever operates your wrapper has no structural moat — so choose one whose incentives you can see, benchmark its take regularly, and keep your configuration portable enough that leaving is a project, not a hostage negotiation.
02Time starts selling ads to AI bots
Time converted its pages to markdown and, with Mobian, began serving sponsored FAQ-style blocks aimed at AI agents rather than humans — targeted contextually or by date range, measured on visibility, favourability and accuracy, with Ally Bank and the Project Management Institute among the first buyers. COO Mark Howard: “This is a growing traffic source, and therefore a growing source of inventory.” Time now sees more bot traffic than human traffic most days, and the inventory prices at a premium — the first credible attempt to turn the crawl from a cost line into a rate card (Digiday, 30 July).
Publisher takeaway
Before you can decide whether this is a model or a novelty, you need the number Time already has: your bot share. Split your server logs by user agent and ASN and find out what portion of your requests are agents, which agents, and against which sections — if bots outnumber humans on your site too, you are already operating a second audience you have never priced. Then be clear-eyed about what’s proven here. Two named buyers is a pilot, not a market; the measurement currency — visibility, favourability, accuracy — has no accepted standard yet; and the premium pricing is Time’s claim, not an audited yield figure. So don’t convert your site to markdown on spec. The move is to make crawl gateable and meterable first — per-crawler controls, logged access, a page structure agents can parse — because that infrastructure pays off whether you end up selling to bots, licensing to them, or blocking them. Time’s real contribution is the reframe: crawl is a traffic source, and traffic sources get rate cards. Work out what yours would say, even if you never publish it.
03Life without Googlebot: the modelling has begun in earnest
The threat covered in our last two issues has moved into the spreadsheet stage. With publisher ad supply down as much as 40% year over year in Q2 and 15.5% of US Google news searches now showing Top Stories inside an AI Overview, Digiday finds blocking Google’s crawlers has gone from conference talk to boardroom slide — Reuters is openly weighing the tradeoff. The history is unkind: Axel Springer and Spanish publishers tried versions of this in 2014 and reversed within weeks, because whoever moves first eats the loss alone. The realistic path is the UK CMA route — opting out of AI features while keeping search placement — which one exec called a “poisoned chalice,” because nobody has the data to model the downside yet (Digiday, 30 July).
Publisher takeaway
Nobody can model the poisoned chalice for you, so build the model yourself before anyone asks for a decision. Three numbers to produce. First, revenue per Google-referred session by section — not traffic share, because sessions that don’t monetise aren’t worth defending. Second, the share of your Google traffic arriving on brand terms, which would partly survive any delisting via direct navigation. Third, your subscription-to-programmatic revenue mix, because that ratio is your actual leverage: subscription businesses can absorb the traffic hit, traffic-dependent ones can’t, and knowing which you are settles most of the argument. History says first movers fold — 2014 proved it — so the scenario worth modelling isn’t the full block, it’s the CMA-style partial opt-out: AI features declined, search placement kept. Your model should output the price at which that trade breaks even for you. Having that number commits you to nothing. Not having it means that when the moment arrives — a licensing offer, a coalition, a board question — you’ll be negotiating from instinct against parties who did the arithmetic.
04Reddit grows ads 64% — and gets marked down 13% for its Google dependency
$805M total revenue (+61%), $762M in ads (+64%) — and Reddit’s shares still fell nearly 13% after hours, because CEO Steve Huffman conceded “search referrals were choppy in the quarter,” with AI Overviews intercepting users before they click through. Huffman also would not commit to renewing the roughly $60M Google licensing deal: “the range of outcomes is wide.” It is the clearest public-market signal yet that investors will discount any content business whose audience arrives via Google, however strong the ad line looks (AdExchanger, 30 July).
Publisher takeaway
The market just told you how it prices your traffic mix, using the largest test case available: a 64% ad quarter could not outweigh one admission that Google referrals were choppy. If your business ever faces investors, lenders, acquirers or a bank renewing a facility, expect the same discount applied to you — so get ahead of it by reporting audience the way the market now reads it: direct and loyal traffic as its own line, Google-referred as its own line, with the trend on each. Growing direct share is now a valuation input, not just an audience-development goal, which changes what the newsletter, the app and the registration wall are worth in your internal maths — fund them accordingly. The second signal is the licensing line: Huffman wouldn’t commit to renewing a $60M Google deal, calling the range of outcomes wide. If Reddit treats a relationship that size as an open negotiation, your content licensing conversations are negotiations too — nothing about these terms is standard yet, and what you accept this year becomes your baseline for the repricing everyone else is visibly planning.
05Google concedes PMax opt-outs for Display and Search Partners
A limited pilot gives Performance Max advertisers two checkboxes to exclude Google Display Network and Search Partners inventory — the first real concession since PMax launched in 2021. Buyers are delighted; Kyle Rovinski of Duncan Channon: “Search advertisers don’t like the Display Network. We don’t want to opt in.” David Dweck of Go Fish Digital calls GDN and GSP “sources of remnant inventory that Google’s forced advertisers to opt into.” One complicating detail: PMG reported a 10% average increase in PMax investment among clients with the new controls — the pool grows even as its distribution narrows (Digiday, 24 July).
Publisher takeaway
Work out your exposure before the pilot generalises. In GAM, pull earnings by demand channel and isolate what arrives through AdSense and AdX from Google-side demand you never negotiated — that is where forced PMax spend has been landing. If a meaningful share of your open-auction revenue traces back to GDN, you have been monetising an opt-in advertisers didn’t make, and Dweck’s “remnant inventory” framing tells you exactly how those buyers see you. Two responses. First, assume the demand that stays after opt-outs is demand that chose you, and give it reasons to: clean ads.txt, accurate inventory declarations, honest ad density and page experience — the things that move an advertiser from “exclude display” to “include this site.” Second, apply concentration discipline: if Google-mediated demand exceeds the threshold you would flag for any other single partner, treat it the same way, with a quarterly reduction plan. Treat PMG’s 10% figure as what it is — a buyer-supplied number from one agency’s client base, directionally interesting, not a forecast. The stable conclusion holds either way: the pool may grow while its distribution narrows, and you want to be inside the narrower distribution on merit.
06Principal media is now paying the holdcos’ AI bills
Rather than invoice for AI separately, holding companies are asking clients to route fixed percentages of budget through principal inventory — one CMO was offered full absorption of AI infrastructure costs in exchange for 70% of spend going principal (Digiday, 27 July). Robert Webster, ex-WPP: “Agencies like to claim they have invested a lot but much of it is manufactured to justify exactly this — skimming money out of media.” The earnings corroborate the shift: Omnicom posted $6.6B revenue, roughly flat, with advertising revenue down low single digits and net income nearly doubled to $585M — and the stock hit a one-year high on the IPG asset disposals, with CEO John Wren reframing Omnicom as “now more of an operating company than a holding company.” The line publishers should circle: third-party service costs, where principal media earnings sit, jumped from $918M to $1.5B year over year (AdExchanger, 29 July).
Publisher takeaway
Principal media is the channel where your inventory is bought in bulk and resold at an undisclosed markup, and this week established two things about it: it is now large enough to fund holdco AI bills, and it is growing fast enough to nearly double a cost line in a year while ad revenue falls. Both come out of the spread between what the client pays and what you receive — so the pressure on your price at source only goes one way. Three checks. Map which of your demand paths touch principal desks: trace resold paths through your sellers.json entries and ask your SSPs directly which buyers transact on a principal basis against your inventory. Price with the mechanism in mind: a larger commitment offered at a lower rate is the classic shape of a principal deal, and the spread you concede is now carrying the agency’s token costs as well as its margin. And use it in direct conversations — when a holdco tells a client its AI comes free, it doesn’t, and the CMOs discovering that 70% of their budget routes through opaque inventory are the same CMOs a transparent direct or programmatic-guaranteed proposal can win. The buy side’s opacity is, for once, your sales argument.
07FreeWheel switches on show-level reporting — seven streamers opt in
A+E Global Media, Fuse, NBCUniversal, Paramount, Spectrum Reach, Warner Bros. Discovery and Xumo will now let programmatic buyers see which shows their ads ran in, roughly two hours after delivery and at no cost, through FreeWheel’s Buyer Cloud. The data stays post-delivery, deliberately kept out of the bid stream to sidestep VPPA exposure. FreeWheel’s Jon Mansell names the tension outright: “channel conflict” — once buyers can see show titles, they can cherry-pick the hits and bid down everything else (Digiday, 29 July).
Publisher takeaway
This one is for CTV and streaming sellers, but the logic generalises. Before opting into title-level transparency — here or on any platform that copies it — do the catalogue maths: what share of your revenue rides on your top ten shows, and what share of impressions sits in the mid and long tail that buyers will bid down once they can see it? If the hits carry you, transparency is a cherry-picking machine pointed at your rate card; if your catalogue is deep, it is a premium argument you have never been able to prove. Either way the prerequisite is the unglamorous work: standardised show titles, consistent content metadata, show-aware tagging through your ad server — reporting is only as good as the metadata underneath it, and buyers will trust the platform’s labels over your corrections. Package deliberately: if cherry-picking is the risk, sell the hits inside bundles rather than as separable line items. And note the compliance design, because it is a template — FreeWheel kept titles out of the bid stream specifically to avoid VPPA exposure. Any pressure to pass content signals in the bid request should be evaluated against that precedent, not against a buyer’s convenience.
08Amazon clears $19.8B in ad revenue — and barely mentions it
Amazon’s ads grew 26% year over year, from $15.7B to $19.8B, and got only a passing mention on the earnings call — 30 new advertisers on NBA coverage, and an Ads Agent that compresses campaign setup from hours to minutes. Executives wanted to talk about AI instead: net income was $62.6B, of which $53.4B came from the Anthropic stake. AdExchanger’s read is the useful one — Big Tech is deliberately reframing itself as AI-first while the ad business, now a near-$80B annualised run rate, compounds quietly underneath (AdExchanger, 30 July).
Publisher takeaway
The number to hold onto is the run rate — nearly $80B annualised — because that is the gravitational mass sitting on the retail and endemic budgets you pitch. But the operational detail is the Ads Agent: campaign setup compressed from hours to minutes removes the friction that used to keep smaller advertisers from defaulting to Amazon, and those are precisely the advertisers a regional or niche publisher’s direct sales team lives on. Your counter is not matching Amazon’s data — you can’t — it is matching the friction. If buying from you requires an insertion order, three emails and a spec sheet while buying from Amazon takes minutes, budgets follow the path of least resistance regardless of how good your audience is. Get a genuinely self-serve or near-self-serve path live for smaller advertisers, and make your inventory legible to agentic buyers as the AAMP tooling matures (see story 11) — publishers who are easy for software to buy will inherit the spend that agents allocate. And keep pitching what retail media structurally lacks: context, content adjacency and non-endemic reach. Amazon knows what people buy; you know what they care about.
09Meta’s ad revenue grew 27%. Its expenses grew 55%.
$59.4B in ad revenue (+27%) against $42B in expenses, up 55% — driven by $1.2B in severance after 8,000 May layoffs, $2.4B in legal costs including youth-related trials CFO Susan Li warned “may ultimately result in a material loss,” and AI infrastructure spend now heading toward debt financing and a one-gigawatt data centre in El Paso with BlackRock. The figure publishers should mark is the performance one: generative models in Meta’s ads retrieval system lifted clicks 8.3% and conversions 15.7% in early tests (AdExchanger, 29 July).
Publisher takeaway
Skip the cost drama — the severance, the lawsuits and the gigawatt data centres are Meta’s problem. The line that is your problem is the performance one: generative models in ads retrieval lifting clicks 8.3% and conversions 15.7%. Treat those as Meta’s own early-test figures, reported on its own earnings call — claims, not audited results — but take the direction seriously, because every point of provable performance Meta adds resets the bar an open-web CPM has to clear. When a buyer can get a measured double-digit conversion lift by moving budget in-platform, “quality environment” stops winning arguments on its own. Your defence has to be denominated in the same currency: outcomes. Push conversion-tracked deals and PMP packages built on first-party data where you can demonstrate performance rather than assert adjacency; get attention or attribution measurement on your highest-value placements so there is a number in the room that isn’t Meta’s; and pressure-test your own reporting — if an advertiser asked you tomorrow to prove your inventory converts, how long would the answer take? The platforms are turning performance claims into quarterly disclosures. The open web answering with reach and adjectives is how the gap widens.
10Xbox tests ad-supported game streaming
New Xbox CEO Asha Sharma is testing ad-supported game streaming as an alternative to paid subscriptions — her assessment of the division: “our business today is not healthy.” Microsoft already owns the ad tech to serve it, which makes this an execution question rather than a build question. A new premium video inventory pool at Xbox scale would change the supply picture for every CTV and video seller competing for the same budgets; the open question flagged in the reporting is whether hardcore gamers tolerate interruption at all (AdExchanger, 30 July, citing Business Insider).
Publisher takeaway
Nothing to execute this week — this is a forecasting item, and it earns its place because of who is building it. Microsoft owning the stack means the distance from test to scaled inventory is short, and the destination is the CTV and online-video budgets you compete for. If you sell video, add the branch to your H2 scenario planning now: a supply expansion in premium video softens pricing for everyone downstream, and your 2027 rate conversations will happen against whatever this becomes. If you monetise gaming audiences specifically, the open question in the reporting is your opportunity — nobody knows whether hardcore gamers tolerate interruption mid-match, and publishers who already reach that audience through native, contextual and non-interruptive formats hold the answer Microsoft is about to spend heavily finding out. Sharma’s own framing — “our business today is not healthy” — tells you the appetite for experimentation is real, not exploratory. Watch the test markets, note which formats survive contact with the audience, and price your gaming-adjacent inventory knowing a very large competitor is about to teach the market what works in that environment.
11IAB Tech Lab ships AAMP 2.3 — agentic buying gets its guardrails
AAMP 2.3 finalises Agentic Audiences v1.0 for programmatic transaction, adds Google Ad Manager reporting and Meta buying support, brings full OpenDirect spec validation, and wires privacy diligence directly into buyer agent workflows via the IAB Diligence Platform and SafeGuard Privacy. For sell-side teams the substantive parts are the guardrails: verification against real inventory rather than derived numbers, and deterministic, provable price commitments with human approval gates. Adform CTO Jochen Schlosser framed the intent precisely — agentic advertising should build “on, not around, programmatic controls” (IAB Tech Lab, 30 July).
Publisher takeaway
Agentic buying keeps arriving as speculation; this week it arrived as a spec you can implement. For a GAM operator the concrete part is direct: AAMP 2.3 adds Google Ad Manager reporting support, which means buyer agents can now read the delivery data your stack produces — so your reporting hygiene (naming conventions, clean key-values, accurate inventory mappings) just became machine-readable sales collateral. The two guardrails are worth implementing rather than skimming: verification against real inventory rather than derived numbers, and deterministic, provable price commitments with human approval gates. Both are protections that only exist for sellers who wire them in — Schlosser’s line about building “on, not around, programmatic controls” is an intent, and intents bind nobody. Assign someone technical to read the spec and come back with a gap list answering two questions: can an agent verify your avails against reality, and can your price commitments be proven rather than asserted? Sellers who can answer yes become safe counterparties for automated budgets; sellers who can’t will find that agents, unlike human buyers, extend no benefit of the doubt. The standard also gives you a floor for what to refuse — any agentic buyer unwilling to transact within these controls is telling you something.
12FTC, Utah and California sue Hims & Hers over health data shared with ad platforms
The complaint, filed in the Northern District of California on a 2-0 Commission vote under the FTC Act and ROSCA, alleges Hims & Hers shared sensitive health information with Meta, Snap and other ad platforms via customer lists and automated tracking on its site — while promising privacy protections — alongside charging for prescriptions immediately after intake and obstructing cancellations, with Utah and California joining as plaintiffs (FTC, 29 July).
Publisher takeaway
The regulatory exposure sits with the advertiser, not the platform and not you — but discovery does not respect that boundary: pixel-based data flows get mapped end to end, and publishers carrying health, wellness or pharma advertisers are on the map. Three pieces of hygiene. Audit what advertiser and platform tags on your own pages actually collect — if your health-content sections carry pixels that see URL paths, article topics or user inputs, you are a conduit for exactly the data flows named in this complaint, and “we didn’t know what the tag did” is a weak position in 2026. Document the diligence: a dated record of which tags run where, what they collect and who approved them is the difference between a defensible posture and an assertion, and it costs your ad ops team an afternoon. And read the ROSCA half of the complaint too — obstructed cancellations drew the same complaint as the data sharing, so if your own subscription flow buries the cancel button, you carry a version of the exposure your advertisers do. The 2-0 vote with two state AGs alongside is the pattern to price in: these cases now travel with state coalitions, and those don’t wind down when Washington changes.