00The week in one paragraph
This was the week the open web’s referees stopped pretending to be neutral. Publicis took the industry’s last independent identity layer in a $2.167 billion acquisition of LiveRamp — and Omnicom and WPP promptly walked — while Nielsen’s CEO defended its $2.15 billion purchase of DoubleVerify as “playing offense,” meaning both of the largest independent verification firms have now left the public markets within a year. The fight over ads served to AI agents turned from thought experiment into live commercial dispute: Perplexity blocked Time’s markdown ads within days and threatened trust-score penalties for publishers who follow. TollBit put numbers on the scraping imbalance — European publishers take four times the AI scrapes of North American sites and get one human visit per 179 bot visits — while the Stealth Bot Prohibition Act gained its cost case: Politico spends 25% of its hosting budget on bot management. Publisher earnings showed the Google replacement is a portfolio, not a product — People Inc. is down to 21% Google traffic, Dow Jones takes 81% of revenue from circulation — and Wall Street marked down nearly every open-web ad tech stock regardless of growth while forecasting Meta, Amazon and Google at roughly 58% of US ad spend, just as OpenAI began assembling an SMB ads unit that needs no publisher inventory at all. The through-line for publishers is control: over who scrapes you, who verifies you, and who owns the pipes your addressability runs through.
01Perplexity blocks Time’s agent-facing ads — and claims the right to set the rules
Three weeks after Time began selling ads to AI bots — the 30 July move we covered in our 2 August issue — the experiment met its first wall. Time had started embedding ads in the markdown versions of its pages, the versions AI agents actually read, built with ad tech firm Mobian and with Ally Bank and the Project Management Institute as early advertisers. Perplexity blocked all markdown advertising on Time.com from influencing its agents and search results, and chief communications officer Jesse Dwyer warned that publishers deploying such ads risk reputation downgrades and trust-score penalties in Perplexity’s index, calling the format “deceptive.” Mobian’s CEO countered that the ads are “current, sourced, brand-verified information at the moment it is consuming the page”; Perplexity’s position is that readers cannot tell sponsored content from reporting. It is the first real test of whether publishers can monetise agent traffic on their own terms — and the ruling, for now, came from the platform (Digiday, 11 August).
Publisher takeaway
The precedent is the story: when an agent reads your page, the publisher did not get to decide what was on it. If you are piloting agent-facing formats — markdown ads or anything adjacent — structure them as their own line items and ad units in GAM, fully separable from your human-facing stack, so a platform block becomes a per-surface pause rather than a fire drill. Before you launch anything, pull your referral logs and size your Perplexity exposure specifically; the retaliation risk is proportional to what that platform sends you, and for many publishers it is currently small enough that testing is cheap. Label sponsorship unambiguously in the agent-readable version — “deception” is the lever every platform will reach for, and clean disclosure removes it. And put platform-penalty risk into your contract with the ad tech intermediary: who carries the cost if a trust-score downgrade hits your whole index presence, not just the ad revenue? One platform’s policy is not the market’s, but assume others will copy whichever position wins. Watch this dispute rather than joining it — the rules of agent monetisation are being written in public, cheaply, on someone else’s traffic.
02The referees pick sides: Publicis takes LiveRamp, Nielsen defends buying DoubleVerify
Two of ad tech’s independent layers changed hands in the same news cycle. Publicis Groupe’s $2.167 billion acquisition of LiveRamp removes the industry’s last genuinely independent identity layer, and Omnicom and WPP have already begun abandoning it to accelerate their own proprietary identity stacks; as GrowthCode’s Jonathon Shaevitz argues, code neutrality is not incentive neutrality — roadmaps and pricing follow corporate interest regardless of architecture (AdExchanger, 10 August). Meanwhile Nielsen’s $2.15 billion acquisition of DoubleVerify — announced 6 August and covered in our 9 August issue — got its public defence: CEO Karthik Rao told AdExchanger “We’re the referee, we tell the score, and people don’t like it – but it comes from a position of strength, and with this acquisition, we’re playing offense.” The deal ends DV’s roughly five-year run as a public company, closes in Q1 2027 with Mark Zagorski staying on, and — coming less than a year after IAS was taken private by Novacap at $1.9 billion — takes both of the largest independent verification firms off the public markets. Rao’s independence argument rests on Nielsen owning no media inventory (AdExchanger, 13 August).
Publisher takeaway
Run the dependency audit this month, not after the first roadmap change. On identity: if RampID or ATS sits anywhere in your addressability stack, you now depend on infrastructure owned by an agency group that competes for the same budgets your inventory does. That is not a reason to rip it out; it is a reason to price the dependency — know what share of your addressable revenue routes through it, and have a second identity partner at least in test so a pricing or roadmap shift is a negotiation rather than a hostage situation. On verification: Rao’s no-media-inventory argument is fair as far as it goes, but the practical change is fewer independent parties to appeal to when a score goes against you, with less public disclosure once DV leaves the markets. Get your methodology questions in writing before the Q1 2027 close, while the company is still publicly answering them. And keep your own log-level evidence — ad server logs, IVT reports, viewability measurement — because when the referee and the scorekeeper share a roof, your appeal rests on the records you kept yourself.
03European publishers take four times the AI scraping — and robots.txt is ignored
TollBit analysed data from 40 scraping vendors across 3,906 publishers, 456 of them European, and put numbers on what EU publishers have described anecdotally for a year. Median AI scrapes per site were four times higher on European sites, which received one human visit per 179 bot visits, and their robots.txt no-scrape instructions were ignored nearly three times more often than North American sites’. The trade is deteriorating fast: the scrape-to-referral ratio went from 150:1 in Q1 to 227:1 in Q2 2026, while just 0.05% of external referrals to European publishers came from AI apps versus 0.16% in North America (Digiday, 14 August).
Publisher takeaway
Treat TollBit’s figures as a claim that directs your measurement, not one that settles it — TollBit sells scraping monetisation, and Cloudflare and DataDome see different regional patterns, so methodology matters. The action is the same either way: measure your own logs this quarter rather than trusting anyone’s narrative. Reconcile your robots.txt directives against server logs by user agent and ASN, and find out which crawlers you formally block are still being served content — that gap is a fixable engineering problem, and it is the difference between a policy and a wish. Put a cost number on the scraping you carry: bandwidth, CDN egress, bot-management tooling, the infrastructure share that bots consume. If you are a European publisher, that number belongs in three places — your licensing conversations, because a documented cost strengthens a price; your trade-body submissions, because EU enforcement runs on evidence from affected parties; and your own build-or-block decision, because a 227:1 scrape-to-referral ratio means the “we crawl, we send you traffic” bargain is no longer a bargain you need to preserve. Separate bot from human traffic in every audience figure you give a buyer, before buyers start asking.
04The Google replacement is a portfolio, not a product
This week’s publisher earnings round-up shows no single Google replacement emerging — so publishers are assembling portfolios instead. People Inc. has cut Google search to 21% of traffic (from 25% last quarter) and grew non-session-based revenue from 39% to 43% of digital revenue; CEO Neil Vogel: “We are nearly on the other side of search being a material driver of value for us. But we’re not there yet.” Dow Jones now takes 81% of revenue from circulation against 17% from advertising. USA Today Co. grew digital-only subscription revenue 6.8% while digital advertising fell 9.2%. And Ziff Davis reports around 50% of relevant Google queries now trigger AI Overviews, up from 36% a quarter ago (Digiday, 13 August).
Publisher takeaway
Nobody found one replacement for Google; they found six smaller ones and accepted the operational cost of running all of them. That acceptance is the actual decision in front of you. Start by measuring your Google dependency the way these companies now report theirs — not traffic share but revenue exposure: what percentage of your digital revenue requires a Google-originated session to exist? People Inc.’s 43% non-session-based figure is the benchmark worth stealing, because it reframes the goal from “replace the traffic” to “make revenue that doesn’t need the traffic.” Subscriptions, licensing, newsletters, events, commerce and direct-sold sponsorships all qualify; open-auction programmatic on search-referred sessions does not. Then accept that a portfolio costs more to run than a pipeline — each channel needs its own owner, its own measurement, and a tolerance for individually unimpressive numbers that only work in aggregate. The Ziff Davis figure is the clock on this work: AI Overviews on half of relevant queries means the erosion is not a scenario you are planning for, it is the baseline you are already operating in. Budget your diversification against that trajectory, not last year’s referral report.
05The Stealth Bot Prohibition Act gets its cost case
The Stealth Bot Prohibition Act — introduced in late July by Reps. Valerie Foushee (D-NC), Laurel Lee (R-FL) and Gus Bilirakis (R-FL), and covered at introduction in our 26 July issue — got the follow-through analysis this week, and with it the number that makes the case. The bill would require AI web crawlers to disclose their identity and purpose to website hosts, with $53,000 fines enforceable by the FTC and state attorneys general, targeting bots that masquerade as human visitors to bypass robots.txt, then scrape and resell content. The cost case is concrete: Politico allocates 25% of its hosting spend to bot management. News/Media Alliance CEO Danielle Coffey called it “hygiene and vegetables” — unglamorous, and structurally necessary (AdExchanger, 10 August).
Publisher takeaway
Politico’s figure is the template: 25% of hosting spend is a legible, invoiceable number, and it does more work in a licensing negotiation, a trade-body submission or a board deck than any amount of adjectives about scraping. Produce your own version this quarter. Pull your infrastructure bills apart — CDN, WAF, bot-management tooling, the compute and bandwidth share your logs attribute to automated traffic — and write down what bots cost you as a percentage of hosting. That number is useful whether or not the bill passes: it is the evidence lawmakers and enforcers will ask affected parties for, it is a cost line you can put on the table when an AI company wants access, and it is the internal justification for the engineering time that blocking properly requires. Because that is the other half of the follow-through: disclosure mandates only help publishers whose blocking actually works. If your robots.txt and your server logs disagree about which crawlers get through — and for many publishers they do — fix the enforcement gap now, so that when identification becomes law you are metering access rather than discovering leaks. Hygiene and vegetables, exactly as Coffey says. Eat them.
06Agentic buying meets supply-path optimisation — prove your path or lose it
Goodway Group’s Andrea Kwiatek offered the week’s most useful counterweight to agentic hype: the agency is feeding campaign briefs to buyer agents and evaluating seller-agent recommendations — while simultaneously reducing its SSP partnerships through rigorous SPO validation. Her framing: “SPO is not about cutting out partners. It is more about understanding what those partners are able to support in terms of outcomes.” Her warning is that agentic buying risks becoming “another media-buying black box” without agent-to-agent transparency (AdExchanger, 14 August).
Publisher takeaway
Read the two halves together: the same agency testing seller agents is actively cutting SSPs, which means the standard Kwiatek describes — what can this partner evidence in terms of outcomes — is the standard your supply will be held to, and increasingly by software rather than by a planner you can take to lunch. The dropping is automated now, and automated dropping is faster and less forgiving than the human kind. So make your path documentable before an agent asks. That means clean sellers.json and ads.txt with no stale or mystery entries, a complete supply chain object on every bid request, and a straight answer to the fee question: how many hops between the buyer’s money and your ad server, and what does each one take? If you cannot answer that today, your SSPs can — ask them, in writing, and consolidate away from the paths nobody will explain. Then go one step further and ask each SSP what it actually surfaces to buyer-side agents about your inventory, because that machine-readable representation is becoming your sales deck. Publishers who can evidence fees, hops and outcomes get shortlisted whether the buyer is a human or an agent. The ones who cannot get dropped without a meeting.
07Walmart’s “independent” gaming publication cuts its entire editorial team
Restart.run, launched in December 2024 under the tagline “independent gaming news powered by Walmart” and operated by gaming agency Moonrock, laid off its entire editorial team — including editor-in-chief Brandy Berthelson — from a staff page of roughly ten names. Neither Walmart nor Moonrock commented, and Berthelson said she had no clarity on what performance thresholds would trigger a funding withdrawal (Digiday, 13 August).
Publisher takeaway
The detail that should stick is the editor-in-chief not knowing what threshold would end the funding — because that is the structural condition of brand-funded editorial, not a Restart.run quirk. Brand money sits in a marketing budget, not a P&L you control, and it is reviewed on a cycle nobody shows you, by people whose success metric is not your publication’s health. None of that makes it bad money; it makes it money you must contract for differently. If you are taking or weighing brand funding, negotiate the exit before the entrance: performance thresholds in writing, so you know what number keeps the lights on; a defined review calendar, so the decision points are visible; a notice period measured in months, not a Tuesday announcement; and wind-down funding that covers severance and an orderly transition. Then run the operation as if the sponsor’s quarter could turn, because it can — keep building programmatic, subscription or direct revenue alongside the funded work so the publication has a floor that one budget review cannot remove. A sponsor who balks at putting thresholds and notice in writing is telling you, before launch, exactly how the ending will go.
08Wall Street marks down ad tech — growth or no growth
Q2 earnings season delivered a verdict that had little to do with performance. AppLovin grew revenue 53% and fell 19.7%; The Trade Desk grew 3% and fell 22%; Criteo declined 11% and fell 24%; Taboola grew 2.4% quarterly and fell 27.5%. PubMatic (+20.8%) and Magnite (+8.6%) were the sell-side exceptions. Madison & Wall’s Luke Stillman notes the walled gardens “were all still very solid,” with Meta, Amazon and Google expected to take roughly 58% of US advertising this year, up from 56%. As one analyst put it, investors are “pricing these companies on what it expects to happen next, not the actual facts of what just happened” (Digiday, 13 August).
Publisher takeaway
What investors expect to happen next, apparently, is more concentration — and that expectation reaches your ad stack whether or not you own a single share. A vendor whose stock has been marked down 20–27% despite growing is a vendor under pressure to cut costs, raise take rates, or sell itself, and each of those outcomes lands on publishers as degraded support, changed pricing, or a migration you did not schedule. So treat partner equity health as an operational signal: know which of your demand and technology partners just got repriced, and hold a contingency plan for your top three — where the demand reroutes, what the integration lift is, how long the switch takes. The PubMatic and Magnite exceptions are worth noting without over-reading: the market currently rewards the sell-side platforms, which supports rather than undermines the case for maintaining more than one. And take the 58% forecast seriously as a planning number — if the walled-garden share of US spend keeps climbing, the open-web pool your programmatic revenue draws from is shrinking in relative terms every year, which is the macro argument for the direct-sold, first-party and non-advertising revenue work elsewhere in this issue. The scoreboard is not about last quarter. Neither should your planning be.
09IAB Tech Lab opens GPP and data-deletion overhauls for public comment
The IAB Tech Lab put the Global Privacy Platform and Data Deletion Request Framework v2.0 out for public comment, with the window closing 11 September 2026. This is not a cosmetic pass: the GPP changes remove the MSPA state-by-state coverage approach, eliminate Service Provider and Opt-Out Option Modes, drop secondary usage consents and simplify notice and choice fields, while DDRF v2.0 clarifies the JWT definitions for identity and deletion requests and improves result feedback. CEO Anthony Katsur frames it as making compliance “more consistent, transparent, and practical” (PR Newswire / IAB Tech Lab, 11 August).
Publisher takeaway
Simplification still means engineering, and the publishers who read the draft during the comment window get to shape the work; the ones who read it after ratification just inherit it. Three concrete steps before 11 September. First, inventory what you actually run: if your consent implementation leans on MSPA’s state-by-state coverage, Service Provider Mode or Opt-Out Option Mode, those code paths are slated for removal, and you need to know how deep they reach into your GAM key-values, your prebid consent handling and your downstream vendor contracts. Second, put the question to your CMP vendor in writing now — what is their migration plan, on what timeline, and who pays for the transition work — because CMP migration timelines have a habit of landing in Q4, next to your best revenue weeks. Third, if the draft breaks something specific in your setup, file the comment; a standards body mid-consultation is the cheapest possible venue to get a problem fixed, and “practical” only describes the outcome if practitioners actually responded. The DDRF changes are quieter but real: tightened identity and deletion JWTs mean your deletion-request plumbing gets retested, not just re-documented. Budget an engineer’s week, not a memo.
10OpenAI starts pouring a fourth walled garden — the SMB ads unit
OpenAI is standing up a dedicated SMB advertising unit: hiring data scientists, growth leads and analytics engineers on packages up to $515K, while outsourcing SMB sales to third-party vendors under a senior vendor manager in Dublin — the exact long-tail playbook Google and Meta have run for years. Building the data infrastructure from scratch alongside the hiring suggests a near-term launch. eMarketer’s Nate Elliott supplies the context: SMBs make up the majority of ad revenue for Google and Meta — which is the point. A fourth long-tail ad machine is being assembled, and it runs on an owned surface and an owned audience, with no need for publisher inventory (Digiday, 10 August).
Publisher takeaway
The tell is the Dublin vendor manager: outsourced long-tail sales with in-house oversight is precisely how Google and Meta built the self-funding machines that everything else sits on, and OpenAI copying the org chart means it intends the same economics. For publishers the threat is not that ChatGPT ads compete for your impressions — they will not run on your pages at all — it is that they compete for your advertisers, specifically the local and SMB accounts in your direct book who want outcomes, tolerate automation, and follow ease of use wherever it goes. Those advertisers left print for Google search and left Yellow Pages for Facebook; a fourth simple, self-serve surface gives them a fourth exit. Defend that book deliberately: audit which of your direct advertisers fit the SMB profile, and make sure your own self-serve or managed-lite offering is genuinely easier than what they will be pitched — first-party audience targeting, evidence of outcomes, a human who answers. And note what this says about the open web’s trajectory: another walled garden is being poured while publishers argue about crawler policy, and it will absorb budget without ever appearing in your bid stream. The competition for SMB money is now four platforms deep and none of them need you.
11CMOs can’t connect AI visibility to sales — and that gap is a brief
73% of CMOs have invested in AI visibility monitoring, but tools like Scrunch, Profound and Semrush can track citations in AI Overviews and ChatGPT without connecting them to revenue, forcing marketers to triangulate conversion data, traffic and custom marketing-mix models. Skyword CEO Andrew Wheeler: “Citation alone isn’t good enough.” Roast’s John Barham: “There is no one tool out there that can paint you a picture of the universe” (Digiday, 13 August).
Publisher takeaway
Read this as a sales brief, not as a marketing-department curiosity. The buy side has a measurement hole it is actively spending to fill — nearly three-quarters of CMOs bought tooling that stops at the mention — and the same gap sits under every AI-era content and licensing pitch you will make. Whoever can evidence a downstream commercial outcome, rather than a citation, gets the budget. Publishers are unusually well placed to supply that evidence if they choose to build it: you own the first-party data on what your audience does after exposure — the newsletter signup, the return visit, the commerce click, the registered account — and none of it requires a browser’s or a platform’s permission to measure. So package it. If you are selling sponsored content, content licensing or first-party audience products, put a post-exposure outcomes section in the proposal: what your logged-in data can show happening after the mention, at what match rate, on what timeline. That turns the conversation from defending a CPM into selling into documented demand. The vendors will eventually close this loop; until they do, a publisher with clean first-party outcome data is selling something 73% of CMOs have already budgeted for and cannot currently buy.
12AppLovin’s 8% of online ad spend sits in a box buyers can’t inspect
AppLovin’s share of online ad spend has gone from 1% two years ago to roughly 8%, even as its market cap fell from $245 billion to $106 billion. Its consumer platform mirrors Performance Max and Advantage+ — pick Discovery or Prospecting campaigns, and very little else. Early adopters like AS Beauty confirmed incremental sales via holdout tests, though Haus analyst Olivia Kory says performance is “cooling off a bit,” and buyer scepticism persists over aggressive SDK data collection and device fingerprinting around Apple’s ATT. One buyer, quoted in the piece: “What’s the catch? There has to be a catch” (AdExchanger, 12 August).
Publisher takeaway
The pattern to watch is not AppLovin specifically — it is that another several percent of budget has moved into a box buyers cannot inspect, and it keeps happening: Performance Max, Advantage+, now this. Each of these products absorbs open-web spend by promising outcomes without placements, and each one shrinks the pool your transparent, auditable inventory competes for. Note the sourcing discipline the story itself models: the incrementality wins are advertiser-reported holdout tests, the “cooling off” is one analyst’s read, and the fingerprinting concerns are reported buyer scepticism — treat all of them as claims, and apply the same standard to any performance number a platform shows you about your own inventory. The practical response is to lean into the one thing the black boxes structurally cannot offer: inspectability. When buyers eventually tire of asking “what’s the catch?” — and the ATT and SDK questions suggest some already are — the safe harbour is supply where placement, context and path are verifiable. Make yours that supply: documented supply chain, log-level reporting you will actually share, placement-level transparency in your direct and PMP deals. You cannot out-algorithm a black box on modelled conversions. You can be the inventory a nervous buyer can actually see.
13Walmart Connect ships negative keywords — seven years after Amazon
Walmart Connect added negative keywords to sponsored products, available in Ad Center and through Pacvue, Skai, Quartile, Teikametrics and DataCaciques — a control Amazon has offered since 2019 and Sam’s Club Connect for years. Podean’s Chris Sheldon described the campaign to get it shipped: “This was not some quiet little request… it really has been a begging to get them to release this” (Digiday, 12 August).
Publisher takeaway
File this under ammunition. Retail media keeps getting positioned as the mature, accountable alternative to the open web — the sophisticated destination budgets should migrate toward — and here is a major network shipping, in 2026, after years of advertisers begging, a control that programmatic display considered table stakes a decade ago and that your GAM stack has carried for as long as you have run it. The next time a budget conversation frames your inventory as the risky option against retail media’s, this is the concrete counterexample: negative targeting, brand-safety exclusions, granular reporting and placement control are things you offer today, not roadmap items advertisers must campaign for. Use it constructively rather than smugly — build a one-page capabilities comparison for your sales team that lists the controls your stack offers against what the major retail networks actually support, because most buyers have never seen that laid out and the gap is wider than the narrative suggests. And if you operate or participate in a retail media network yourself, take the operational lesson from the other direction: advertiser controls are not a nice-to-have you defer, they are the credibility of the channel, and “a begging” is not a process any partner should describe.