1 Google Ads will hold you to your own targets — and CPAs could double
Starting August 17, Google Ads will make budget-limited Target CPA and Target ROAS campaigns actually deliver at their stated targets. A campaign converting at a $5 CPA against a $10 target will drift toward $10; a campaign running 400% ROAS against a 200% target will slide to 200%. The change hits Search, Shopping, Performance Max, Demand Gen and Travel campaigns (App and Video are excluded), plus SA360, DV360, Editor and the API. Google announced it June 15 via Ads Liaison Ginny Marvin and shipped a Bid Target Adjustment Tool on July 6 — a six-week runway that is now half gone.
The era of setting a loose target and pocketing the overdelivery is ending: the target field is about to become the price you actually pay. Practitioners are calling it "one of the most self-serving Google-centric changes in years"; Marvin's answer — "set targets that actually mean something to your business" — is also the to-do list. Audit every campaign with a stale or padded target before August 17, because after that, Google will spend up to the number you wrote down.
2 Apple deleted the word "Apple" from where its ads can run
On July 14, Apple emailed advertisers that its Advertising Services terms change on July 28 — a 14-day runway. The key edit is in Section 6(a): ad delivery language that read "Apple software applications or Apple devices" now reads "devices, operating systems, software or web applications, or other platforms or properties." Analyst Eric Seufert called the new wording "conspicuously broad." It lands months after Apple confirmed ads in Maps and as its ad business approaches an estimated $8.5 billion.
Read plainly, Apple just wrote itself permission to run an ad network beyond its own devices — potentially the most interesting new supply source since retail media, wired to Apple's own attribution rails (AdAttributionKit already covers 77% of App Store referral conversions). Nothing is announced yet, but terms changes precede products. If you buy app installs or compete with Apple Search Ads inventory, put July 28 on the calendar and watch what ships next.
3 Chrome now shops without you
Google's July 14 Chrome drop gave US AI Pro and Ultra subscribers "auto browse" — an agent that books travel, fills forms, schedules appointments and shops within a set budget, hunting discount codes and even using saved Password Manager credentials along the way. It shipped alongside a Gemini 3 side panel, a UK desktop rollout, and native support for the Universal Commerce Protocol co-developed with Shopify, Etsy, Wayfair and Target. This is happening inside the browser that holds 66% of the market.
The visitor on your site may now be software with a spending cap. That makes structured product data, clean checkout flows and UCP readiness a merchandising discipline, not an engineering nice-to-have — agents can't be persuaded by a hero banner. One caution worth escalating: prompt-injection attacks against browsing agents succeed at rates up to 86% in benchmark tests, so brand and security teams need to be in the same meeting now.
4 The open web's ad supply fell 40% — and got more expensive
Benchmarking data from Ozone — the platform serving premium publishers including the Guardian, News UK and the Wall Street Journal — tracked roughly 20 billion impressions and found Q2 publisher ad requests down 32–37% year over year in the US and 39–41% in the UK, as zero-click AI search choked referral traffic. First-half programmatic spend fell 30.6%. But prices moved the other way: June eCPMs rose 30% in the UK and 7% in the US, and apps were the only channel to grow, with spend up 23%.
This is scarcity repricing, the same dynamic TV went through: fewer impressions, higher rates, buyers paying up for what's left. The 15-year assumption of infinite supply is dead. If premium display is becoming a scarce asset, buy it like one — on attention, context and first-party data quality rather than volume — and prioritize publishers with resilient logged-in, app and newsletter audiences over those still living on search referrals.
5 ChatGPT answers without referring — a 45,000-household study proves it
A Bocconi University study posted to arXiv on July 8 analyzed clickstream data from 45,386 US households and found ChatGPT sends an outbound click in just 5.2% of conversation sessions, versus 31.1% of Google queries. Fully 74.4% of ChatGPT-using households never generated a single referral in ten months. And each time OpenAI widened ChatGPT Search access, traditional search queries fell 9.4% on average — deepening to 17% after twenty weeks — with academic (−32.8%), reference (−26.5%) and news (−13.4%) referrals hit hardest.
The informational top of the funnel is eroding fastest, which means being the cited answer — answer engine optimization — is now the visibility game, and session counts understate your actual reach. One counterintuitive wrinkle for publishers: sites that block AI crawlers in robots.txt actually received more runtime referrals than sites that allow them (6.48 vs 4.48 per domain). Blocking training doesn't cost you citations — those are separate pipes.
6 Reddit said the quiet part: its AI deals might eat its ad moat
Reddit's EVP of ads monetization Roelof van Zwol admitted the company is "intensely debating" whether licensing its community data to OpenAI and Google undercuts the targeting advantage its ad business is built on, per Digiday's July 15 report. Reddit Max and Community Intelligence are both sold on the premise that Reddit's conversations offer signal no other platform can replicate — the same pool of posts being licensed to train the models. Van Zwol's defense: what's licensed is raw content, while ad targeting runs on engagement signal Reddit keeps to itself.
This is the honest version of a question every data-rich platform — and every retailer with a media network — now faces: is your data worth more as a moat or as a product? For advertisers, it's a due-diligence prompt. When a platform pitches "unique signal," ask how durable that uniqueness is once the same data is training the AI systems your customers ask for recommendations.
7 Criteo drew a 50% premium bid — private equity is buying the plumbing
A takeover approach for Criteo from two private investment firms, disclosed in early July at roughly a 50% premium to its share price, is still the most-read story in adtech this week. It landed in the same 48-hour window as Integral Ad Science installing a new CEO — one year after its own $1.9 billion buyout — and IAB Europe's AdEx report valuing Europe's digital ad market at €131.1 billion, with growth concentrated in video, social and retail media.
When public markets discount ad infrastructure, private equity buys it — and PE ownership changes vendor behavior: pricing discipline, roadmap cuts, consolidation. If Criteo, IAS or any PE-held vendor sits in your critical path for retargeting, measurement or commerce media, this is the week to diversify dependencies and make sure your contracts have real exit clauses. The pipes you rent are changing landlords too.
8 Amazon plugged 1,000 iHeart sellers into its ad machine
Amazon and iHeartMedia expanded their partnership: iHeart's 1,000+ local sellers now resell Amazon inventory across Twitch, Amazon Music, Fire TV and Alexa (they've resold Prime Video for three years), while Amazon DSP buyers gain access to iHeart's digital display, audio, podcast and creator inventory — targeted with Amazon's authenticated shopping signals and measured through iHeart's AudioGraph. iHeart reaches 90% of Americans monthly, so the local footprint here is enormous.
Retail data is colonizing audio and local — channels that were awareness-only line items are getting closed-loop, commerce-signal measurement. If audio has been parked in your "can't attribute it" bucket, the blocker is shrinking fast. And note the pattern: Amazon keeps renting other companies' sales forces and audiences to extend its ad machine into places its own DSP couldn't reach alone.
9 AI ad voices passed the blind test — and regional accents won
Azerion's "Voice of the Future" study, released July 14 with fieldwork across 3,000 UK respondents, found only 29% could correctly identify an AI-voiced ad; 37% confidently called it human. The AI voice matched or beat the human read on all six creative dimensions tested, including naturalness (78% vs 75%). The sharper finding: regionally accented AI voices drove 9% brand uplift versus 3% for the standard human read, and lifted recommendation intent to 33% versus 10%.
The production-cost excuse for running one generic voiceover everywhere just died. But the win isn't swapping your human voice for a synthetic one — it's localization at scale: accent-matched, dynamically assembled variants per market, which used to be unaffordable and now isn't. The fairness caveat matters too: personalized versions performed best (24% uplift) when listeners perceived the personalization as fair. Transparency is a performance variable now.
10 Dentsu turned creator marketing into an API call
Dentsu announced an API partnership with Meta that wires Creator Marketplace and Partnership Ads directly into its dentsu.connect operating system — social listening, creator selection and paid activation in one dashboard, blending Dentsu's data with Meta's. Chief media officer Toby Benjamin says what took days — connecting, contracting, getting partnership ads live — "now happens in hours." UK pilots have run; the US market is next. Context: L'Oréal now works with roughly 500,000 creators a year, and micro/nano influencers will claim nearly half of US influencer budgets this year per eMarketer.
Creator marketing is becoming programmatic plumbing, and the holdcos are racing to own the pipes before clients build their own. Once briefing-to-activation collapses from days to hours everywhere, speed stops being a differentiator — creator relationships, creative judgment and taste become the scarce assets again. Expect the other holding companies to announce near-identical API deals within the quarter, and expect the same plumbing to reach mid-size advertisers soon after.