> ## Content Index
> Fetch the complete content index at: https://adjacent.media/llms.txt
> Use this file to discover other available public pages before exploring further.

# The Adjacent Brief — July 22, 2026
- URL: https://adjacent.media/briefs/2026-07-22/
- Published: 2026-07-22T14:15:04.000Z
- Updated: 2026-07-22T14:15:05.000Z
- Description: Google’s AI Mode is pulling roughly 40% of human traffic away from publisher sites, per Cloudflare data covering June 2025 through April 2026, while simultaneously ranking as the second-most cited domain in its own AI search results.
- Author: Jonathan Greene
- Tags: #brief

**TL;DR:** Google's AI Mode is pulling roughly 40% of human traffic away from publisher sites, per Cloudflare data covering June 2025 through April 2026, while simultaneously ranking as the second-most cited domain in its own AI search results. Separately, 70% of top retailers lack the structured data infrastructure to appear in agentic commerce at all, and South Korea announced a $540B chip hub it doesn't yet have the power grid to run.

## Worth Reading

- [Google Maps quietly demoted restaurant star ratings — and nobody noticed](https://hackaday.com/2026/07/20/google-maps-killed-the-restaurant-star/?ref=adjacent.media) — The platform that made star ratings culturally dominant is now burying them; another quiet edit to how discovery works.
- [AT&T's subsidy restructuring would hit low-income internet customers hardest](https://www.androidcentral.com/phones/att/at-and-t-could-raise-home-internet-prices-for-the-people-who-can-least-afford-it?ref=adjacent.media) — The people most dependent on carrier subsidies have the least leverage to push back.
- [Netflix CTR up 40%, session length down 18% — something in the recommendation engine is breaking](https://open.substack.com/pub/lewislin/p/m41h-netflix-ctr-is-up-40-session) — Clicks are easy to optimize; time spent is harder to fake.
- [A former Google DeepMind employee tried to stop its Pentagon deal — then quit](https://open.substack.com/pub/transformernews/p/i-tried-to-stop-google-deepmind-pentagon-deal-then-quit) — First-person account of what internal dissent at a frontier AI lab actually looks like in 2026.
- [LG ships monitors with McAfee pop-up adware pre-installed](https://www.theverge.com/tech/967983/lg-monitors-mcafee-adware-gamers-nexus?ref=adjacent.media) — Hardware-bundled adware was embarrassing in 2005; in 2026 it's a brand-destroying decision.
- [Mullvad VPN loses customers after co-founder donates millions to a populist party](https://www.theregister.com/software/2026/07/21/some-mullvad-vpn-customers-tunnel-for-exit-after-co-founder-donates-millions-to-populist-party/5275012?ref=adjacent.media) — Privacy products sell on founder trust more than most categories; political donations break that faster than a data breach.
- [Samsung grows in India despite repeated price hikes](https://www.sammobile.com/news/samsung-only-mobile-brand-growing-india-despite-multiple-price-hikes/?ref=adjacent.media) — Brand equity absorbing margin pressure in the one emerging market that still has room to run.

## Connected World

**A $540B ambition running on a grid that isn't there yet**

South Korea announced a plan to build a [southwest semiconductor manufacturing hub requiring $540B in investment](https://thenextweb.com/news/south-korea-honam-chip-hub-power-water?ref=adjacent.media) — but the Honam region lacks the power and water infrastructure to support it. The sequencing problem is real: chip fabs are among the most energy-intensive industrial facilities on earth, and South Korea's grid is already under pressure. The announcement lands as institutional investors are watching Korean semiconductor stocks as a leading indicator for global AI chip demand, which means the gap between the ambition and the infrastructure will get priced in quickly if the build timeline slips.

**The resilience question hyperscalers can't answer cleanly**

Google Cloud's outage this week exposed something more uncomfortable than downtime: customers still [can't accurately read hyperscalers' real resilience posture](https://www.theregister.com/off-prem/2026/07/21/google-cloud-outage-shows-its-still-hard-to-understand-hyperscalers-real-resilience-regimes/5275405?ref=adjacent.media), even after years of SLA documentation and architecture guides. The opacity is deliberate — uptime guarantees are commercially negotiated rather than engineering-verified. For enterprise buyers, the practical consequence is that redundancy planning has to be done defensively, treating the hyperscaler's stated resilience numbers as a floor estimate and building capacity well above them.

**A dual-battery iPhone is a design concession that engineers adopt despite its drawbacks, never a genuine feature.**

iOS 27 code references [an iPhone variant with two separate batteries](https://www.macrumors.com/2026/07/20/ios-27-dual-battery-iphone/?ref=adjacent.media). The charitable read is thermal management; the honest read is that Apple hasn't solved battery density at the size and thinness it wants, and splitting the cell is the engineering workaround. Whether this reaches production matters less than what it says about the physical limits the industry is working against as devices get thinner and AI inference loads get heavier.

## Culture & Signal

**Google is draining the web it depends on**

The New York Times, citing Cloudflare data, reports that [human traffic to publisher and business sites fell roughly 40%](https://www.nytimes.com/2026/07/20/technology/google-ai-open-web.html?ref=adjacent.media) (paywall) between June 2025 and April 2026, attributed directly to Google's AI Search serving answers in-place rather than routing users to sources. AI Mode ingests the web, synthesizes an answer, and the click never happens. Publishers lose the ad impression; Google keeps the session. The pattern has been building since AI overviews launched, but the Cloudflare figure puts a number on what was previously anecdotal complaint.

The harder problem for Google is structural. The web it's summarizing was built by publishers who depended on search traffic to fund the reporting, research, and original work that makes the answers worth anything. Erode the economics long enough and the training corpus degrades. Google's answer to this — that it's creating a "new kind of traffic" — has not yet been demonstrated at the scale that would offset the losses.

**Solar's policy problem has nothing to do with the panels**

States are [restructuring net metering rules in ways that reduce the financial return on residential solar](https://www.nytimes.com/2026/07/21/upshot/solar-panels-prices-batteries.html?ref=adjacent.media) (paywall), shifting costs back to panel owners and making payback periods longer. The technology works fine; the problem is that utilities and regulators are repricing the implicit subsidy that made rooftop solar math pencil out for middle-income households. Consumer energy choices that seemed locked in are proving more fragile to regulatory revision than buyers anticipated when they signed the contracts.

## The New Consumer

**Google citing Google is a structural advantage nobody can replicate**

Search Engine Journal's analysis finds that [Google is the second-most cited domain in its own AI Mode results](https://www.searchenginejournal.com/google-is-ai-modes-no-2-cited-domain-data-shows/582941/?ref=adjacent.media), primarily through Business Profiles and Product Knowledge Panel cards — structured data Google controls. For brands, this is the key mechanic to understand: AI Mode doesn't just favor structured data, it favors *Google's* structured data. That means your Google Business Profile is now a more consequential asset than your homepage for AI search visibility, and optimizing it is no longer optional SEO housekeeping.

**Automation is hollowing out comment sections**

Rachel Karten's piece in Milk Karten examines [how Manychat's DM automation tools are degrading Instagram comment sections](https://open.substack.com/pub/milkkarten/p/manychat-ruining-comment-sections) — brands prompt users to comment a keyword, which triggers a bot-delivered DM, and the comment thread fills with identical one-word replies that crowd out actual conversation. The immediate problem is aesthetic; the downstream problem is that comment engagement becomes a metric that means nothing. Automation optimizes for a platform's visible signals while hollowing out the underlying behavior those signals were meant to measure. Creators who built audiences on perceived community are now presiding over comment sections that look active and feel empty.

**AI-generated fake bird sightings constitute a genuine data quality problem.**

The Next Web reports that [AI-generated false bird sightings are polluting citizen science databases](https://thenextweb.com/news/ai-slop-fake-bird-sightings-citizen-science?ref=adjacent.media) in Brazil, with fabricated species location data entering records that researchers rely on for conservation decisions. The mechanism — plausible-looking AI outputs entering open-contribution platforms — is not specific to ornithology. Any citizen science or crowdsourced data system with low submission friction and limited verification is exposed to the same problem. The databases that power ecological research, local weather modeling, and public health surveillance all share the same architecture.

## Commerce Rewired

**Most retailers aren't in the agentic commerce game yet — and don't know it**

Search Engine Journal reports that [70% of top retailers lack the structured data and API infrastructure](https://www.searchenginejournal.com/70-of-top-retailers-are-invisible-to-agentic-commerce-heres-why/581198/?ref=adjacent.media) that agentic commerce tools require to surface their products. When an AI agent is tasked with buying something, it doesn't browse — it queries structured feeds. Retailers without product schema markup, real-time inventory APIs, and machine-readable pricing are invisible to the transaction. The companies that solved this for Google Shopping a decade ago have a head start; the ones that treated SEO as an afterthought are in the same position now, one layer deeper.

**Samsung's memory pricing test is a leverage play**

Chinese smartphone manufacturers are [pushing back against Samsung's latest DRAM price increase](https://www.sammobile.com/news/chinese-phone-makers-are-rejecting-samsungs-latest-memory-price-increase/?ref=adjacent.media), declining to absorb the hike during a period when their own device margins are already compressed by tariff exposure. The standoff tests whether Samsung has enough pricing power in memory to push costs downstream — or whether the concentration of Chinese OEM volume gives buyers real negotiating leverage. If Chinese manufacturers hold, Samsung either absorbs the margin hit or loses share in the segment of the market growing fastest.

## Brand & Growth

**The US market is becoming a selective filter for global brands**

Yanko Design catalogs how [Polestar, OnePlus, DJI, and a cluster of other globally prominent brands](https://www.yankodesign.com/2026/07/20/from-polestar-to-oneplus-to-dji-the-us-is-becoming-a-harder-place-for-globally-influential-brands/?ref=adjacent.media) are being squeezed out of or hobbled in the US market through a combination of tariffs, security reviews, and retail channel pressure. The mechanism varies by brand — DJI faces export restrictions, Polestar faces EV tariff structures, OnePlus faces carrier relationships that never materialized — but the outcome is consistent: companies with strong global brand equity and genuine product quality are finding the US a market they can't fully operate in. For brand strategists, the relevant question is what it means when the world's largest consumer market becomes selectively inaccessible, and whether building US-market relevance is still a viable ambition for a globally-oriented brand that can't guarantee US distribution.

---

*\*AI-generated fake bird sightings constitute a genuine data quality problem.\**