// The Adjacent Brief

TL;DR: Brands are reworking their content to get cited by AI answer engines. Digiday reports on brands using recurring content series and cultural hooks like Jimi Hendrix, and Ahrefs data shows ChatGPT's most-cited sites lean toward primary sources. Google appears to be serving degraded results to rank-tracking tools, which makes conventional SEO measurement harder to trust.

Worth Reading

Brand & Growth

AI engines cite the original source and skip the summaries built on it

Ahrefs data shows ChatGPT's most-cited sites tilt toward primary sources rather than aggregators and roundup pages. That changes what content investment buys. A brand that republishes other people's findings is competing for citations it is unlikely to win. A brand that originates data, documentation, or a distinct point of view gives the model something to point at.

Digiday's reporting fits that pattern. Brands are building recurring content series, and in one case leaning on Jimi Hendrix, to improve AI search performance. The series format matters more than the celebrity. Consistent, structured, ownable content is easier for a model to attribute than a one-off campaign page. For CMOs, that means budgeting for content that works as reference material and not only as campaign creative.

Rank trackers stop being ground truth

Search Engine Journal's analysis argues that Google is poisoning results served to rank trackers. If that holds, the tools many SEO teams use to justify budgets are measuring a version of search that real users don't see. Search is one problem and vendor dependency is another. Any growth team whose reporting rests on third-party scrapers of a platform that doesn't want to be scraped should expect the same from AI answer engines. The defensible metrics are first-party ones: referral traffic, branded search volume, and conversions you can trace.

Enterprise AI standards get set in the team chat

Forrester argues that multiplayer AI is becoming the enterprise standard by accident. Shared AI sessions get adopted inside collaboration tools teams already use, before anyone writes a policy for them. Anthropic's Claude Code leadership projected this week that roughly 90% of workplaces will adopt AI collaboration tools. IT and brand leaders face a different decision than whether to allow shared AI workspaces: which vendor's defaults become the governance model once usage is entrenched.

Connected World

Consoles have been pronounced dead before

Business Insider research covered by The Next Web says home consoles face extinction as cloud gaming grows. Treat the forecast with the skepticism its predecessors earned. Google shut down Stadia in 2023 with the infrastructure in place but no reason for players to switch. Cloud gaming still has to answer the same questions: latency, ownership of purchased games, and whether a subscription catalog beats a box that just works. The growth figure measures streaming revenue. It doesn't show that players are leaving hardware behind. Platform holders are already hedging with hybrid offerings, and that hedge is the more realistic near-term shape.

Local consent decides where compute lands

Carbon County, Utah, a place built on coal, is pushing back on a nuclear- and gas-powered data center proposal. Energy-heritage communities are not automatically welcoming the next energy industry. Residents who have lived through boom and bust cycles know who keeps the upside. For hyperscalers and their power partners, site selection is becoming a community-relations exercise with real veto power, and it is happening alongside regulatory wins like this week's Georgia approval for Google's power deal.

An old renewables fight offers a contrast. Research shows painted turbine blades can save millions of birds, a cheap visual fix for one of wind power's most persistent siting objections. Low-cost mitigations that address a specific local harm win more permits than promises about national strategy. Data center developers have not yet found their equivalent.

Culture & Signal

Labor risk hides in the ingredient list

Verisk Maplecroft's analysis finds beauty ingredients among the highest-risk inputs for human rights abuses, with eight widely used cosmetics ingredients exposed to child and forced labor risk. Seafood has the same problem. Mongabay reports that Chinese squid tied to labor abuses finds an easy route into Western markets because catch from international waters moves without consistent verification.

In both cases, abuse sits several tiers below the brand, inside commodity inputs that change hands in bulk. Brands that audit their direct suppliers and stop there are exposed. Investigative reporting and risk-index providers are mapping those deeper tiers faster than many compliance teams are. Beauty and food brands making sustainability claims should assume the ingredient-level data will become public before they've seen it themselves.

Automating the backlog moves the error rather than removing the work

The UK Home Office's asylum AI failed one in five accuracy checks while processing backlogs kept growing. A 20% failure rate in a system that shapes asylum decisions can't be fixed by tuning. Every flagged output needs human review, so the promised throughput gain gets spent on checking the machine. Public-sector AI keeps getting procured for speed and evaluated, late, on accuracy. That order should be reversed.

The New Consumer

One paragraph replaces page one

When someone asks ChatGPT, Perplexity, or Gemini about a person or company, they get a single synthesized paragraph instead of ten blue links. The Next Web lays out what AI reputation management actually involves. There is no page two to push bad results onto. Executives, founders, and consumer brands should audit what these systems say about them the way they once audited Google results. Given the primary-source citation pattern noted above, the fix is to publish authoritative first-party material and keep everything visible.

AI fluency gets performed more than practiced

WIRED reports that men keep faking how much they know about AI: 51% of men versus 25% of women claimed familiarity with fictional AI leaders. That is stronger evidence than a typical attitude survey because it catches people claiming knowledge of figures who don't exist. For managers, self-reported AI proficiency is a weak hiring and staffing signal. Ask people to show the workflow.

Creator income follows a power law nobody markets

Forbes notes that making money at podcasting is easy for a few and impossible for most. Brands buying audio should read that as pricing information. The top tier commands premium rates because the middle can't sustain an audience. Niche shows with small, loyal audiences can be acquired cheaply, and few buyers are looking at them.

Machines & Minds

If you can't evaluate it, you haven't built it yet

Hamel Husain argues that "It's hard to eval" is a product smell. When a team can't say what a good output looks like, the cause is usually a fuzzy product definition rather than a missing metric. It is the most useful frame in today's feed for anyone buying or building AI. It also accounts for the Home Office failure above: an asylum assessment tool with no clear standard for a correct decision will fail audits that apply one.

It also accounts for the attention AI math results get. A quote making the rounds on Kottke says of OpenAI's latest release that "if a human did this, it would be an instant Fields" Medal. Set the hype aside. Math is where models keep posting credible wins because proofs can be checked. Capability claims can be trusted only where they can be evaluated.

McKinsey's practitioners on the AI-powered future of supply chain planning describe the same principle in an enterprise setting. Demand forecasts and inventory plans have measurable outcomes, so the value loop is visible quarter by quarter. Enterprise buyers should start AI deployments where the scorecard already exists.

Free app-building is a bet on software with one user

Google now lets people build personal Android apps for free with just prompts through AI Studio. The business case is the pull into Google's model ecosystem and developer surface rather than the apps themselves. The behavior fits this week's finding that most surveyed SMBs are building custom AI tools instead of buying them. If a usable tracker or internal tool takes a paragraph to describe, the low end of the SaaS market faces pressure from free, disposable software. Whether these apps get maintained past week two is the adoption test.

Commerce Rewired

The capex bill arrives before the receipts

The Register reports that AI infrastructure spending is forecast to outstrip expected revenue. Gary Marcus highlights a new report finding it remains "very difficult" to find materially measurable financial impacts from AI spending.

Cost alone doesn't settle it. The question is what the money buys and who gets paid. Suppliers of chips, power, and data center construction are booking real revenue now. The gap sits further down the chain, with enterprises that bought licenses without a defined outcome to measure, which is the Hamel problem at balance-sheet scale. CFOs should not pull AI budgets. They should fund the deployments with existing scorecards, like planning, forecasting, and support deflection, and make every other pilot name its metric before it gets a renewal.

19 articles across 6 themes · 15 sources · Powered by Folo + Claude