This week’s AI and marketing signals expose a dangerous paradox: the smarter our AI tools get, the more confidently wrong our decisions become — and two stories from opposite ends of the tech stack prove it with uncomfortable precision.
Cerebras Built the Enterprise Brain Notion Never Could
Cerebras — an AI chip company with no note-taking app to sell — published a production-ready blueprint that ingests an entire organisation’s Slack history, Confluence wikis, GitHub repos, and custom databases into queryable embedding space, making any institutional question answerable by anyone without touching individual platform APIs. This is the first credible, non-hype demonstration of the “second brain” concept at enterprise scale, and it works right now with tools most large teams already have. For any team running Confluence and Slack, the gap between “we have all this knowledge” and “anyone can query it” just collapsed from a multi-year roadmap item to a single engineering sprint.
Read the Cerebras blueprint this week and map their ingestion pipeline against your actual Confluence and Slack environment — the architecture is replicable today.
AI Advice Makes You Wrong — and Certain About It
A study covered by The Next Web — earning 309 HN upvotes and 174 discussion comments — found that AI-generated advice measurably reduces decision accuracy while simultaneously boosting the decision-maker’s confidence. Marketing teams are already using AI recommendations for budget allocation, copy selection, and audience targeting, which means this isn’t a future risk: it’s a live workflow problem producing confident, degraded decisions right now. The confidence boost is the most dangerous variable, because teams don’t self-correct when they don’t feel uncertain.
Audit one AI-assisted decision workflow this week and add a mandatory “what would make this recommendation wrong?” prompt before acting on any AI output.
Your AI Search Visibility KPI Is Probably Measuring Nothing
Search Engine Journal published a framework — backed by arxiv research — arguing that citation counting is the wrong metric for AI search visibility, because AI citations are not recommendations and search results are too session-unstable to snapshot reliably. Every marketing team currently reporting AI visibility as a citation-count KPI is tracking a metric that doesn’t correlate with whether AI actually recommends their product when users ask task-oriented questions. The gap between citation presence and recommendation quality is widening as models get better at synthesising without explicit attribution.
Replace one AI citation count report this week with a repeated-query test — run the same task-oriented question across multiple sessions and models, and track recommendation rate instead.
Read the full story →
Try it yourself →
W3C’s Attribution API Will Break Your Measurement Stack
Digiday published a practitioner explainer on W3C’s Attribution API proposals — the browser-level standard that will determine whether conversion measurement is legally and technically possible after third-party cookies fully disappear. Most marketing teams are still treating post-cookie attribution as an ad-tech vendor problem rather than a first-party data architecture decision they need to own, and the W3C spec defines the hard constraints on what will even be possible. Getting ahead of the proposals now means your measurement infrastructure doesn’t break on enforcement day.
Share the Digiday explainer with your analytics and ad ops teams this week and answer one specific question together: is your current attribution setup compatible with W3C’s privacy constraints?
Read the full story →
Try it yourself →
Perplexity Has a Hidden Power-User Layer Most People Miss
Fast Company identified three advanced Perplexity techniques — built directly into the prompt box — that go well beyond basic querying and are invisible to users who’ve only ever typed into the main search bar. Perplexity’s real differentiated value sits in Spaces, Connectors, and Artifacts, which transform it from a search alternative into a genuine research pipeline; practitioners missing those tools are evaluating a diminished version of the product when deciding whether it belongs in their stack. Worth noting: Perplexity is simultaneously facing HN criticism for using stealth, undeclared crawlers, so every advanced workflow built on it carries data-access risk that publishers are actively working to block.
Spend 20 minutes this week specifically exploring the prompt box tools Fast Company references — treat it as a skills audit before making any decision about Perplexity’s role in your content workflow.
Read the full story →
Try it yourself →
Join the discussion →
Claude Claims a Mathematical Breakthrough — Verify Before You Share
Anthropic’s Claude Fable model produced a claimed counterexample to the Jacobian Conjecture — a long-standing open problem in mathematics — earning 132 HN upvotes as the first widely-publicised case of a commercial AI model potentially contributing to unsolved mathematics. If the counterexample validates under mathematician review, it shifts AI’s mental model from “synthesis and retrieval tool” to “hypothesis generation tool” and accelerates every timeline for AI as a technical research accelerator. Gary Marcus has documented multiple cases of AI producing mathematically confident but ultimately wrong outputs, and a claimed counterexample to a famous conjecture is precisely the high-stakes scenario where that failure mode is most consequential.
Watch the HN thread and math community response over the next 48–72 hours before drawing any conclusions — the signal value is entirely contingent on expert validation, not the claim itself.
Read the full story →
Try it yourself →
Kagi Orion: The First Privacy Browser That Keeps Your Extensions
Kagi’s Orion browser launched publicly with 105 HN upvotes as a WebKit browser that uniquely supports both Chrome and Firefox extensions simultaneously — removing the single biggest reason practitioners stay on Chrome despite preferring privacy defaults. Every privacy browser before Orion forced a choice between extension ecosystems and privacy controls, which kept most marketing practitioners on Chrome regardless of preference; Orion removes that tradeoff for the first time, including for SEO, analytics, and ad tools with no Safari-compatible equivalents. Kagi is subscription-funded and small, so long-term compatibility guarantees carry platform risk that Chrome and Firefox simply don’t.
Run a 30-minute real-world test with Orion this week using your actual daily extension set — specifically verify your SEO, analytics, and ad tools before making any workflow commitment.
Read the full story →
Try it yourself →
Join the discussion →
Eminent Domain Is Now an AI Infrastructure Risk Factor
Fortune and the New York Times reported that power companies and tech firms are invoking eminent domain — including on Native American land — to seize territory for data centre expansion, legally framing it as “public use” in a new category of infrastructure conflict. AI infrastructure buildout timelines are now subject to litigation and regulatory risk that wasn’t priced into most cloud and AI service expansion projections, and 45 states have already enacted eminent domain reform laws that narrow the legal pathway for this acquisition strategy. For practitioners planning AI-dependent marketing infrastructure over a two-to-three year horizon, this is a latency and availability risk to track, not just a social ethics story.
Flag this as a macro risk signal when evaluating long-term commitments to AI infrastructure services — cloud capacity expansions in contested regions may face delays affecting SLA reliability and pricing.
Read the full story →
Join the discussion →
More from Rafal Reyzer
For deeper dives on AI and marketing strategy, visit my YouTube channel →
Rafal Reyzer
Hey there, welcome to my blog! I'm a full-time entrepreneur building two companies, a digital marketer, and a content creator with 10+ years of experience. I started RafalReyzer.com to provide you with great tools and strategies you can use to become a proficient digital marketer and achieve freedom through online creativity. My site is a one-stop shop for digital marketers, and content enthusiasts who want to be independent, earn more money, and create beautiful things. Explore my journey here, and don't forget to get in touch if you need help with digital marketing.