Publicação de Łukasz Trzeciak

"UK regulators take a stand against Google's AI content scraping. News publishers can now opt out of feeding the search giant's algorithms. #AIregulation #DigitalPublishing #ContentScraping #UKRegulators @DemandNexus" https://lnkd.in/dt6WwaMr

Risky ad claims can drain your budget before results. Spot these red flags early: - Unrealistic promises - Vague benefits - Non-compliant language ZebraTruth scans your AI videos for compliance before you spend a dime. How do you catch risky claims?

44.2% of Perplexity's citation algorithm is driven by content freshness. Most law firms publish once and assume it keeps working. It doesn't — there's a ~30-day window before decay sets in. Our founder Jacob Shamis breaks down the mechanic and what to do about it. Worth a read if your firm is investing in AI visibility. https://lnkd.in/eGqvsQzG

Your firm's best content is expiring right now! Not being penalized. Not being removed. Just quietly losing citation weight in every AI engine that matters — ChatGPT, Perplexity, Gemini, Copilot. New research analyzing 216,000+ pages found that temporal freshness accounts for 44.2% of Perplexity's citation algorithm. There's roughly a 30-day window before decay sets in. After that, content that isn't updated starts disappearing from AI recommendations — regardless of how strong it was when published. Most law firms operate on a quarterly or annual publishing cycle. A piece goes up, gets promoted, and then sits there unchanged for months. That's not a content strategy anymore. It's a decay schedule. The firms staying on AI shortlists aren't publishing more. They're maintaining more — running monthly refresh passes on practice area pages and attorney profiles, updating data points, adding current references. Small edits. Consistent cadence. Sustained visibility. If your firm published strong content six months ago and assumes it's still working, it probably isn't. New on the Selectio.ai blog: the mechanic behind content freshness in AI search, and the five-step refresh cadence that keeps law firms visible. https://lnkd.in/edaCpqhE #LegalMarketing #AEO #AISearch #LawFirmMarketing

A Brief Interruption 🫸to Your Regularly Scheduled AI Content 🤖: When 15+ Million Requests Meet Reality I wrote about what it takes to scale a .NET ingestion pipeline when traffic jumps from “a few thousand” to 15M+ requests a day 🚀. It’s a practical look at transactional outbox 🗒️️, background workers 👨🏭, retries, and idempotency when reliability matters more than buzzwords. https://lnkd.in/dhd9YJNr

A client called me panicking. Their AI chatbot — 6 months in production — was confidently giving wrong answers to customers. Not vague answers. Wrong ones. The culprit? Their RAG pipeline was retrieving outdated docs because nobody had built a freshness filter into search. One afternoon fix: → Added a recency weight to retrieval scoring → Filtered chunks older than 90 days for time-sensitive queries → Added a “last updated” field to every document’s metadata Hallucinations dropped by 60% in two weeks. The lesson: RAG failure is almost never the LLM. It’s almost always the retrieval. Garbage in. Garbage out. Still true in 2026.

tokenmaxxing follows Goodhart's law: "when a measure becomes a target, it ceases to be a good measure" tokenmaxxing is like measuring an engineer's number of lines coded...easy to game, and easy to waste resources to hit a gamified metric. today number of lines coded is even less valuable thanks to AI and tokenmaxxing, yet people measure their AI proficiency based on how much their claude code spend was last month. in a matter of months, companies MSFT cancelled claude code licenses and UBER blew past their AI budget. what were they expecting?

After dealing with an agentic (claude) AI tool for a few weeks I am of the opinion that a fundamental shortcoming in the way code generation takes place is that there is no PoC behind using a library/structure. For example say you want to use RapidJSON/simdjson for json parsing. You'd want to first see the features the library brings that's good for use. Maybe have a small mock structure to see how it'd fit with the least amount of "gluing" code. This in itself doesn't guarantee failure avoidance but does build some level of confidence. Common strategies people suggested like using /caveman and SKILLS markdown file was bogus in my opinion.

The summer is heating up as much as possible in 2026, But something is getting "COLD" Yes, you heard it right, and it's "Cold email" You may write the best email possible for a proposal, but it ends up landing in the spam section, and that's because you really wrote a good one and got flagged as AI. Don't worry, you are not alone, almost every folk is facing the same issue. Feel free to share your rants in the comments section. Looking forward to hearing from you and helping with the ways I have figured out.

Googlers didn't hold back in this new piece "Optimizing your website for generative AI features on Google Search" but still in the myth busting section they didn't exact mention - self promotional listicles. There are some other bits in this piece which sort of directionally point to listicles but would have been cool if they exact mentioned self promotional listicles & busted the myth -> classified it as spamming.