Daily Edition · Page 7

THE DAILY INDEX

Technology, AI, software, security, crypto and business

NEWS & TRENDS

TLDR Marketing

CLIPPING IS A SHOO-IN MEDIA TREND OF THE YEAR

Clipping is where long-form video is broken into short 30-60 second segments for distribution on social platforms. It's driving massive reach, with some podcasts and creators gaining far more influence through clips than through their original full-length content. A growing ecosystem of paid “clippers” now distributes and amplifies these snippets, sometimes earning significant income based on performance. While it started as a way to promote longer content, clipping is increasingly becoming the main product itself.

HOW CEOS VIEW CMOS

A survey of 150 US CEOs found that the CEO-CMO relationship is stronger than ever, but performance expectations are outpacing delivery. 60% of CEOs now classify marketing as a cost center, nearly double last year's 35%, and top CMO ratings declined from the previous 2 years. The core tension is that CEOs want revenue impact and financial accountability while many CMOs still report in marketing terms rather than business outcomes. CMOs also face a unique pressure point around AI, being held 4x more accountable for AI ROI than any other C-suite executive.

SEO & AI TIP: OPTIMIZING FOR FREE VS. PAID CHATGPT

Free and paid ChatGPT operate as different systems, so your brand can appear with conflicting answers depending on the user tier. In one test, the free version ignored real search results and fabricated a 90 to 91 degree forecast, while the paid version used a structured data tool to return an accurate 68 degree result. This creates a targeting problem since B2B buyers often use paid tools while consumers rely on free ones, meaning your effective AI presence depends on your audience mix. Audit both tiers by running identical prompts and comparing outputs to map the true exposure surface. Measurement needs to align with the tier your ICP uses, or reporting will reflect a version of reality your customers never see.

HOW ONE RETAILER USED AI TO TURN ‘OUT-OF-STOCK' INTO IN-STORE TRAFFIC

The Salvation Army ran a retail campaign that advertised recently sold, one-of-a-kind thrift items rather than available stock, using scarcity and FOMO to drive store visits. Gemini evaluated thousands of product photos across the nationwide inventory while image generator Nano Banana transformed basic stock images into editorial-quality creative. The hyperlocal campaign targeted specific ZIP codes near stores through Google's Demand Gen, serving custom imagery across YouTube, Discover, Gmail, and the Display Network. In its first 30 days, the campaign achieved an $11 cost per store visit, outperforming the industry benchmark by 138%, with over 58% of clicks leading to in-store searches.

DUDE PERFECT WANTS TO GRADUATE BEYOND BEING A "CREATOR COMPANY"

Dude Perfect is positioning itself as a media company rather than a creator business, with its CEO noting that partnerships now resemble sports sponsorships and long-term brand affiliations rather than one-off sponsored posts. More than half of the company's content is now consumed on TV screens through YouTube, and 80% of viewing happens with kids and parents together, making it one of the few digital properties that reliably reaches family units. The company is expanding beyond its original five performers by launching new channels and developing owned IP and formats.

LOOKING FOR THE “I” ON THE INTERNET

“Frictionmaxxing” is the trend of reintroducing intentional inconvenience into an overly smooth digital life. It pushes back on a frictionless internet where saving, scrolling, and consuming happen automatically. The concept highlights the impulse to collect online fragments, from memes to videos to aesthetic references, most of which are never revisited. This accumulation turns users into informal archivists shaped by algorithms and overload, raising questions about identity in a system built on saving without return.

HOW AMAZON USES LLMS TO RECOMMEND PRODUCTS

Amazon's COSMO system tackles a fundamental limitation in product search by teaching recommendation engines why people buy things, not just what they buy. The team fed millions of purchase behavior pairs into large language models, then built a multi-stage pipeline to filter out the roughly 65%-91% of outputs that were circular, generic, or trivially obvious. A smaller instruction-tuned model called COSMO-LM now generates fresh commonsense knowledge on the fly at a fraction of the original inference cost. A/B tests showed a 0.7% lift in product sales and 8% increase in navigation engagement on just 10% of US traffic. Amazon projects billions in potential revenue if extended across all traffic.

TLDR IndexPage 7