Subscribe | Free AI | CRO Audit | Cool Tools

Which Test Won?

On a product page, which variation had more clicks to add to cart?
See the answer below.

News This Week

• Claude Opus 4.7 Is Imminent, and Claude Desktop Is Now an Execution Engine

  Anthropic is preparing to launch Claude Opus 4.7 alongside a new AI design tool. Simultaneously, Claude Desktop received a wave of major updates, including Routines for scheduled and event-triggered automation, which transformed it from a chatbot into a local task operator that can touch files, control a browser, and run workflows without human babysitting. (Link) (Link)

• OpenAI Launches GPT-5.4-Cyber, Its Answer to Anthropic's Mythos

  OpenAI unveiled GPT-5.4-Cyber, a specialized cybersecurity model released exclusively to thousands of verified defenders through its new Trusted Access for Cyber (TAC) program. Like Anthropic's Mythos Preview, access is tightly controlled, marking a new industry norm: the most powerful AI models are being handed to security experts before the general public. (Link)

• Google Gemini Launches on Mac With Screen-Aware AI and Agentic Shopping

  Google's Gemini app is now available on Mac, letting users summon AI assistance from any application with a keyboard shortcut and share their screen in real time. Separately, Gemini is testing a native shopping cart, hinting at a future where users browse and buy products without ever leaving the AI assistant. (Link)

• Stanford Research: AI Adoption Is Up, But Worker Trust Is Down

  New Stanford research shows that while AI adoption in workplaces has risen sharply, trust in AI tools among workers has declined in parallel. Employees increasingly report that AI creates more review work than it eliminates. One copywriter told The Guardian that her workload became dominated by correcting AI-generated "workslop" that looks polished but has major flaws. (Link)

• Saying "AI-Designed" on Your Product Can Cut Purchase Intent by Up to 29%

  Researchers ran 9 experiments and found that labeling a product "AI-designed" versus human-designed reduced purchase intention by as much as 29%, with shoppers perceiving 75% less human involvement. The fix: frame AI as a collaborator, not the author. "Our designer Alex used AI" outperforms both fully human and fully AI branding, boosting purchase intent 13% above AI-only messaging. (Link)

• Meta Is on Track to Surpass Google in Global Ad Revenue for the First Time

  Meta is projected to capture 26.8% of global digital ad spend in 2026 vs. Google's 26.4%, the first time any company has threatened Google's position at the top. The driver is Meta's AI-powered ad automation, which is faster to deploy and easier to justify to CFOs than Google's more complex offering. Marketers with Google-heavy budgets are urged to stress-test their media mix now. (Link)

• Google AI Overviews Have a 10% Error Rate, and Most Users Don't Double-Check

  A New York Times analysis found that Google's AI Overviews are inaccurate roughly 10% of the time, yet most users treat the AI-generated summaries as authoritative without verifying them. For SEO, this creates both a hazard (misinformation from your category appearing above your content) and an opportunity (factually reliable, well-cited content is increasingly likely to be selected as the source). (Link)

• MCP for Websites: How to Make Your Site Readable by AI Agents and LLMs

  Anthropic's Model Context Protocol (MCP) is emerging as a key infrastructure layer for AI search visibility and acts as a standardized bridge that lets large language models cleanly access and interpret your website's content. Digital Brand Expressions covers why forward-thinking technical SEOs are beginning to implement MCP servers to ensure their sites are accurately represented in AI-generated answers and agent workflows. (Link)

• When AI Customer Service Goes Wrong and the Fixes That Actually Work

  A Forbes analysis documents the most common failure modes of AI-powered customer service, including confident wrong answers, inability to handle emotional escalations, and over-automation of high-stakes interactions, alongside the interventions that restore conversion and satisfaction. The core finding: AI in CX needs clear handoff protocols to humans at defined trigger points, not just a "contact us" fallback. (Link)

• Google Data Studio Separates from Looker Studio and Returns as a Free Reporting Hub

  Google is bringing back Data Studio as a standalone, free reporting tool for Ads, Sheets, and BigQuery, completely separating it from the enterprise-focused Looker Studio. For conversion teams managing multiple campaigns, this removes a major dashboard stitching headache and gives marketers a cleaner, no-cost path to attribution reporting across Google's ecosystem. (Link)

• Anthropic co-founder confirms the company briefed the Trump administration on Mythos (Link)

Cool Tools

AI tools for sales

• Pippit - Video generator for marketers (Link)

• PixelSeo - Get meta and schema data for images (Link)

• Umanwrite - transform AI slop into human text (Link)

• AdsCreator - enter a URL and get ads for Meta, Google (Link)

Which Test Won Answer

Variation A, with a percentage saved, was the winner. Clicks on Add to Cart had a lift of 12.5% @ 92% statistical confidence.

This is called price anchoring. Every e-commerce store owner must understand this concept because it sets the stage for the value of a product in the shopper’s mind. “This value is $49, but today you can get it for X!” This test has to do with how to show the amount off. The general rule of thumb is that if a product is over $100 US dollars, then use $ off. If it’s under, then use % off.

If you look at this in extremes, it makes sense. For example, a $10 product that is 25% off sounds like a bigger savings than $2.50. A $500 product that is 25% off sounds like a smaller savings than $125.

With all tests, you must test this on your site before implementing it. A test result from one site does not indicate it will be the same on another site due to many factors.