One category label change. New Balance swapped "athleisure" for "athletic footwear" in its web copy and product listings. Its visibility in AI-generated search answers jumped from 1% to 90%. Search Engine Land has the full breakdown. That's not a typo. One semantic tweak. Massive swing. For any small business owner who has ever wondered whether this AI search stuff actually matters yet, that number is your answer.

Here's why it works. AI search tools like Google's AI Overviews don't browse your site the way a human does. They read structured signals: how you label your category, how your copy describes what you do, what your schema markup says you are. If those signals use the words an AI is looking for when someone asks "what kind of shoe should I buy for running," you get recommended. If they don't, a competitor does. No algorithm change required on their end. You just weren't speaking the right language.

How to fix it: Go look at how you describe your business category on your website, your Google Business Profile, and your directory listings. Are you using the plain, common-sense label a customer would type into a search bar, or are you using insider jargon or a branded term you invented? Change it to match the common label. That's the whole strategy.

Want to check your website for AI visibility? Use our free scanning tool. Making simple changes can help, just ask Michael from Murals & More…

AI Overviews now show up in 43% of searches. A year ago it was 15%. Similarweb's 2026 report tracked that jump and also found that Google's AI Mode grew from 126 million to 279 million monthly visits in the same period. Shelly Palmer has a name for the new metric that matters: "share of prompt." That's how often an AI system picks your business when it's constructing an answer to someone's question. Not your page-one ranking. Not your click-through rate. Whether the AI thinks of you at all.

This is early days and the playbook isn't fully written yet. But the New Balance example tells you something useful right now: the businesses that get surfaced in AI answers are the ones whose copy is clear, literal, and categorically specific. Fancy brand language works great for humans. AI systems want the boring, obvious version.

ChatGPT quietly changed how it handles style prompts. If you have been writing "write this email in the style of [author name]," that prompt now gets pushback. OpenAI has adjusted ChatGPT's behavior so it refuses direct style-copying requests for named living authors, offering instead to match a "similar feeling." The change has real implications for small business content workflows. Prompts that worked last month may now stall. The workaround is to describe the style in concrete terms rather than attributing it to a person: "punchy, short sentences, plain language, no corporate tone" gets you further than a famous name. It also signals where the legal guardrails are heading across all the major AI tools. If your content process relies on named-author style prompts, now is a good time to rework them.

·         Hardbook -- Booking link that collects dates and gets a contract signed in one step. For freelancers and service businesses losing deals between verbal yes and signed paper.

·         Liminal -- Shared workspace where your AI agent writes files and your team sees them as clean UI in real time. No Slack pasting, no hosting setup.

·         Cekura -- Automated QA for voice and chat AI agents. Simulates calls before launch and monitors live production. For anyone who has deployed an AI agent and wants to know it actually works.

·         SUB/WAVE -- Turns your local music library into a self-hosted AI DJ radio station. Built for cafes, retail, gyms. Free under MIT license, so no streaming licensing headaches.

·         Claude's voice mode now runs on Sonnet and Opus and can call Gmail, Google Calendar, and Slack mid-conversation. Hands-free AI delegation just got meaningfully more useful.

·         Forbes rounds up five practical owner-relevant developments this week: OpenAI's small business training program, Microsoft Copilot deeper in Excel, Intuit's QuickBooks credit card, Claude's "Record a Skill" screen-watcher, and a call to redesign business sites so AI agents can read them.

·         Anthropic published its position on open-weight AI and declined to sign the industry letter that OpenAI, Google, and 150+ others signed. Useful context if you're evaluating open-weight tools.

·         Kimi K3 weights are now public on Hugging Face -- the largest open-weight model released for commercial use, free to download (NOTE: It’s HUGE!)

·         Zapier's AI Workflow Index got its first public walkthrough -- four roles AI actually plays in real workflows, and why productivity gains stall.

·         Sam Altman visited the White House to preview OpenAI's most powerful model as the administration finalizes its AI regulatory position.

·         Anthropic published a prompting guide specifically for Opus 5 -- the new model behaves differently enough that existing prompts may need updating.

·         Inc. has a practical piece on staying relevant as an entrepreneur in a period of real disruption.

·         NVIDIA is leading a new Open Secure AI Alliance framing open-source AI as a cybersecurity asset.

-Scott

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.