89% of AI search demand has no owner yet, and neither does most of your clients' patch
Big week for anyone trying to work out where AI actually earns its keep: turns out most AI search categories are still up for grabs, until you clock how much technical groundwork you need just to be in the running, Notion's growth still comes down to unglamorous community grunt work rather than luck, and even the slickest AI video ad we've seen this year only worked because someone deliberately made it look less like AI. Who would've thought humans were still needed? Here are your five.
1. 89% of AI search demand has no owner yet, and neither does most of your clients' patch
Kevin Indig's pulled six months of ChatGPT answer data from Semrush, 1,094 categories, five prompts each, over 600,000 citations, and found that only 15.2% of categories have a genuine owner (one brand leading four of five prompts by five points or more). Just over half are wide open with no clear front-runner. Worse for anyone chasing the big prizes: the highest-volume categories, worth 98% of all AI search demand, are the least settled of the lot. Add it up and 89% of that demand sits in categories nobody's locked down. Worth flagging to any client sat on a "we're roughly visible in ChatGPT" report: a one or two point lead isn't a lead worth defending. Indig found leaders only get overtaken when the gap's under about three points, once you're past that, you tend to hold your spot. Citations matter less than you'd think too, the most-cited brand in a category was only the most-mentioned one about a fifth of the time. Worth building into next quarter's reporting: pick the 10-20 categories that actually matter to a client, track the same five prompt types (definition, comparison, alternatives, use case, buying decision), and stop treating citation count as the whole strategy. It gets you in the room, it doesn't make you the name ChatGPT says out loud. Search Engine Journal
2. The technical SEO checklist AI crawlers actually bother reading
Telepathic's put together a proper technical playbook for AI visibility: which bots to allow in robots.txt (GPTBot, ClaudeBot, PerplexityBot, Google-Extended), schema markup beyond the usual Article and FAQ types (HowTo, Speakable, entity IDs that link your brand, people and products together), and site structure a bot can actually follow in three clicks or fewer. Worth taking seriously for any client whose site leans on heavy JavaScript: most AI crawlers execute little to no JS and time out in five to ten seconds, so if the words aren't in the raw HTML, they don't exist as far as AI search is concerned. Worth a five-minute check on any client site this week: right-click, view source, and see if the actual page copy is sitting there in plain text. If it isn't, that's the fix that needs to happen before schema or backlinks get a look in. Telepathic
3. Notion's growth wasn't luck, it was an ambassador programme, a lot of TikTok, and knowing what not to build
First Round's deep-dive into Notion's marketing is worth the full read: an Ambassador programme that started as a single tweet and pulled a thousand applicants within weeks, a template gallery built almost entirely from user submissions instead of an in-house content team, and an influencer push under Lexie Barnhorn (ex-Curology) that took #Notion past a billion views on TikTok. Worth reading in full for the mistakes section alone: expecting community to prove its ROI in month one, burning weeks testing community tools instead of just picking one, and hiring a community lead for the job title rather than genuine passion for the product. Worth forwarding to any client who treats community as a nice-to-have rather than a moat: Notion's founding community hire is blunt about it, "just go viral" was never the strategy, the grants, swag logistics and a consultant certification programme were. First Round Review
4. The AI ad that converted didn't look like an AI ad
Shlomo Genchin's walked through the AI ad he built for HiBob that performed so well it's being turned into a CTV ad: a real Reddit testimonial about payroll frustrations sourced via Perplexity, a consistent AI character built in Gemini, animated through Enhancor and Kling, voiced in ElevenLabs, then deliberately desaturated in the edit so it read as authentic rather than AI-glossy. Worth the reminder for any client nervous their AI video will look cheap: the bit that actually worked wasn't better AI, it was starting from a real pain point instead of a made-up concept, then editing the shine back out afterwards. Worth stealing even if a client never touches an AI tool: search Reddit or reviews for how their actual customers describe the problem before anyone writes a script. The best line in the ad wasn't written, it was found. Unbore
5. Where AI actually belongs in a newsletter (hint: not the writing)
Dan Oshinsky's rounded up how he uses AI across his newsletter business, and writing isn't top of the list. It's process (coding one-off tools and templates in an afternoon), ideation (a brainstorming partner that pokes holes in an idea), and analysis (sorting survey responses, spotting subject line patterns), the voice stays entirely his. Worth sitting with if you or a client run a newsletter: the productivity gains are real, but they're all in the scaffolding, not the sentence readers actually notice. He also makes the case for a plain-English AI policy linked in the footer, so readers know exactly what's touched by AI and what isn't. Worth trying this week regardless of the newsletter: take a recent survey's open-text answers, or a pile of subject lines and their open rates, and ask AI what it notices. Free research that used to take an afternoon of scrolling. Inbox Collective
Small business shout-out
This week's shout-out goes to Dispute Buddy, an app from New Zealand-based founder Jenny Rudd that scans years of text messages for patterns of coercive control, gaslighting, and financial or emotional abuse, then turns them into a lawyer-ready report. Rudd built it after spending 20 hours screenshotting and organising her own messages for Family Court, on top of thousands in lawyer fees, so she built the tool she wished had existed. It's a one-off, lifetime-use purchase, iPhone only for now. Know a small business we should feature? Reply and let us know.
