The Future of Search and the Death of Links // BRXND Dispatch vol 96
Tom Critchlow shares data on declining clicks, why AI overviews are changing search behavior, and why he's still optimistic about the open web's survival.
You’re getting this email as a subscriber to the BRXND Dispatch, a newsletter at the intersection of marketing and AI.
Claire here. This week, we’re spotlighting Tom Critchlow’s talk from BRXND NYC on the future of search. Tom has been working in search for over 20 years and currently leads the audience growth team at Raptive, which works with 6,000 publisher sites and sees about 20 billion Google page views annually.
Tom argues we’re moving from an era of links to an era of recommendations. AI overviews now appear on 30-40% of search results, and when they do, clicks drop 20-40%. Google’s AI Mode sends six times fewer clicks than traditional search. But Tom is actually optimistic about the web’s future. He thinks recommendation feeds like Google Discover will grow to replace declining search traffic, and that AI will get so good at understanding us that it’ll proactively surface relevant content rather than waiting for us to search.
The shift Tom describes feels significant beyond just SEO mechanics. When search worked through links, you actively chose what to click—it was manual research, even if imperfect. Recommendations flip that relationship. AI decides what you should see based on what it thinks it knows about you. That creates interesting pressure on brands: they can’t just game algorithms with SEO tricks anymore. They need genuine value propositions because AI won’t recommend bad products. But it also raises questions about what happens to our relationship with information when we move from active searching to passive receiving. Does “doing research” mean something different when content is curated for you rather than found by you?
Tom’s talk is engaging, informative, and certainly worth a watch. Watch it here:
What caught our eye this week
Noah got a shoutout from Boris Cherny, one of the creators of Claude Code, on Every’s AI & I podcast with Dan Shipper.
Ticketmaster is running AI-generated ads on Facebook that swap out virtual families and team allegiances based on who’s seeing them—a bit crude now, but a preview of where this is heading. On Meta’s Q3 earnings call, Zuckerberg laid out the company’s vision: advertisers will eventually just give Meta “a business objective and a credit card” and AI handles everything else, from generating personalized video creative to finding the right customers. The real revenue opportunity for Meta isn’t consumer-facing products like ChatGPT, but rather B2B tools that let brands create infinite ad variations and serve them only to people they’re most likely to work on.
OpenAI, Anthropic, and Cohere are hiring more forward-deployed engineers who can code and talk to customers. OpenAI’s FDE team launched this year and expects to hit 50 engineers in 2025, while Anthropic is growing its applied AI team 5x. Palantir pioneered this almost two decades ago, sending pairs called “Echo” (to figure out what customer needs) and “Delta” (to build it). Nic Prettejohn at Palantir says “the only valuable software is not how exquisite its code is... It’s only valuable if it means something for the end customer.”
Anthropic published how code execution makes MCP efficient: instead of loading all tool definitions upfront (150k+ tokens), agents write code to discover tools on-demand from a filesystem, dropping to ~2k tokens (98.7% savings). Agents write code that talks to MCP servers instead of making direct tool calls, keeping intermediate results in the execution environment.
If you have any other questions, please be in touch. As always, thanks for reading.
Noah and Claire