SearchPilot ads

Podcast listeners - we just released a new episode on the feed - this time with Christine Ullmann of John Lewis talking about how premium brands get found in AI search. Don't miss out - The Business Class Lounge wherever you get your podcasts.

What's the hardest AI search question you're trying to answer right now? 🤔 Bring it to dinner. Craig Bradford and I are getting ten ecommerce and growth leaders from large UK retailers round one table to compare notes on what we're all seeing. Good food (on us), plenty of questions, and hopefully a few answers too. Wednesday 11 November, two nights before MozCon London. Retail only this time. We've got two seats left. If you fancy joining us, drop me a DM. Or tag the person at your company who should be there.

My conversation with Britney Muller, in which she told some WILD anecdotes and combined deep technical expertise with extreme clarity of thought - now available in the podcast feed: https://lnkd.in/euQUcW-M

Anyone else?

I've been in the lab. I wanted to see if I could model the impact of combining randomness in LLM outputs with the branching chaos of conversational paths. I wondered how stable the recommended product and brand would be. The answer? Extremely unstable. Given a static persona and buying objective, one of my experiments "bought" 9 different products in 12 runs. The one pictured below "bought" 5 different products from 4 different brands from 4 different retailers.One-shot prompt tracking isn't anywhere near capturing this complexity. I have more questions than answers, but I'm excited to do more testing.

Guardrail metrics have long been a staple of many forms of testing, but they have been less common in SEO testing. I wrote up three uses for guardrails: https://lnkd.in/ecQag83X