I am very interested in @randfish's recent research into the growth of LLMs, their impact on people's search habits, and their impact on Google. My summary: 1. Google is still growing, not shrinking as some would have you believe 2. Not everything people do on LLM-based chat platforms are "searches" 3. Even if you are fixated on AI-powered discovery, Google is the biggest for *that* too This is not to say we should ignore the growth of the others, or get complacent, and in fact I think there's a lot of scope to test the impact of our activities on other LLM recommendations, but we should keep things in perspective. It reminds me a lot of the growth of social media circa 2007. It was clearly a coming revolution, and there was noise about it being catastrophic for search because people were making the argument that users would prefer to ask their friends where to eat rather than searching for restaurants. It turned out, of course, that they did both.

PSA claude can drive google docs in suggest mode to finally have an AI make suggestions and leave comments all over your work. You know, if that kind of thing is your bag.

I’m struck by the similarity to something I’ve said a lot about jobs as AI gets more capable: “it’s our job to want things”

Pulling out one of the threads of thought from my post this week, I’ve long thought we are heading quickly to a world where every search is “search volume 1”. 👀 What do you think? Do you agree? 🤔 What are you doing about it?

Did you know the Google quality graders consider (among other things) "content effort" to differentiate things that somehow deserve to rank? Ever since speaking to @CyrusShepard about it, I've been thinking about what that might mean these days. I actually DON'T think it necessarily or only means "written by a human" - it may include consideration of the effort and organisation needed to gather the data, insights or information underpinning the content for example. Whatever it flows towards meaning, you'll be well-served to have a listen to Cyrus on the subject. Remember, if it were easy, everyone would be doing it, and then it by definition couldn't work in a competitive space! I'm always reminded, in these conversations of the old name for what we do: competitive webmastering. Get your competition on. (and if you want to hear more from Cyrus and me, register here for the full content: searchpilot.com/winning-ai-c… )

Bought new scales. Part of the setup involves entering your weight. Someone hasn’t thought this through

I don’t even think my P(AI did it | something kills me in the next decade) is 10%

I have become very interested in the developing AI capabilities for forecasting. My favourite tool at the moment is @FUTURESEARCHAI which takes a simple forecasting question, builds a panel of researchers and emulates the best human forecasters' approaches to quantitative forecasting. I recently set it the task of forecasting: - Change in Google search *traffic* to large retailers over the next 3 years (-15% central estimate) - Change in total online revenue for the same retailers over the same time period (+31% central estimate) How can these both be true when organic is such a large channel? Well one key is that the clicks become higher-intent on average: "while session volume may compress overall, nominal price increases (AOV) and higher-intent surviving clicks will likely keep SEO-attributed dollar revenue roughly flat" The other is obviously the growth of GEO: - Change in revenue attributed to combined organic search + AI discovery over the same time period (+20% central estimate) What do you think? See more here: futuresearch.ai/app/conversa…

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: searchpilot.com/resources/bl…

Shower thought: if / when AI gets good at writing, its guardrails will in theory mean it won't talk about certain things still, so there's a long-run niche for human novelists writing about crime and dystopia.

who do I know IRL who has a P(doom) >= 10%?

Suppose I wanted to opt in to share my data with Profound or Datos, how would I do that? I think Profound may be using Datos data? Is there an extension to install, or what?