Analytics Roundup – Updates from October 2025

Overall Grade for Comet AI Browser: C

When I logged in to ChatGPT recently, I was prompted to try out their new Atlas browser. I’ve been hearing that AI browsers are the next-big-thing-since-the-last-big-thing, so I was eager to try it out. I followed their prompt and started the install, but unfortunately it silently failed. The one time ChatGPT has nothing to say…

My next move was to install Perplexity’s Comet, which came out of beta on October 2nd.  Comet has an AI sidebar that can read websites, summarize and analyze content, and assist with a variety of browser-related tasks. To test it out, I gave it a series of assignments and graded the results. Here’s my report card:

• Analyze a website and suggest Google Analytics custom events and parameters: B
• Review my past roundups and summarize this article in keeping with my style: D – looks like you’re stuck with human me for the time being.
• Review my Feedly feed and suggest other sources to follow: B+
• Identify insights in a Google Analytics Traffic acquisition report: A-
• Convert the table of cities by population from this page into a CSV file for downloading: F

Of course I could do any of these things in Claude or ChatGPT, but it does feel more streamlined to have it right there. It also performed well as a browser.

Given the mixed results, I’m not sure how much my behavior will change in the short term, but it seems inevitable that we will all be doing AI-assisted browsing soon enough.

Product Updates

• GA: Benchmarking support for unnormalized metrics
  Until recently, GA4’s benchmarking feature only supported proportional metrics like Engagement rate and Average engagement time per session. Now you can see benchmarking for metrics like Sessions and Active users. I’d be more excited about these additions if I didn’t hate the benchmarking UX so much.

• GA: In addition to the already-supported Reddit, Snap and Pinterest, you can now set up automated imports of cost, impressions and click data from Meta and TikTok.
• Looker Studio: New Google Ads metrics
  The native Google Ads connector now supports Conversions (by conv. date), Conv. value (by conv. date), All conv. (by conv. date), All conv. value (by conv. date). The fact that Google Ads otherwise attributes conversions by the ad impression or click date can result in significant discrepancies with Google Analytics, especially if you sell high-ticket items with a long consideration cycle. Reporting conversions and revenue by conversion date can reduce discrepancies significantly.

Workflow

• Elevating SEO Analysis with Google Search Console and BigQuery, Sarah Crooke
  This article starts with the most basic queries possible, but ends with an approach for doing query intent classification with AI. The code she shares could be easily adapted to other classification/data-enhancement tasks. I also love her visualization for ‘Landing page CTR vs Engagement rates’.
• Deploy GTM Monitoring System on GCP with Terraform and Dataform in 30 Minutes, Krisztián Korpa
  The ability to monitor tag firing has been on my wish list for GTM forever. This article describes how to do it step-by-step, using the GTM addEventCallback API. Brilliant.
• When a New Browser Launches: What I’m Seeing in Google Analytics for ChatGPT’s Atlas, Dana DiTomaso
  Practical implications for tracking. The article goes into detail about Atlas, but most of it also applies to Comet and Dia (the other AI browser that’s getting a little buzz). I hadn’t really thought about the fact that the introduction of a new browser would inflate user counts. Which highlights the fact that “users” as a metric isn’t really users at all.
• Data Catalogs 2.0 – Get Value from Metadata, Peter Baumann
  I’m not sure how I feel about Data Catalogs in practice. Like a lot of documentation initiatives, the vision is compelling, but the reality usually falls short. But as data complexity and regulatory requirements increase, doing nothing is not an option. This article does a great job of outlining what to work towards and why it matters.

Ideas

• AI, Data Work, and the In-Between, June Dershewitz
  A thoughtful and informed perspective on the state of data work. My favorite quote: “It’s a bit like a game of musical chairs, and the music hasn’t stopped yet.”
• What’s Your Edge? Rethinking Expertise in the Age of AI, Ravikiran Kalluri, MIT Sloan Management Review
  In some ways this is a perfect companion to June Dershewitz’s article. The author outlines specific organizational strategies for creating meaningful work and staying competitive as we incorporate AI into workflows. A quote that struck a chord for me in this one was, “Higher education teaches the art of knowledge creation: how fields establish truth, evolve understanding, and challenge paradigms.” I wholeheartedly agree, as long as we let great teachers teach.
  Reading the full article requires a subscription, or you can buy a PDF for $7.95.
• The Rise of Data Platform Monoliths, Madison Mae
  I mostly cover Google analytics & data products, but Fivetran buying dbt is a really big deal in the data world. This article and others I’ve read cast it as a necessary-for-survival move in the face of Snowflake’s introduction of Openflow. I personally don’t see how Fivetran’s revenue model can survive and it would be a huge shame if dbt ends up going down with the ship. Historically, Fivetran has been able to charge a premium for data extract-and-load, but extract-and-load is becoming commoditized. A few other perspectives: Tom Tunguz, Timo Deschau
  

Privacy

• OOPS! The Consumer Signal You Might Be Overlooking in Your Privacy Strategy, Ariana Wolf, Cardinal Path
  Opt-out Preference SIgnals (OOPS), also known as Universal Opt-out Mechanisms (UOOM), “are now legally recognized under a dozen U.S. state laws, yet many organizations remain unaware of, or unprepared for, this emerging compliance requirement.”
  I checked the CMPs I work with on client sites, and they support GPC, which appears to be the most established protocol for Universal Opt-out. But it’s a confusing landscape and I still encounter plenty of organizations that have built their own consent solutions (bad idea) or use a janky tool that is probably not up to the latest standards.