This Week in AI: June 13, 2026

Six things landed this week that touch how you measure, build, and decide what to build with AI. Here is what each one changes for your setup.

The updates

1. GA4 now tracks AI assistant traffic on its own What happened: Google Analytics added a native AI Assistant channel to its default channel group, first confirmed in the Analytics Help Center's "What's New" documentation on May 13, 2026, with broader availability across properties reaching most accounts by around June 7. Sessions referred from ChatGPT, Gemini, and Claude are classified automatically, with no tagging. The number is forward-only, so there is no back history, and it only catches referrals that survive the trip to your site. Treat it as a floor, not a full count. Source: https://support.google.com/analytics (Help Center "What's New") (writeup: https://www.searchenginejournal.com/google-analytics-adds-ai-assistant-as-default-channel-group/574974/) What to change in your setup: Turn the channel on in your reporting view now so you start accumulating history. The thing to optimise is not ad creative, it is the source content and structured data an assistant pulls from, so pair the number with content built to be cited.

2. Google's AI Mode ads answer the question instead of taking a slot What happened: At Google Marketing Live 2026 (May 20), Google announced two new AI Mode ad formats, Conversational Discovery ads (Gemini builds the creative live, matched to the exact phrasing of the query) and Highlighted Answers (ads eligible to appear inside AI Mode's own recommendation lists), plus two more rolling out separately across Search, AI-powered Shopping ads and a Business Agent for Leads. Both AI Mode formats were still in testing as of the announcement, with no confirmed public launch date yet. Source: https://blog.google/products/ads-commerce/google-marketing-live-search-ads/ What to change in your setup: The lever moves from creative variants to the feed the model reads. Audit what Gemini can truthfully say about you: product data, structured answers, on-site content. You no longer control the final asset, only the inputs, and it's worth doing this audit now even while the formats are still in testing.

3. Your Google Business Profile now plugs into Gemini and GA4 What happened: Two linked moves in the same week. You can connect a Google Business Profile to Gemini and build a Business Notebook over your reviews, questions, and performance data, then ask it for trends or to update profile details. Separately, you can link the Business Profile to Google Analytics to track calls, bookings, and direction clicks alongside the rest of your traffic. Source: https://blog.google/innovation-and-ai/products/gemini-app/gemini-features-for-businesses/ (GA link: https://x.com/googleanalytics/status/2065094180053852527) What to change in your setup: If you or a client runs on local intent, connect both. Local data is moving into the assistant layer, and the GA link finally makes local actions measurable cross-channel. Both are rolling out globally through June 2026 in phases (the Business Notebook piece specifically excludes the UK and EEA for now), so don't be surprised if the option isn't visible in every account yet.

4. A method for using LLMs as a concept pre-screen, not a verdict What happened: PyMC Labs and Colgate-Palmolive published a method called Semantic Similarity Rating. Instead of asking a model to rate a concept 1 to 5, you have it roleplay a demographic and write free-text reactions, then map those to a scale by semantic similarity. Tested against 57 real surveys and 9,300 US participants, it reproduced human purchase-intent distributions at about 90% of human test-retest reliability. That means about as consistent as people are with each other, not 90% accurate at predicting real purchases. The viral version of this overstated it. Source: https://arxiv.org/html/2510.08338v1 What to change in your setup: Add a synthetic pre-screen step before any real research. Use it to kill weak concepts and draft survey questions overnight, then take the two strongest to real buyers. Do not let it stand in for one real conversation with one real customer.

5. ChatGPT shipped personal finance, and turned a category into a feature What happened: OpenAI rolled out a Personal Finance preview for ChatGPT Pro users in the US on May 15, 2026, built on Plaid, connecting to 12,000+ financial institutions (read-only: it can see balances, transactions, and holdings, but cannot move money, trade, or pay a bill). A decade of fintech apps competed on exactly those features. Coverage framed it as the platform pulling standalone apps down into infrastructure. Source: https://www.bloomberg.com/news/newsletters/2026-06-05/ai-personal-financial-advisers-chatgpt-claude-threaten-jobs (also: https://www.inc.com/lucia-auerbach/chatgpts-personal-finance-test-is-rolling-out-in-the-u-s-with-a-major-warning-label/91346331) What to change in your setup: Run the moat test on anything you are building. Could a frontier lab ship this as a feature next quarter. If the answer is yes, the edge has to be a niche workflow, proprietary data, or proof, not something an assistant can copy in one release.

6. Someone turned the Claude Code spinner into an ad slot What happened: Andrew McCalip (ShiftKeys, Inc.) launched Kickbacks on June 11, 2026, an extension that swaps Claude Code's idle spinner for a short sponsored line sold in a live auction, then shares the revenue with the developer. The developer share started at 50% and was bumped to 70% within the first day. Early reported earnings are small and inconsistent: one tester made about $1.30 in 20 minutes of active coding, another 43 cents over a full workday, a third $4.43 from 407 impressions across three hours. McCalip's own suggestion that it "could eventually cover an entire AI subscription's monthly cost" or earn "north of $1 per active coding hour" is his own forward-looking speculation about where the market could go, not a measured result. The launch tweet passed 5.5 million views in 24 hours, and two competitors (SpinYield, Claude Code Ads) launched within 48 hours. Source: https://x.com/andrewmccalip/status/2065049432652189933 (project: http://kickbacks.ai) What to change in your setup: This is a signal, not a must-install. The useful idea for your own agent tooling is the boundary: keep anything that competes for attention or trust out of the context window.

The one to actually act on this week

Turn on the GA4 AI Assistant channel today. The data is forward-only, so every day you wait is history you cannot get back. It costs nothing and it gives you the first real read on how much of your traffic already comes through assistants.