Table of Contents
Welcome to this week’s SEO News Roundup from Pathfinder Marketing. A big week for analytics and AI tracking changes, a fresh Ahrefs schema study that’s going to ruffle a few feathers, plus Shopify rolling out official connectors for ChatGPT and Claude. Let’s get into it.
Google Analytics Adds AI Assistant as a Default Channel Group
Google Analytics is rolling out a native “AI Assistant” channel group in GA4. Up until now you had to build a custom channel group with regex to separate AI assistant traffic from referrals. Google is starting to do this natively, and some of you might already be seeing it in your GA4 accounts.
Two things worth knowing. First, it only works moving forward, so historical data won’t be reclassified. Second, it only works if the referrer string is intact. So if the click comes from the ChatGPT app, where the referrer gets stripped, that traffic will still land in Direct.
When GA4 detects an AI assistant referrer, three dimensions update automatically: the medium becomes ai-assistant, the channel becomes “AI Assistant”, and the campaign is labelled (ai-assistant).
Google Stops Sending the Text Fragment URL for AI Overviews
Another change coming to AI tracking in Analytics: Google has dropped sending the text fragment URL on AI Overview and AI Mode clicks. Previously when AI Overviews or AI Mode linked out to a page, the URL would include a #:~:text= fragment that jumped users straight to the relevant passage. It looks like Google has stripped that out.
If you were using that fragment in GTM to identify AI Overviews traffic in GA4, that workaround no longer works. Search Console still has an “AI Overviews” filter under Search Appearance, but at the URL level there’s no longer a reliable way to separate AIO clicks from regular organic.
Adding Schema Did Not Boost AI Citations in Ahrefs Study
Ahrefs ran a study on 1,885 pages that were already ranking in ChatGPT and AI Overviews to see if adding schema markup helped them get cited more. The answer was no. Adding schema markup to already-ranking pages did not meaningfully improve citations on any platform.
The numbers across their matched difference-in-differences analysis: AI Overviews dropped 4.6% relative to control pages (statistically significant but small), AI Mode came in at +2.4%, and ChatGPT at +2.2%. Both of those last two are statistically indistinguishable from zero.
| AI source | Effect on citations |
|---|---|
| Google AIO | −4.6% |
| Google AI Mode | +2.4% |
| ChatGPT | +2.2% |
One important caveat: every page in the study was already getting 100+ AIO citations before adding schema. So the data speaks to whether schema lifts already-cited pages, not whether it helps pages break in for the first time.
AI Citation Ranking Factors From Cyrus Shepard
Cyrus Shepard recently put together a report pulling out 54 experiments, case studies, and patents on what actually moves the needle for AI citations. He’s classified the findings into factors and given each a score. Based on all the studies, these are the things he reckons influence AI citations the most.
The top of the list is mostly traditional SEO: URL Accessibility (9.5), Search Rank (9.4), Fan-out Rank (9.3), Preview Control (9.2), Query-Answer Match (9.2). Further down you get content-specific factors like Answer Near the Top (8.8), AI-ready Structure (8.6), and Factually Specific (8.3). Domain Authority lands at 5.0 and LLMs.txt at the bottom on 2.0.
Some of the scores need a grain of salt. Structured data is at 5.6, but the Ahrefs study we just covered showed it’s not really considered at all. That said, it’s worth knowing what the different studies are showing and the report is worth a look.
Chrome Silently Installs a 4 GB Local LLM on Your System
It was recently discovered that Chrome installs a 4 GB AI model on every system that meets the requirements. Basically there’s a local LLM sitting on your machine, and if you delete it Chrome just downloads it again. Privacy researcher Alexander Hanff goes into detail on how it was discovered and what it does.
The model is Gemini Nano, and it lives in OptGuideOnDeviceModel/weights.bin inside your Chrome profile. To turn it off properly, head to chrome://flags, find “Enables Optimization Guide On Device”, and set it to Disabled.
Someone also figured out how to actually use it. Once you enable the right flag, you can go to chrome://on-device-internals, load the default model, and query it directly. According to Jason Packer, the training data cutoff for this local model is around October 2023.
Google Drops FAQ Rich Results
Google has updated its FAQ schema documentation to say that as of 7 May 2026, FAQ rich results are no longer appearing in Google Search. Around 2023 Google had already heavily cut these out, reserving them for government and health sites, but now they’re gone completely.
In June 2026 the FAQ search appearance, the rich result report, and FAQ support in the Rich Results Test will all be removed. Support for the FAQ rich result in the Search Console API gets pulled in August 2026.
If you’ve got FAQ schema on your pages, don’t just delete it. It’s still useful for other systems that read it, including the AI engines. But the SERP feature you were hoping to earn from it isn’t coming back.
Google Tests Web Bot Auth to Verify AI Agent Requests
Google is testing Web Bot Auth, an experimental cryptographic protocol that lets bots sign their requests rather than relying on user-agent strings and IP addresses. So instead of just trusting whatever the user agent says, Google’s AI agents will be able to cryptographically prove they are who they say they are.
It’s still early days and you don’t need to do anything just yet. But once this is fully in effect, the old trick of switching your user agent to access content that’s served differently to bots is going to stop working.
Verified Google AI agent requests will carry a Signature-Agent header set to g="https://agent.bot.goog". Google’s still recommending you keep IP-based and reverse-DNS verification as a fallback, since not every request is signed yet.
n8n’s MCP Server Can Now Build Workflows From a Prompt
n8n has updated its MCP server so it can now create new workflows, not just run existing ones. If you’re heavy on n8n usage, you can build a brand new workflow just by talking to the MCP from Claude, ChatGPT, Cursor, or any other MCP-capable client.
It generates a TypeScript representation of the workflow rather than raw JSON, so the AI client can validate and fix errors before anything touches your instance. The team’s internal testing showed Claude Code outperforms Claude Chat for the same prompts. It’s available across all editions, including the free Community Edition. You’ll want n8n version 2.18.4 or higher.
Shopify’s Horizon Theme Adds ?variant= to Every Product URL
Looks like Shopify’s new Horizon theme appends a variant URL fragment to every product URL by default. Even if a product doesn’t have variants, or only has one variant, you’ll see this string stuck on the end of the URL.
There is a canonical pointing back to the clean URL. But as most of you know, canonicals don’t always get respected by Google, and this can lead to indexing and crawl budget issues, especially on larger stores.
One workaround floating around the Shopify forums is editing the liquid template to hardcode the clean product URL:
<a href="{{ product.url }}">That’s similar to how you might already be stripping the collection path out of product URLs. The catch is that merchants have reported it can break related products widgets, so there’s no quick and easy fix just yet. Worth being aware of if you’re migrating clients to Horizon.
Shopify Releases Official Connector Apps for ChatGPT and Claude
Shopify has released native connector apps for both ChatGPT and Claude that let you manage a lot of things on your Shopify store directly from the AI client. These are official, first-party Shopify connectors built and maintained by Shopify.
The Claude connector exposes 25 tools, including search_products, get-shop-info, create-product, update-product, list-orders, and list-customers. So you can look up an order, update a product price, ask how a collection is selling, or apply a discount across a set of products without leaving the chat.
It’s a meaningful productivity win for solo founders and small merchants, and worth flagging for any agency clients on Shopify.
Google Releases a Guide on Building Agent-Friendly Websites
Google has released documentation on how to build agent-friendly websites. It walks through how AI systems read your site: by taking screenshots, by parsing the raw HTML, and by reading the accessibility tree.
The recommendations cover a lot of the things you should already be doing for accessibility, but with agents now as part of the audience. Use semantic HTML like and rather than throwing everything into a
. Set cursor: pointer on anything actionable. Use the for attribute on elements to link them to inputs. Keep your layout stable so screenshot-based agents don’t get confused by buttons moving between categories. And make sure interactive elements have at least 8 square pixels of visible area, otherwise visual analysis will filter them out.
If you want to make your website agent-friendly and WebMCP-ready, this guide is worth reading end to end.
That’s all we have for this week. If you’ve got tips, suggestions, or news we should cover next week, drop them in the comments and we’ll try to include them. Cheers 👋
![How much does SEO cost in Australia [year]? 2](https://www.pathfindermarketing.com.au/wp-content/uploads/2024/07/SEO-Pricing.webp "How much does SEO cost in Australia [year]? 2")
![Best Perth SEO Agencies in [year] [Ranked + Compared] 3](https://www.pathfindermarketing.com.au/wp-content/uploads/2026/07/blog1.webp "Best Perth SEO Agencies in [year] [Ranked + Compared] 3")
