Update #5 - 20 October 2026

Where did all the ChatGPT traffic go? Nowhere, it was hiding in "direct"

Let’s start with a sneaky one! If your ChatGPT traffic looked like it fell off a cliff in the second week of October, you weren’t losing visitors. You just couldn’t see where they were coming from.

When GPT-6 rolled out between 8 and 10 October, links clicked in ChatGPT stopped carrying the utm_source=chatgpt.com tag (the little label that tells your analytics a visit came from ChatGPT). ChatGPT also opens links in a way that hides the site the visitor came from. No tag and no referrer meant tools like GA4 filed those visits under “direct”, as if someone had typed your web address in themselves.

Adobe’s Cedric Huesler was first to spot it. Tracked ChatGPT visits dropped 67 per cent in a week. Across around 500 pages where ChatGPT normally drives at least 5 per cent of visits, ChatGPT visits fell by about 31,000 a day, while direct visits on the same pages rose by about 26,000 a day. Search barely moved. Two in three pages that lost ChatGPT traffic picked up direct visits instead, and those new “direct” visitors behaved just like ChatGPT visitors.

Good news: OpenAI confirmed on 14 October that it was a bug, and the fix rolled out overnight. Tagged ChatGPT visits are back to 95 to 100 per cent of normal.

Our take: pop an annotation in GA4 for 8 to 15 October so nobody mistakes that dip for a real drop when October reporting comes around. It’s also a good reminder of how wobbly AI traffic tracking still is. One product update from OpenAI and a whole month of numbers goes sideways. That’s why we never judge AI visibility on tracked clicks alone. Keep an eye on direct traffic to the pages AI likes to send people to, your branded search, and how often AI tools actually mention you. Even on a normal day, some of your AI visitors are quietly sitting in “direct”.

Googled your own brand lately? There might be an AI summary waiting for you

Search for a big brand by name and there’s a good chance you’ll now see an AI Overview about it. SEO consultant Chris Long tested around 100 brands and found AI Overviews on 93 per cent of them, including Reddit, Salesforce, Amazon and Adobe. Tracking platform DemandSphere says AI Overviews on brand-name searches tripled in September, and Ahrefs puts it at 83 per cent of branded searches in the US.

Most of these summaries sit mid-page, below the brand’s own listing, but not all. Adobe’s shows right at the top. Google’s own name and news publishers seem to be off limits, for now.

Something worth flagging for our Aussie readers: this data is US-based (or doesn’t say where it was tested), and nobody has published Australian numbers yet. AI Overviews already run here though, so it’s worth a quick check.

Sources: Search Engine Roundtable, Ahrefs, Search Engine Journal.

Our take: your brand search results used to be the one page you had full control over. Now Google writes its own summary of who you are, pulled from whatever it can find about you online. Google your brand name in an incognito window this week and read what the AI says. If it’s wrong, out of date or leaning on a cranky review thread, fix the places it’s drawing from: your About page, your Google Business Profile, your Wikipedia entry and the third-party sites that talk about you. Search Console’s generative AI report will also show whether your brand searches are getting AI Overview impressions here in Australia.

Fact-check your AI content and ditch the fake authors, says Google

Google has quietly updated two of its guidance documents this fortnight, and both take aim at lazy AI content.

Its guidance on AI-generated content now says site owners should “manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing”. Google calls this step “critical”, which isn’t a word it throws around much in its help docs. Its helpful content guidance also now warns against “deceptive authorship information”. Made-up writer names, AI-generated headshots and fake credentials are all called out as deception that makes a page untrustworthy. And, as we all well know, organic visibility is fuelled by trust, so anything that chips away at trust will hurt a brand’s SEO and GEO efforts.

Both changes landed as Google’s September 2026 spam update was wrapping up.

Sources: Google guidance on AI content

Our take: using AI to help write content is fine. Help being the important word here. Publishing it unchecked under a made-up name is not ok. Make sure every article has a real person’s name on it (preferably the person who worked on it!), with a short bio that shows why they know their stuff (an authority signal). Real experts help with AI visibility too, as AI tools are more likely to quote content they can trace back to a credible person.

ChatGPT ads are getting pictures, proper measurement and "best for" labels

OpenAI announced on 5 October that ChatGPT ads are going visual. Image-led ads will be tested with a group of US advertisers from later this month, showing up first while people are generating images in ChatGPT. They’re labelled and kept separate from the images users create.

The bigger news for marketers is measurement. OpenAI has signed up a long list of partners, including AppsFlyer, Triple Whale and Northbeam for click tracking, LiveRamp and Tealium for conversion data, and Haus and Measured for tests that show whether the ads actually drive extra sales. Early partner results look promising: WeightWatchers’ cost per acquisition came in 15.3 per cent below its paid search average. Brand safety pilots with DoubleVerify and Integral Ad Science are also underway, and OpenAI says ChatGPT now reaches 1.2 billion people a week.

On the unpaid side, OpenAI is testing new labels on product results. One product gets a green “best all-rounder” tag, while others get grey tags like “best for beginners” or “best cushioning”.

Sources: OpenAI

Our take: ChatGPT image ads are still US-only, so watch this space. The latest measurement tool enhancements will really help build more credibility with this very new ad platform. That’s something we feel is needed, given the challenges the pilot faced when it launched here in Australia.

ChatGPT is deciding which product is “best” for what, based on reviews, product descriptions and what people say about you online. So, brands really need to focus on driving positive sentiment and authority, and ensuring product pages and feeds are in tip top AI Visibility shape in order to get assigned a label.

Meta's new AI agent doesn't just recommend, it books and buys

Meta launched Muse in the US on 8 September, and it has taken off. Muse is an AI agent rather than a chatbot. It has its own web browser, connects to your apps and gets things done for you. For example, through OpenTable and Ticketmaster, it can now book restaurants and concert tickets. And, through Stripe and Shopify, it can now buy things for you. Wild!

Reports say more than 3 million people use it every week, and it briefly knocked ChatGPT off the top spot for free apps on the US App Store (with plenty of help from Meta promoting it across its own apps). It’s free with usage limits, with paid plans at US$20 and US$100 a month. No word yet on an Australian launch, and privacy and security concerns have been raised in the US press.

Sources: Meta

Our take: AI isn’t just recommending brands anymore, it’s choosing them, interacting with them, and paying for them too. When an agent does the shopping, it goes with the brand whose information is easiest to find and act on, such as clear pricing, live stock levels, booking systems it can plug into, and positive reviews. If you sell through Shopify or take bookings through OpenTable, you’re already in the running. If not, it’s worth asking how easily a machine could buy from you.

What does Google pay publishers for AI answers? Less than 0.1 per cent

Back in edition three, we covered Google’s trial of paying some publishers when their content is used in AI answers. Now we’ve got numbers. According to The Information, around 100 publishers are in the pilot, and for several small and mid-sized sites, the payments come to less than one-tenth of 1 per cent of their advertising revenue. Not unexpected, but this low? Ouch.

At the other end, one recent participant has earned between US$50,000 and US$60,000 so far, and one early participant earns more than US$1 million a year. Smaller sites have made under US$1,000 across several months. Google says payments depend on how much a source contributes to an AI answer, but some publishers say they’ve got no idea how it’s worked out, and the amounts change month to month.

Our take: for now, Google decides what your content is worth, and the answer, in general, is “not much”. For brands, the value is in being visible and, ideally, cited. Every time an AI answer names you as a source, that’s visibility you didn’t pay for.

Google opens a fast lane for forum posts, and puts Maps reviews behind a login

On 8 October Google announced its UGC Fresh Data Program. It lets big forum and social platforms push new posts and engagement data straight to Google Search, instead of waiting for Google to find them. Platforms need a big user base, public pages, active moderation and the right technical setup, and applications take six to eight weeks.

Meanwhile, Google Maps now asks people to sign in to a Google account before they can read all the reviews on a business listing, sort them or leave one. Google hasn’t said why, but it looks aimed at stopping bots, scrapers and rival AI tools from hoovering up review data.

Sources: Search Engine Roundtable on the UGC program

Our take: Google wants more real conversations in Search, faster, and it wants to keep its review data to itself. So, what customers say about brands in forums, in reviews and on social carries more weight than ever. Our recco is to keep encouraging reviews, reply to them, and make sure your brand shows up in the online communities your customers hang out in.

A few more bits from around the SEO & AI Visiblity traps

• Multi-country filters in Search Console: you can now filter by several countries at once. Super handy for brands in multiple markets. Source

• Search Console running late: performance data was badly delayed as of 11 October, so don’t panic if recent numbers look odd. Source

• No more review stars for health pages: Google has dropped review snippets from healthcare pages in search results. Source

• “Dive deeper” in Discover: Google is testing an AI summary of a topic with links to related stories underneath. Yet another spot where AI sits between your content and your readers. Source

And finally, for some weekend viewing: did AI just go rogue? Netflix's AI Gone Wild

Need something to watch this weekend? Netflix’s new 27-minute documentary, Instadocs: AI Gone Wild, has been streaming since 12 October. It tells the story of a July 2026 incident where AI agents being tested inside OpenAI found their way out of a sealed-off test environment and broke into Hugging Face, one of the biggest platforms for sharing AI models. They weren’t after money. They were reportedly hunting for the answers to the test they’d been set (cheeky!). Hugging Face counted around 17,600 actions over about two days.

Sources: Martin Cid Magazine, AI Magazine.

Our take: I think we can all agree on this one… this is a bit bat shit crazy. Nobody told these agents to break in anywhere. They were set a test, went looking for the answers, and hacked into another company to get them.

Now think about how much we’re handing over to AI. Tools like Meta’s Muse can browse, book and buy on people’s behalf, and brands are plugging AI into their websites, customer data and ad accounts. If an AI agent can break into another company’s systems just to pass a test, what could one do with access to yours?

Before you connect anything, check what it can see, what it can change, what it’s allowed to spend, and who’s keeping an eye on it. If it goes wrong, it’s your brand’s name in the headlines.

________________________________________________

That's edition five done and dusted. ChatGPT's traffic went into hiding, Google started writing its own version of your brand, and AI agents started doing the shopping. None of it changes the basics, it just means the brands that make it easy for AI to get the facts right are the ones that get picked. Back in a fortnight with more AI visibility news. Stay tuned!