SEO tracking: Google and ChatGPT are changing the rules of monitoring

Between September 27 and 29, 2026, a quarter of the Google search result pages monitored by rank trackers showed completely different sites than usual. If you saw a drop in rankings during that period, it might never have actually happened.

Does this really affect me, or is it only a problem for big sites?

It concerns anyone who uses a rank tracker or measures traffic from ChatGPT in Google Analytics. If you have an e-commerce site, a professional firm, or a manufacturing company with an optimized website, your October reports might contain incorrect data.

In October 2026, two independent things happened in parallel that made two of the most used metrics in SEO monitoring unreliable. The first concerns Google and rank trackers (the software that checks your site's position every day). The second concerns ChatGPT and how Google Analytics counts visits coming from the AI assistant.

The practical result: if you submitted a report in October, or if you are looking at data from that month, some figures might be telling a false story. Not because of you, not because of the agency, but because of how measurement systems work right now.

What are Google's 'poisoned' SERPs and why do they exist?

Google has started serving altered results pages to automated bots to defend against scraping by AI systems. Rank trackers, which function as bots, have been receiving these fake pages instead of the real ones.

A rank tracker works like this: every day it runs thousands of automatic searches on Google and records the position your site appears in for each keyword. Google has detected this automated traffic and, according to analysis by the company Nozzle, has responded by serving modified results pages to these bots.

Nozzle analyzed 17.8 million search result pages collected between September 13 and October 9, 2026. 8% of those pages were ‘poisoned’: the sites that normally appear for that keyword were almost all absent. The peak was between September 27 and 29, when about a quarter of the pages were altered. From October 1 onwards, the phenomenon stabilized between 5% and 11% per day.

On the affected pages, the sources cited in AI Overviews (the summaries generated by Google at the top of results) maintained only 42.3% of the sites present on the normal page, against the 68.8% benchmark. The AI Overviews themselves appeared in 34.9% of cases instead of the usual 54.2%, according to the analysis published on DEV Community by Nozzle .

The problem doesn't affect your real users: they see normal results. It only concerns the data your rank tracker recorded during that period.

What happened to ChatGPT traffic in Google Analytics?

After the release of GPT-6 on October 7, 2026, ChatGPT stopped adding the identification tag to outbound visits. Google Analytics no longer recognized those visits as coming from ChatGPT and counted them as 'direct traffic'.

Google Analytics recognizes where a visit comes from thanks to a parameter called UTM (Urchin Tracking Module), which is a label added to the link's web address. ChatGPT used the tag utm_source=chatgpt.comto signal that the visit came from its assistant.

With GPT-6, that tag is gone. Visits were arriving without a label and with the attribute noreferrer , which prevents the browser from communicating the origin. Result: Google Analytics classified them as ‘Direct traffic’, the residual category that collects everything that does not have an identifiable source.

OpenAI confirmed the bug on October 14 and deployed the fix on October 15. According to data from Adobe reported by NoticeMeSenpai , at 9 AM on October 15, UTM tags were back to 95–100% of the previous week's levels. The compromised data window runs from October 8 to 15, 2026.

If you see a drop in ChatGPT traffic in your reports during that period and an increase in direct traffic, it's not a real problem: it's a measurement issue. Note this anomaly in your reports before someone draws the wrong conclusions.

Why are people talking about this now: is AI traffic really growing?

Referral traffic from AI platforms has grown 16x in two years. Measuring it poorly today means missing out on strategic insights for a channel that carries more and more weight.

According to SE Ranking , AI platforms now account for 0.32% of all global web traffic. It might seem small, but it is 16 times the value of 2024. For an e-commerce site with 50,000 visits per month, that means about 160 visits per month from AI, compared to 10 two years ago.

It's not just ChatGPT. The composition of AI traffic is changing rapidly: ChatGPT was dominant with 89.1% of AI referral traffic between May and August 2025, but it dropped to 62.6% between March and April 2026. Claude went from 1.4% to 18.5%, Gemini from 2.4% to 10.6% in the same period, according to data from Goodie/HiGoodie .

This means that monitoring only ChatGPT is no longer enough. A company that only measures traffic from ChatGPT in GA4 is ignoring almost 40% of the AI traffic it receives. And if that 40% is misclassified, decisions on where to invest in content become even less reliable. To learn more about how SEO is adapting to this scenario, read our article on ChatGPT citing primary sources and what changes for visibility .

What changes with Google AI Overviews and featured links?

Google is moving links to higher positions in AI responses and testing new dedicated sections. For those monitoring rankings, this creates a misalignment between what the user sees and what Search Console reports.

AI Overviews are the boxes generated by Google that appear at the top of search results and synthesize an answer. According to an Ahrefs analysis on 300,000 keywords , in March 2026 AI Overviews appeared in 48% of all queries, up from 34.5% in December 2025: a 58% increase in three months.

Google is now moving inline links (links inserted directly into the AI response text) higher up in responses. It is also testing a dedicated section at the top of some responses with links to official pages. The problem for those monitoring: in Search Console, every link within an AI Overview receives the same position as the AI Overview as a whole. So if a link moves up within the AI response, Search Console does not record it as a position improvement.

To be included as a ‘preferred source’ in Google carousels, your site must be enabled for generative AI features in Search Console. If you haven't done this yet, you're missing out on visibility that doesn't show up in any report. This topic connects directly to what we analyzed in the article on AI Overviews on brands and SEO visibility .

What does Bing offer to understand where your site isn't appearing in AI answers?

Microsoft added a tool called Topics Portfolio to Bing Webmaster Tools that shows which topics your site isn't mentioned for in AI responses, while competitors are.

Bing Webmaster Tools (Microsoft's free tool for monitoring visibility on Bing) has introduced a beta section called Topics Portfolio within the AI Performance report. It groups the queries for which your site is cited in AI responses into macro-topics, with approximate data on the total volume of citations per topic.

The most useful part for those looking to grow is the list of ‘opportunity queries’: searches where other sites are cited by Bing’s AI, but yours isn’t. It’s a direct way to identify content gaps. Heads up: the list might also include queries related to competitor brands, which aren’t relevant to you. Microsoft itself recommends using this data as directional guidance, not as precise numbers.

Three concrete scenarios: how does all this translate for an Italian company?

An e-commerce site, a consulting firm, and a manufacturing company react differently to these changes. Here is what to check in each case.

Illustrative examples: numbers are meant to show the reasoning, they are not real customer data.

The topic of visibility in AI responses doesn't just concern large publishers. As we explained in the article on SEO without Google index and what changes for visibility , even SMEs need to start monitoring their presence in AI systems, not just in traditional SERPs.

How do you do the right double-check for every number in the report?

Every SEO metric now has an alternative control source. Using them together is the only way to distinguish a real problem from a measurement artifact.

What are the most common mistakes made when data looks weird?

The most common mistake is reacting to data without verifying it: optimizing pages that haven't actually lost rankings, or panicking over traffic drops that never happened.

Here are the three most frequent mistakes in this type of situation:

• Optimize in response to false data. If the rank tracker shows a drop and you edit the page content, you risk breaking something that was working. Always check with Search Console before touching anything.
• Presenting the ChatGPT drop as a real traffic loss. A report showing -45% ChatGPT sessions in October without a note about the UTM bug leads the client to the wrong conclusions. Always add context.
• Ignore Claude and Gemini in monitoring. If you only measure traffic from ChatGPT, you're missing out on almost 40% of AI traffic. With Claude at 18.5% and Gemini at 10.6% of total AI referrals, they are no longer negligible.

A less obvious error concerns Search Console and impressions. As documented by Garrett Digital , Google's removal of the &num=100 parameter in September 2025 caused a drop in impressions for 87.7% of the 819 sites analyzed. Those unaware of this change interpreted the drop as a real visibility issue.

How to tell if your SEO data is reliable right now?

There are precise indicators that allow you to distinguish a measurement problem from a real visibility issue. Here they are in a reference table.

This data from Search Engine Land show why monitoring visibility in AI Overviews has become critical: even those in the first position lose almost 60% of clicks when an AI Overview appears. Knowing if you are cited in the AI Overview is more important than knowing if you are in the first position. The topic is explored in depth in our article on Google auto-expanding AI Overviews and the impact on organic CTR .

How to get started: what are the concrete steps to get monitoring in order now?

Five practical actions are enough to have a reliable monitoring system that covers both traditional SEO and visibility in AI answers.

So, is it worth reviewing your company's SEO monitoring right now?

It is worth it if You have an active rank tracker, you measure traffic from ChatGPT in GA4, and you present monthly reports to clients or management. The data from September–October 2026 must be verified and annotated before anyone makes the wrong decisions.

A middle ground You only use Search Console without a rank tracker and have little traffic from AI. The direct impact is minor, but it is the right time to enable the generative AI report and start monitoring visibility in AI responses.

Not yet if your site doesn't have active SEO optimization yet and you aren't measuring traffic by source. In this case, the tracking issue is secondary: first comes building a measurable organic presence.

How we see it at SHM

For us, technology is a means: the goal is for companies to make decisions based on real data, not measurement artifacts. What happened in October 2026 with rank trackers and ChatGPT's UTM tags is a reminder that no number should be read without understanding how it was collected.

When we work on technical SEO and AI visibility for our clients, the first thing we do is build a multi-source monitoring system: rank trackers, Search Console, GA4 with segments for AI sources, and now also Bing Webmaster Tools for opportunity queries. Not because any single tool is wrong, but because no single tool tells the whole story.

If you want to understand how your site appears in Google, ChatGPT, and Bing responses, or if your current reports are reliable, you can start with an audit using our service of SEO for AI or explore how integrate AI into your digital strategy . For those who want to better understand the topic of visibility across all AI engines, the starting point is our collection on SEO and visibility for AI .

Data changes. Measurement systems change. What doesn't change is the need to understand what they are actually measuring.

Frequently Asked Questions

The problem isn't that the data has changed: it's that many were reading it as if it had always been reliable.