GEO Roundup October 2026: More links in AI Mode, no click data

Google is adding more links to AI Mode and still won't say how often anyone clicks them. Other important GEO stories of October 2026: the spam update has finished, tools for measuring AI visibility are getting better while classic rank trackers are getting worse, and a large study shows which sources ChatGPT, Claude, Gemini and Perplexity actually cite.

TL;DR

More links, no click data: Since 15 October, Google shows more links in AI Mode and AI Overviews, placed higher up. Search Console still provides no click data for them.
Spam update complete: The September update ran from 24 September to 8 October. Programmatic SEO remains the number one risk.
Measurement is getting better and worse at once: Bing now groups AI visibility by topic. According to Nozzle, rank trackers have been receiving partly wrong Google results for weeks.
What gets cited is mostly what others write about you: More than 84% of AI citations come from earned media, only around 15% from owned channels.
The question decides the format: For how-to questions, 68% of citations come from practical guides; for trend questions, 41% come from expert analysis.

1. Google: more links in AI Mode and AI Overviews

What happened: On 15 October, Google announced several changes. Inline links appear more often and higher up in responses. Where there are only a few official places for a task, Google is testing a dedicated section with those links at the top of the response. Article carousels for timely topics are expanding to more countries and languages. The number of people using Preferred Sources has grown more than sevenfold since the start of the year.

Context: More links do not mean more clicks. Search Console's AI reports still show no click data, as Barry Schwartz points out at Search Engine Land. At the same time, more and more search is moving into AI surfaces: according to AlsoAsked, AI Mode now answers 97% of People Also Ask questions, up from 86% in August (sample: 19.2 million English-language answers).

What to do now:

Check whether your site is the official destination for its core tasks: pricing, sign-up, documentation, contact. Those are exactly the links Google wants to show prominently.
Ask loyal readers to mark your site as a Preferred Source. The feature is growing, and the carousels highlight these sources.
Don't judge AI visibility on clicks alone. The data simply isn't there.

2. Spam update done, programmatic SEO in the crosshairs

A tower of identical template pages topples while Blink carries away a single solid page

What happened: Google's September 2026 spam update started on 24 September and, according to the Google Search Status Dashboard, was complete on 8 October. It applies globally and to all languages. Google did not say what it targeted.

Context: John Mueller had set the tone in early September. Writing on Bluesky about a specific site, he said programmatic SEO "often leads to a site that's either spam, borderline spam, or low quality", and that Google's systems had possibly lost faith in the site. A prediction model by Dan Taylor also expects a Panda-like update against scaled AI content. That is a forecast; Google has confirmed nothing of the sort.

What to do now:

Look for pages built from templates and data combinations: city × service, tool × attribute, "what does X mean?" in series.
Merge thin variants or remove them from the index. Twenty good pages beat two thousand mediocre ones.
If you use AI-written text, every page needs something only you can provide: data, experience, a reasoned opinion.

3. Measuring AI visibility: new reports, unreliable rank trackers

What happened: Google has rolled out Search Console's AI performance reports and AI controls globally, along with a new filter for multimodal searches. On 13 October, Bing Webmaster Tools launched Topics Portfolio as a beta inside its AI Performance report. It groups the queries for which your site appears in AI answers into topics and shows demand and competition per topic.

Context: Classic rank tracking, meanwhile, is getting less reliable. According to the vendor Nozzle, Google has been serving rank trackers non-standard results pages since 20 September, peaking at 25% and running at 5 to 11% a day since 1 October (reported by Search Engine Journal). That is one vendor's measurement; Google has not commented.

What to do now:

Set up Bing Webmaster Tools if you haven't yet. Bing powers Copilot and currently offers the most detailed AI visibility data.
Check the AI report in Search Console every month and record the trend.
Treat third-party ranking data as a direction, not as the truth.

4. Which sources AI systems cite

Blink pushes its own page into a large stack of third-party sources that the AI answer cites from

What happened: At the end of October, Muck Rack and Generative Pulse published their study "What Is AI Reading?". It is based on more than 40 million AI responses and 207 million citations from 2026, plus an analysis of 1,020 prompts in September. Around 99% of citations are unpaid, and more than 84% come from earned media and third-party sources. Journalism's share fell to 22% (May: 27%), while companies' owned channels rose to 15.2% (May: 13.7%).

Context: The platforms differ markedly, as Axios summarises:

ChatGPT cites government and nonprofit sources most often (34.6%).
Gemini relies most on company-owned content (34.4%).
Perplexity draws most on journalism (27.8%).
Claude mainly uses third-party corporate content (23.5%).

The type of question matters even more. For how-to questions, 68% of citations come from practical guides; for best-of queries, 45% come from buyer's guides and reviews; for trend questions, 41% come from expert analysis.

What to do now:

For every important page, ask which question it answers and pick the matching format: guide, comparison or analysis.
Plan for mentions on other sites: trade media, associations, studies, guest articles. GEO does not stop at your own domain.
Don't optimise for "the AI". ChatGPT, Gemini, Perplexity and Claude prefer different sources.

5. Crawling and discovery

What happened: On 15 October, Cloudflare and OpenAI published results from a pilot in which Cloudflare sends signals about new and changed pages to OpenAI's search crawler. In a cohort of around 1% of Cloudflare's network, this helped surface hundreds of millions of new pages for ChatGPT search. DeepSeek is opening its search to developers through a Search and a Fetch API and says it filters out SEO spam and "GEO manipulation". And Google reminded site owners to serve a 503 status code during maintenance, prompted by two critical PHP vulnerabilities.

Context: If you want AI search to find you, you have to be technically reachable. It sounds obvious, yet it regularly fails because of bot blocks nobody set on purpose.

What to do now:

Check in Cloudflare (AI Crawl Control) or your firewall which AI crawlers you block, and whether that is intended.
Serve a 503 during planned downtime, not an empty page or a 200 with an error message.

My take

Google shows more links and won't say whether anyone clicks them. My reading: clicks are not the metric Google wants to be judged by. For you, that means measuring AI visibility yourself instead of waiting for the report that would answer the question.

For me, the Muck Rack numbers are the real story of the month. More than 84% of citations come from third parties. If you see GEO purely as on-page optimisation, you're pulling the smaller lever. And if you're thinking about generating thousands of pages with AI to show up in AI answers: Mueller has already said everything there is to say.