
Are AI Overviews killing website traffic? Partly, yes. AI Overviews and AI Mode now answer more queries right on the results page, and I think clicks from Google will keep falling, especially for informational questions. But a lost click is not a lost channel. What matters now is whether Google's answer names your business, and GA4 sessions and Search Console clicks can't show you that.
Are AI Overviews killing website traffic? My short answer
Clicks will keep shrinking. Google now shows the AI Mode prompt box directly inside AI Overviews for some queries, in place of the "Show more" button. That moves people deeper into the AI answer and further from the list of links.
Google says it is working on the problem. In October it announced five updates meant to send more people to the web, including more inline links placed higher in the answer. In November it brought Gemini 4 Hydrogen to Search. Google says the new model's improved query fan-out links to a wider range of sites. (Query fan-out means the AI splits one question into many smaller searches.) But Google has not shared traffic figures for these changes. I would plan for fewer clicks per query.
The channel itself is not dying, though. For many questions, the research step has moved inside the answer. If the answer names you, the customer can still arrive, through a brand search, a direct visit or a phone call. If it names a competitor, you may lose the customer before anyone has a reason to click. That is the real risk, and a clicks chart doesn't show it.
What Reddit's drop does and doesn't prove
On 29 October 2026, Reddit CEO Steve Huffman told CNBC that search referrals from Google were "down meaningfully in this quarter" because of the "shift to AI mode and AI overviews". He expects the trend to continue in the fourth quarter. According to Search Engine Roundtable, by 2 November Reddit's stock had fallen 36% over the previous week. Huffman called Google's AI shift "a fundamental shift in how the Google product works, from being a traffic driver to whatever's going on now."
The same article adds an important caveat. SEO consultant Glenn Gabe noted that Reddit's AI-translated pages, tens of millions of URLs, lost heavily in the May 2026 core update and the June 2026 spam update. So part of Reddit's drop may come from rankings, not from AI Overviews.
My takeaway: even for Reddit, AI effects and ranking effects are mixed together. Yours probably are too. Don't blame AI Overviews by default.
Why Search Console clicks and GA4 sessions can't judge AI search
Your usual reports measure the last step of a journey that now often starts somewhere else. There are three reasons.
- Google's AI report shows no clicks. Search Console's generative AI performance reports launched in June 2026 and show impressions only. In November, Google's Gary Illyes pushed back on claims that click and query data were coming. He said the team wants to add "data that makes sense" but has nothing to announce.
- The first summary of your brand is Google's. On many desktop brand searches, Google now shows an AI Overview in the right-hand column, where the knowledge panel used to be. People read it, and no click gets logged.
- Referrers break. A UTM tag is a label in a link that tells analytics where a visit came from. In October 2026, ChatGPT stopped adding UTM tags to outbound links for about a week, so those visits showed up as direct traffic. OpenAI confirmed the bug and fixed it. Gemini only added UTM parameters that same month.
Calling the channel dead because the clicks line is going down is, I think, the market's main mistake right now. You are counting the visitors who still needed your page after reading the answer, and ignoring whether the answer mentioned you at all.
What I'd measure instead
| Metric | Question it answers | Where to get it |
|---|---|---|
| Share of Model | Do AI answers name us on our key questions? | Manual checks, OpenAI Business Hub, Bing AI Performance |
| AI bot access | Can AI bots actually read our facts? | Server logs, CDN logs, bot reports |
| Brand demand | Are people still looking for us by name? | Search Console brand queries, Microsoft Clarity |
How often AI answers mention you on your key questions
Share of Model is the share of AI answers on your key questions that mention your business. Build a fixed list of 10–20 questions your customers really ask. Run them on a regular schedule in ChatGPT, Gemini and AI Mode, Perplexity and Claude, using the same wording every time. Note who gets named instead of you and which sources the answers cite. Our research on measuring Share of Model explains the method in detail.
First-party data is starting to appear as well. Businesses in ChatGPT shows verified businesses their mentions, profile views and website clicks in a Business Hub. Claiming a profile is still in beta, insights appear only once there are enough conversations, and it shows no prompts or competitor data. Bing's AI Performance report now includes Topics Portfolio, which shows queries where competitors are cited and you are not. Microsoft says its numbers are approximate.
Whatever the tool, the principle stays the same: measure before and after each change. That is how we judged the Sereda.ai case. Two researchers independently scored ChatGPT and Gemini answers to the same 4 questions, before and after the AI Page went live on 22 October 2025. The averaged score went from 24 to 87 in ChatGPT and from 28 to 91 in Gemini. One case is not a guarantee, but this is the kind of measurement I mean.
Whether AI bots can read your facts
AI bots read a site differently from Google and from people. They need clear, unique facts. Check your server logs: which bots visit (OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-SearchBot, Googlebot), which pages they request and which status codes they get back.
An open robots.txt is not proof of access. OpenAI's Aura safety check is a central blocklist that decides whether ChatGPT's live search and agents can open a site. It blocked one publisher by mistake from May 2026, even though its robots.txt was open, and the publisher's logs showed no ChatGPT requests at all. Chetver Pulse shows which AI bots visit your site, what they read and where they hit errors, and it says what to change for each problem.

Whether brand searches hold up
Filter Search Console for branded queries. In Microsoft Clarity's AI Visibility dashboard, you can now customize your brand terms, including abbreviations, typos and translations. If brand demand stays steady while non-brand clicks fall, I read that as AI answers taking over the research step, not as lost demand. The warning sign is falling brand demand.
How to tell a real drop from noise this November
| Event | Dates | What happened | What to do |
|---|---|---|---|
| Search Console click spikes | 31 Oct – 2 Nov 2026 | On several sites, clicks jumped 2x–6.5x while impressions and rankings stayed flat, mostly on desktop and on pages ranking around positions 11–55 | Annotate the dates. Don't report the spike as growth. |
| November 2026 core update | Started 10 Nov 2026, up to two weeks | Global, all languages | Compare pages only after the rollout finishes |
| Rank-tracker errors | Since 20 Sep 2026 | Nozzle reports that Google serves some incorrect results to automated requests | Cross-check tracker swings against Search Console |
The early-November click spikes
According to Search Engine Journal, Marie Haynes thinks the spikes come from outside bots or a Google logging bug. Google has not confirmed either cause. You don't need to change anything on your site, but leave these days out of any month-over-month comparison.
The November 2026 core update
The Google Search Status Dashboard lists the update as starting on 10 November. As of today it may still be rolling out. Gemini 4 Hydrogen also arrived in Search in November, so log both. Judge your top pages by comparing before and after the completion date, not halfway through the rollout.
Other noise
Nozzle reports that Google has been serving incorrect results to rank trackers since late September. Google has not commented. Tagging bugs like the ChatGPT one above can move visits into "direct". Before you blame AI Overviews, rule these out.
Make your business facts easy for AI to quote
You can't buy a mention, but you can make your facts easy to find and hard to get wrong.
- A facts page for AI. One page with your prices, opening hours, contacts and services. Chetver's AI Page is built from your website, published on aipage.yoursite.com and comes with an llms.txt file. llms.txt is a plain-text file that points AI tools to your key content. Our overview of llms.txt and other AI-crawler signals covers how it fits with robots.txt.
- Schema markup. Use Organization or LocalBusiness, Product and Offer with prices, opening hours and services, and keep it matching the visible text.
- Clean bot access. Let answer bots through robots.txt. Make sure your CDN or firewall doesn't challenge them, and that your key facts come back with a 200 status in plain HTML.
- Training is a separate choice. Opting out of AI training is separate from being readable in AI answers. Google, Microsoft, OpenAI and Anthropic plan to support a shared robots.txt token for training opt-outs. Bing will add separate training and grounding controls to Webmaster Tools on 15 December 2026.
Our research on moving from SEO to GEO explains why these basics now carry more weight than another blog post.
What I would do in your place
- Clean your November data. Annotate 31 October to 2 November. Mark 10 November and the date the core update finishes, then compare your top pages before and after.
- Split brand and non-brand queries. On informational pages, compare impressions with clicks and check whether an AI Overview now appears for those queries.
- Run your key questions. Ask your 10–20 questions in ChatGPT, Gemini and AI Mode, Perplexity and Claude. Write down who is named and which sources are cited, then repeat on a fixed schedule.
- Check real bot access. Look at your logs and CDN, not only at robots.txt.
- Publish or refresh your facts. Use a facts page with llms.txt, plus schema that matches your pages.
Clicks alone undercount AI search, and they will keep undercounting it. Measure whether the answer includes you. A simple place to start is the free AI check, which shows how AI sees your business and what to fix first.
Frequently asked questions
Are AI Overviews really reducing traffic to my website?
Often, yes, especially on informational queries. AI Overviews and AI Mode answer the question on the results page, and Reddit's CEO said in October 2026 that this cut Reddit's Google referrals. But drops often have other causes too, like core updates or tracking bugs, so check those before blaming AI.
How can I tell if a traffic drop is from AI Overviews or the November 2026 core update?
Wait until the core update finishes rolling out. It started on 10 November 2026 and can take up to two weeks. Then compare your pages before and after. If impressions are steady but clicks fall on informational pages that now show an AI Overview, AI answers are the likely cause. If rankings and impressions fall, look at the update.
Why did my Search Console clicks spike between 31 October and 2 November?
Several sites saw clicks jump while impressions and rankings stayed flat, mostly on desktop. Marie Haynes attributes this to outside bots or a Google logging bug, and Google has not confirmed the cause. Annotate those dates and don't report the spike as growth.
Does Search Console show clicks from AI Overviews and AI Mode?
No. Search Console's generative AI performance reports show impressions only. Google's Gary Illyes said in November 2026 that there is nothing to announce about adding click or query data.
Should I block AI bots to protect my traffic?
I wouldn't. Blocking the bots that fetch pages for AI answers removes your facts from those answers, and I don't expect it to bring the clicks back. If you don't want your content used for AI training, you can opt out of training separately. Bing, for example, is adding separate training and grounding controls. Keep the answer bots able to read your key pages.