SEO in the Age of AI Browsers: A Framework for Visibility in the Citation Economy

On October 3, 2026 my website ranked first on Google for “google update august 2026”. The AI Overview above it did not mention it. For “google spam update 2026” it was the other way round. I was not on the first results page, yet the AI Overview listed me as its second source.

In December 2025 I wrote down a framework for the citation economy here. I still think the core idea is right: visibility increasingly happens in AI answers, even when nobody clicks afterwards. Some of the evidence from back then does not hold up to a second check though. Above all I declared schema markup the key to citations. That was wrong.

Today I no longer have to guess. I measured where Google names my website in AI Overviews. Since this year Bing and Google also show themselves how often a website appears in AI answers. The terms behind all this are sorted out in my post on the interplay of SEO, AIO, GEO and LLMO.

By citation economy I mean this: part of a page’s value now comes from an AI answer naming it as a source. The click still matters. It is just no longer the only form of visibility.

The reason is search behavior. According to SparkToro, 68.01 percent of Google searches in the US ended without a click in the first four months of 2026. The figures come from Similarweb’s clickstream panel. For 2024 SparkToro gives 60.45 percent, measured with a different panel. Pew Research measured the effect of AI summaries. With a summary, US users clicked a traditional result in 8 percent of visits in March 2025. Without one it was 15 percent.

Ranking does not lose importance because of this. Google picks the sources for AI Overviews and AI Mode from its index. If you are missing there, you will not be cited either. In my view the citation economy therefore does not replace ranking. It adds a second question: does the answer name me, even if nobody clicks?

An AI browser builds AI directly into the browser. It summarizes pages, takes open tabs into account and on request completes tasks in several steps. When I wrote this post in December 2025, that looked like the next big shift. Ten months later the market looks like this:

• Perplexity Comet: released for free to everyone on October 2, 2025, launched on Android in November 2025 and on iOS in March 2026
• Dia: The Browser Company has belonged to Atlassian since October 21, 2025. Dia is free for macOS and Windows.
• ChatGPT Atlas: launched on October 21, 2025 for macOS. OpenAI announced it would discontinue it on August 9, 2026.
• Microsoft Edge: Microsoft retired Copilot Mode on May 13, 2026. Agentic browsing is now called “Browse with Copilot” and is only open to Microsoft 365 Premium subscribers in the US.
• Chrome with Gemini: Since January 2026 Google has offered “auto browse” in the US for AI Pro and Ultra subscribers. Germany, Austria and Switzerland are missing from Google’s list of countries for Gemini in Chrome.
• Opera Neon: The waitlist was dropped in December 2025. The agentic features cost 19.90 dollars a month.

In my view agentic browsing is therefore not yet a mass phenomenon, certainly not in German-speaking countries. For your visibility what counts most is where the AI systems get their sources.

What sets Google, ChatGPT and Perplexity apart?

The three systems draw on different sources. A good Google ranking does not automatically cover ChatGPT and Perplexity as well.

AI Overviews and AI Mode compile their answers from the Google index. seoClarity analyzed around 362,000 AI Overviews from the US on October 12, 2025. In 90 percent at least one source came from the organic top 10. Of the 5.1 million citations, however, only 56 percent came from the top 20. My own measurement further down shows the same pattern on a small scale.

According to OpenAI, ChatGPT now has more than one billion weekly active users. Its sources overlap little with Google. Ahrefs analyzed 15,000 prompts, with data from July 2025. Only 8 percent of the URLs ChatGPT cited in its answer text were also in Google’s top 10 for the same prompt.

According to its CEO Aravind Srinivas, Perplexity processed 780 million queries in May 2025. I have not found a more recent official figure.

What data shows the decline in clicks?

The most thorough measurement I know of comes from Seer Interactive. Seer tracked 3,119 informational search terms from 42 clients between June 2024 and September 2025. Behind them are 25.1 million organic and 1.1 million paid impressions.

• Organic click-through rate for queries with an AI Overview: from 1.76 to 0.61 percent, a drop of 61 percent
• Paid click-through rate for the same queries: from 19.70 to 6.34 percent, a drop of 68 percent
• Organic click-through rate for queries without an AI Overview: from 2.72 to 1.62 percent, a drop of 41 percent

I find the last line the most important. Users click less even without an AI Overview. Seer suspects they increasingly look for answers elsewhere, for example in ChatGPT, in Perplexity or on social platforms.

Seer itself now reports a countermovement. In a new, broader dataset covering 53 brands, organic click-through rate for queries with an AI Overview rose from 1.3 percent in December 2025 to 2.4 percent in February 2026. This cannot be compared directly with the series above. For me it means: the low point of 2025 does not have to be the end.

How often Google shows an AI Overview at all also varies widely. Semrush measured one for 6.49 percent of keywords in January 2025, for nearly 25 percent in July and for 15.69 percent in November. For my own 50 SEO-related queries it was exactly 47 in October 2026.

In AI Mode the effect is even stronger. According to Search Engine Land, Eric Van Buskirk of Clickstream Solutions analyzed around 846,000 US Google search sessions from February and March 2026. In 64 percent of AI Mode tasks users did not click at all. In 88 percent they took the AI’s shortlist as it was.

According to Google, AI Mode is particularly helpful for questions that need comparisons or further exploration. In my view that is the most important difference between AI Mode and AI Overviews.

No provider fully discloses how its AI system picks sources. For Google there are two clues: what Google documents and what can be measured.

What Google itself says

For AI Overviews and AI Mode Google names one requirement in its documentation on AI features. A page has to be indexed and eligible to appear in Search with a snippet. According to Google there are no additional technical requirements. No special optimizations are necessary either, the familiar SEO fundamentals still apply.

One technique explains a lot. Google calls it “query fan-out”: the AI issues several related searches on subtopics for one question. That way pages can end up in the answer that do not even rank for the original query. This way Google AI Overviews work fits what I measure below.

On trust Google is clear in its guidance on helpful content. Of the four parts of E-E-A-T, trust is the most important. According to Google, E-E-A-T itself is not a specific ranking factor though. Google uses a mix of factors meant to identify content with good E-E-A-T.

What my own measurement shows

On October 3, 2026 I read the first results page on Google Germany for 50 queries, AI Overview included. 38 come from my Search Console, 12 from my content plan.

47 of the 50 queries showed an AI Overview. My website appeared as a source in 12 of them. I got six of these citations without a spot of my own among the first ten organic results. Conversely, I was on the first page for three queries and was not cited, once even from position 1.

I was cited mostly for Google updates, for new Google features and for the term “tl;dr”. For basic terms from my content plan such as “canonical tag” or “meta description” my website did not appear once. The most-cited other domain was YouTube with 40 mentions. developers.google.com followed with 17 and sistrix.de with 15.

In its Webmaster Tools Bing shows the phrases its AI uses to retrieve a website’s content. I also ran 32 of my website’s most-cited phrases on Google. Every one of them showed an AI Overview, four of them without any source. Google named my website only three times. Being cited often by Bing does not automatically make you a source on Google.

For me this means: in my measurement a spot on the first results page was neither necessary nor sufficient. This is an observation on one website on one day. It is not general evidence.

Rankings and clicks capture only part of your visibility. For the rest, the search engines themselves have offered reports since this year.

Bing is furthest ahead. Since February 10, 2026 the AI Performance report in Bing Webmaster Tools shows how often your content appears as a source in Copilot, in AI-generated summaries in Bing and in select partner integrations. It adds the cited URLs and a sample of the phrases. For my website Bing counted exactly 18,604 citations from January 9 to September 30, 2026.

Google announced its own generative AI reports in Search Console on June 3, 2026. According to Google they have been rolled out to all websites worldwide since August 31, 2026. They show impressions in AI Overviews, in AI Mode and in the AI features of Discover. You can break them down by pages, countries, devices and dates. Google does not mention clicks in the announcement.

Whether citations also turn into visits you can see in GA4. In its Analytics help Google itself describes a dedicated “AI assistants” channel for referrals from ChatGPT, Gemini, Microsoft Copilot, Claude and Perplexity. Create it in a custom channel group and move it above “Referrals” in the order.

For ChatGPT, Perplexity and AI Mode your own queries remain. Write down ten questions your customers really ask. Ask them again every few weeks and note who gets named. A single query is not a result.

What a citation is worth can only be roughly estimated. Seer Interactive measured a 35 percent higher organic click-through rate for brands cited in the AI Overview, 0.70 versus 0.52 percent in the third quarter of 2025. Seer writes itself that it cannot prove whether the citation is the cause. Strong brands might just as well simply get cited more often.

The plan combines Google’s recommendations for generative search with what I see on my own website.

Step 1: Check what your content offers that is your own

In its guide to generative AI search, Google writes that unique and useful content will likely influence a website’s presence there more in the long run than any other suggestion. As an example Google names a first-hand review based on personal experience. A summary of existing content, by contrast, simply restates what is already available elsewhere.

Go through your most important pages with these questions:

• Does the page contain your own data, your own tests or your own cases?
• Does it describe an experience only you can provide?
• Does the first sentence of a section answer the question it is about?
• Is it clear who wrote the text and why this person can judge it?

Pages that only summarize go to the top of the list. They compete directly with the AI answer.

Step 2: Cover the topic completely

Because Google runs further searches on subtopics for an answer, the whole topic counts. A hub-and-spoke model arranges the subtopics around a central page and connects them with internal links.

1. Define the core topic: What is it about at its core, for example “AI browsers”?
2. Collect attributes: What belongs to it, such as how it works, providers and privacy?
3. Capture related topics: What is directly connected, such as AI Overviews, ChatGPT or zero-click search?
4. Close gaps: Which of these questions does your website not answer yet?

Google draws one clear line. Creating a separate page for every possible fan-out query violates Google’s scaled content abuse policy if it is done primarily to manipulate. Write the pages your readers are missing.

Step 3: Secure the technical basis

Technically, what counts first is that AI systems may use your page at all. Check with the URL Inspection tool in Search Console whether the page is indexed. Look for nosnippet, data-nosnippet or max-snippet. According to Google these controls limit what is shown from your page in Search.

For ChatGPT and Perplexity the robots.txt decides. OpenAI writes that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, at most as navigational links. Perplexity recommends allowing PerplexityBot. Bing recommends IndexNow so that its AI knows the current version of a page.

Structured data remains useful because many rich results require it. For AI features, however, Google does not require any special markup. Ahrefs’ measurement of 1,885 pages found no increase in citations. What matters is that your markup matches the visible text. I corrected the same mistake about structured data and AI Overviews in the post on that topic on October 6, 2026. You do not need an llms.txt for Google, according to Google its Search ignores such files.

Step 4: Prove trust

According to Google, trust is the most important part of E-E-A-T. You can only prove it with things a reader can check.

• Author: a page with qualifications, photo and links to other publications
• Own material: your own screenshots, measurements and photos instead of stock images
• Sources: links to primary sources such as studies, documentation and authorities
• Transparency: legal notice, date of the last change and open corrections

I take the last point seriously myself. In my view an error that stays costs more trust than an open correction note like the one at the end of this post.

Step 5: Build channels beyond Google

The more answers end without a click, the riskier a single channel becomes. My own measurement shows where it pays to look.

• YouTube: In my sample YouTube was the most-cited other domain in AI Overviews. A short video on a core question can be a second source there.
• Reddit: Reddit had ten mentions in my sample. According to Sistrix, reddit.com is currently the fourth most visible domain in Google Search in the US. Answer where your audience really asks instead of scattering links.
• Bing and Copilot: For my website Bing counted 18,604 citations in nine months. A regular look at Bing Webmaster Tools is worth it.
• Email: A mailing list belongs to you. No search engine update takes it away from you.

In my view this development rewards websites that have something of their own. An AI can summarize what is on the web. It cannot make up for a measurement nobody has done.

My own numbers fit, though not the way I would have expected. The page Bing cites most often explains the abbreviation tl;dr. Google names it too, for four spellings of the query. For my basic articles on classic SEO terms, on the other hand, I do not get named. In my view they offer nothing there that other explanations do not offer as well.

Ranking is still the ticket in. Google picks from its index. If you are missing there, you will not make it into any answer. In my view the only thing that has lost is the formula “more content, more links, more rankings”. The better question for every briefing is: what is here that nobody else has?

In December 2025 I described a new discipline with its own levers here. Today I see it more soberly. The fundamentals stay the same. Google says so too. What is new is measurement. For the first time Bing and Google show you where your pages appear in AI answers.

This is how I see the shift:

• from keyword density to topics that are answered completely
• from summarized content to your own data and your own experience
• from ranking alone to ranking and citation
• from one channel to several channels

Take one of your most important articles and go through it with the questions from step 1. Then check in Bing and in Search Console whether it already gets cited.