AI Search Is Splitting Your Audience Across Engines

Your rankings look fine. The keywords you have owned for years still sit near the top of the page. And yet the clicks that used to follow those rankings keep thinning out, month after month, with no drop in position to explain it. If you run a publishing business, you have probably stopped trusting the rankings report, because it says one thing while your traffic says another.

The question underneath that gap is not "is AI search replacing Google." Your readers have not left. They have scattered. The same person who used to run three Google searches now asks one question inside an AI answer, checks a second answer somewhere else, and never lands on a results page at all for the third. Two large studies published in the last few weeks put numbers on that scattering, and the numbers are worse and stranger than the usual "clicks are down" headline. This is where publisher traffic actually went in 2026, and it is what you do to get found across the places it went.

The clicks did not just shrink, they scattered

Start with the size of the problem, because it is now measured rather than guessed. Ahrefs studied 963 French domains using Google Search Console data from April to July 2026, France being an early rollout market for AI Overviews. Across all of those domains the median click-through rate fell 5.7 percent and the median site lost 8.8 percent of its clicks. That alone is a rough quarter to a third of a year of erosion. For the sites most exposed to AI Overviews, the ones where an AI answer appeared on more than twenty percent of their queries, the click-through rate dropped 23.1 percent, and 82.2 percent of those heavily exposed domains lost clicks. The health category, which triggers AI answers constantly, saw its click-through rate fall 20.4 percent. Ahrefs reduced the whole pattern to one blunt rule of thumb: every extra point of AI Overview exposure costs you roughly one point of click-through rate.

The second study is the one that should change how you plan, because it is not a correlation pulled from log files. Researchers at the University of Pennsylvania and Northeastern University ran a randomized controlled trial with 1,100 people who finished the experiment, splitting them into three groups for seven days: normal Google, Google with AI features removed, and Google AI Mode. A randomized trial is the difference between "sites with more AI answers happen to have fewer clicks" and "the AI answer caused the clicks to fall." The causal reading held. Assignment to AI Mode cut click-through to outside websites by 18.8 percentage points. News sites took a 12.5 point hit, Reddit lost 21.2 points, and even Wikipedia dropped 9.9 points. Turning AI features off did the opposite and lifted click-through by 8.8 points, which tells you the AI layer, not search itself, is the thing draining the clicks.

Here is the part almost nobody is talking about, and it is the reason this article exists. AI Mode did not just keep people on Google reading the answer. It pushed a measurable share of them off Google entirely. The trial found that assignment to AI Mode raised the fraction of users searching on a competitor engine, Bing, DuckDuckGo, or Yahoo, by 11.2 percentage points. Trust in the results fell by 0.34 points on a seven point scale, and satisfaction and usefulness both dropped by more than half a standard deviation. When people were forced into AI Mode and disliked it, some of them voted with their feet and tried a different search engine. Google's own AI is not only compressing the clicks on Google, it is handing a slice of your potential readers to the engines next door.

Why optimizing for one engine is now the risky bet

For twenty years the entire discipline of getting found had a single center of gravity. You ranked on Google, or you did not exist. That was a concentrated bet, but it was a rational one, because Google was where the demand sat and the demand mostly stayed put.

That center is coming apart in two directions at once. The first split is on Google itself, between the ten blue links, the AI Overview, and the full AI Mode conversation, each of which sends readers to your site at a different and lower rate. The second split is across engines, as some of the readers who dislike the AI experience move to Bing, DuckDuckGo, Yahoo, or straight into a standalone assistant like ChatGPT, Gemini, or Perplexity. Your audience is now distributed across a set of surfaces that did not meaningfully compete for it a year ago, and each surface has its own way of deciding who to cite.

A number one ranking on one engine is not a strategy anymore. It is a single point of failure that happens to be performing well today.

This is the exact reason Publisher in a Box treats diversification for stability as the identity of a publishing business rather than a nice-to-have. A publishing business is not one Facebook page and it is not one Google ranking. It is an operating position across Facebook, Google Discover, content syndication, AI search, and eventually the sale of the asset, so that no single algorithm change or engine decision can take the whole thing down at once. The fragmentation of AI search did not create that principle. It just made the cost of ignoring it show up on the traffic report faster than it used to.

What "getting found" means when the search box multiplies

If demand is spreading across engines, then the thing you are optimizing for has to change from a rank on one page to a presence across many answers. That presence is what we call AI Citation Presence, the degree to which the engines quote your brand and your pages when they build an answer in your category. It is the AI-era version of showing up, and it does not move in lockstep with your old rankings.

You measure it with AI Share of Voice, the percentage of AI answers in your category that cite you rather than a competitor. This is the number that actually matters when the click is disappearing, because being the source an engine leans on is what carries your brand into a reader's head even on the queries that never produce a click. Publisher in a Box measures its own AI Share of Voice the same way, and the honest read is that we are still building it. Our last self-audit put our Share of Voice near the floor on unprompted category queries, with Gemini the worst of the engines, confusing the brand rather than citing it. That is exactly why the number has to be measured engine by engine rather than assumed from one, and why the work is a loop and not a one-time claim.

Two older ideas do the heavy lifting underneath the score. Entity Positioning is whether the engines understand what you are, as a defined thing they can name and cite, rather than a loose collection of pages. Topical Authority is whether they see you as a deep and consistent source on a subject rather than an occasional visitor to it. Engines assemble answers from sources they can identify and trust, so a publisher with clear Entity Positioning and real Topical Authority gets cited across many engines at once, while a publisher without it stays invisible on all of them at once. That is the difference between fragmentation working for you and against you.

The measurement loop is the actual work

The passive framing of the last decade said search visibility was mostly out of your hands, something the algorithm did to you. That framing was always half wrong, and in a fragmented landscape it is fully wrong, because almost everything moves once you read your own data and act on it. The real answer we give, and the thing our GEO work and Turnkey Management actually deliver, is a continuous loop of analysis and optimization, not a one-time setup you walk away from.

You cannot manage what you refuse to measure, and single-engine reports are exactly how publishers stay blind to where their readers went.

The loop runs like this. First you read what is already happening. Google Search Console now separates AI-surface performance from ordinary search, so you can see which of your pages are being pulled into AI answers, measured in impressions. That report shows impressions only, no clicks and no click-through rate, so you read it next to your ordinary Search performance to watch the gap open between the two. The Ahrefs study surfaced the pattern you will recognize once you look, sites gaining impressions while losing clicks, one example showing impressions up 183 percent against clicks down 1.9 percent. Impressions without clicks is the fingerprint of an answer being built from your page while the reader stays put.

Second you look across engines, not just Google, because that is where the fragmentation is hiding your wins and your gaps. You check how ChatGPT, Gemini, Perplexity, and Google's AI surfaces answer the real questions in your category, and you record who they cite. That record is your AI Share of Voice, and it tells you which engines already treat you as a source and which ones do not know you exist. Then you push more of what is already earning citations, deepen the topics where your authority is real, and fix the entity signals on the engines where you are absent.

The technical tease for the operators who want it

For anyone who wants to run this rather than read about it, the stack is not exotic. The signals that help engines identify and cite you are structured data on your pages, clean schema that states what your organization is and what each article covers, consistent naming across the web so your entity resolves to one thing, and internal linking that makes your topical depth legible. The measurement side is a set of scheduled checks: pull the AI performance rows from the Search Console API on a schedule, and run a fixed set of category questions through each assistant to log citations over time. You can wire the whole pull into an automation tool like n8n, run it as a Make scenario, or hit the APIs directly with scheduled jobs, and land the results in one sheet you actually look at each week. The point is not the tool. The point is that AI Share of Voice becomes a number you watch move, the same way you once watched rankings, so optimization has something to aim at.

Do not abandon Google, spread the bet

None of this is an argument to walk away from Google. Google still carries the largest share of search demand by a wide margin, and the readers who stay on the blue links are still real readers worth earning. The controlled trial even points the other way for the click itself, since removing AI features raised click-through, which means classic search is still the surface most likely to send someone to your site today. Cutting your Google work would be the wrong lesson.

The right lesson is parallel effort, not replacement. You keep earning the click where the click still lives, and at the same time you build the citation presence that carries your brand through the answers where the click has gone. And you do the same thing one layer up, across your whole revenue base, because search is only one of the surfaces that can be taken from you overnight. This is where the durable channels earn their place. On Facebook the two levers that move the most are Curation, the discipline of what you publish and how you shape it to the audience, and Virality, the reach that turns one strong post into a monetized event, and they feed each other when you distribute your best earning content to the pages that carry the revenue. Google Discover, content syndication through partners like MSN and Yahoo and Apple News, and an owned email audience are each a surface that keeps paying when one engine changes its mind. A publisher operating system is the whole set working together, so the fragmentation of one channel is an inconvenience rather than an emergency.

The publishers who will be fine in 2026 are not the ones who guessed which engine would win. They are the ones who stopped needing a single engine to win, because they measured where their readers actually were and showed up in more than one place.

Where to go from here

If your rankings look stable while your clicks keep sliding, the missing number is your AI Citation Presence, and you cannot fix what you have never measured. The GEO Authority System runs the LLM Visibility Evaluation across the major engines, scores your GEO Readiness, and hands you the GEO Authority Playbook that turns the gaps into a working optimization loop, so you can see your AI Share of Voice engine by engine and act on it. It is a one-time product at $499, and it is the fastest way to find out where you are already cited and where you are invisible.

Learn more and get your evaluation at the GEO Authority System.

Frequently asked questions

Is AI search actually sending my readers to other search engines?

In a randomized controlled trial from the University of Pennsylvania and Northeastern University, people assigned to Google's AI Mode were 11.2 percentage points more likely to search on a competitor engine like Bing, DuckDuckGo, or Yahoo than people on normal Google. Trust and satisfaction with AI Mode fell at the same time. So yes, at least some of the audience that dislikes the AI experience is moving, which is why optimizing for one engine now leaves demand on the table.

My rankings are stable but my clicks are falling. What is happening?

That gap is the signature of AI answers. Your page can still rank and still show impressions while the reader gets what they came for inside the AI answer and never clicks through. The Ahrefs study of French domains found cases of impressions rising while clicks fell, and measured a 23.1 percent click-through drop for the sites most exposed to AI Overviews. The rank is not lying, it just no longer guarantees the visit.

Does this mean I should stop optimizing for Google?

No. Google still holds the largest share of search demand, and the same controlled trial found that removing AI features raised click-through, so classic search is still the surface most likely to send you a visit today. The move is parallel effort, keep earning the click where it still lives while building the citation presence that carries your brand through the answers where the click has gone.

What is AI Share of Voice and how do I measure it?

AI Share of Voice is the percentage of AI answers in your category that cite you rather than a competitor. You measure it by running the real questions in your niche through each assistant, ChatGPT, Gemini, Perplexity, and Google's AI surfaces, and recording who gets cited over time. Tracking it engine by engine is the only way to see which engines already treat you as a source and which do not know you exist.

What can I actually control here?

More than the passive framing suggests. You control the structured data and schema that tell engines what you are, the consistency of your entity signals across the web, the depth of your Topical Authority on the subjects you own, and the measurement loop that tells you what is working. Reading your own Search Console AI performance data and your cross-engine citations, then pushing more of what already earns them, is the work that moves the number.

Key takeaways

• A randomized trial found Google AI Mode cut click-through to outside sites by 18.8 percentage points and pushed 11.2 percentage points more users toward competitor search engines, while trust and satisfaction fell.
• An Ahrefs study of 963 French domains measured a 23.1 percent click-through drop for the sites most exposed to AI Overviews, with roughly one point of click-through lost for every point of AI Overview exposure.
• The real change is fragmentation, not disappearance. Your readers are scattered across Google's AI surfaces and across other engines and assistants, each with its own way of deciding who to cite.
• Optimizing for a single engine is now a concentrated bet with a single point of failure. Diversification for stability is the identity of a publishing business, not an extra.
• The metric that matters is AI Citation Presence, tracked as AI Share of Voice engine by engine, built on clear Entity Positioning and real Topical Authority.
• Do not abandon Google. Run parallel effort, keep earning the click where it still lives, build citation presence where it has gone, and diversify the channels underneath so no one engine controls your business.

Sources

• Chen et al., University of Pennsylvania and Northeastern University, "AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence," August 2026. arXiv
• Search Engine Journal, "Research Shows Google AI Mode Sends Less Clicks and Is a Poor User Experience," September 2026. Article
• Ahrefs, "AI Overviews Impact in France," September 2026. Ahrefs Blog
• Inc., "Google's AI-First Overhaul Annoys Users So Much They Are Actually Switching Search Engines," 2026. Inc.
• Digiday, "Media Briefing: Overheard at the Digiday Publishing Summit, Google Zero edition," September 2026. Digiday