Ask ChatGPT a question and it might name your company by name, right there in the answer, with no link attached. Ask Perplexity the same question and you often get a numbered citation you can click straight through. Ask Gemini and the behavior shifts again.
For a marketer trained on classic SEO, this is disorienting. The entire discipline was built around one assumption: if you rank, you get a click. If AI answers can satisfy the user without ever sending them anywhere, what exactly are you optimizing for?
We went looking for real data on how often AI answers actually link back to their sources, rather than just naming a brand in prose. The honest answer is that the research is partial and platform-specific. But what exists is enough to draw a clear picture, and it is not the picture most marketing teams are planning around.
Traditional search was built on a simple contract. You earned a ranking, the user saw a blue link, and a click carried them to your site. Analytics, attribution, and most performance marketing budgets still assume that chain holds.
We wrote about the mechanics of that shift in the death of blue links, and the short version is that the chain is breaking in the middle. The ranking still happens, in a sense, when a model decides which sources to draw on. But the link to the user is now optional, not guaranteed, and increasingly rare.
That matters because a citation without a link is still a form of visibility. Your brand can be the reason someone chooses a product, a vendor, or a tool, without a single session ever showing up in your web analytics. That is the core tension this post is trying to untangle with actual data instead of assumption.
It is worth remembering that the click was already disappearing before ChatGPT existed. SparkToro's analysis of Datos panel data found that in 2024, roughly 58.5% of Google searches in the US and 59.7% in the EU ended without a single click to any website, including Google's own properties. Out of every 1,000 US searches, only about 360 sent a visitor to the open web.
You can review the methodology in SparkToro's 2024 zero-click search study. Featured snippets, knowledge panels, and instant answers had already trained a large share of searchers to get what they needed without leaving the results page, years before generative answers became mainstream.
That context matters. AI chat products did not invent the zero-click habit, they inherited it and extended it into a new surface. The reasonable expectation, based on that trend line, is that AI answers make the click even less central to how information moves, not that they broke something that used to work reliably.
The most rigorous public dataset we found comes from Ahrefs, which analyzed 15,000 long-tail queries across ChatGPT, Gemini, Copilot, and Perplexity and compared the citations each platform produced against the same query's Google rankings.
A few findings stand out. On average, only about 12% of the links cited by ChatGPT, Gemini, and Copilot also rank in Google's top 10 for the same prompt, meaning roughly 80% of what these models cite would never surface through classic organic search at all. Perplexity behaves differently: close to 29% of its citations overlap with top 10 rankings, consistent with its design as a retrieval-first product that leans on live web results rather than a trained model's internal recall.
Read the full breakdown in Ahrefs' AI search citation overlap study. The practical takeaway is not just about overlap with rankings. It is that citation behavior varies enormously by platform, and any post that quotes a single "AI answers cite sources X% of the time" number is almost certainly flattening real, meaningful differences between how ChatGPT, Perplexity, Gemini, and Copilot are built.
The variation is not an accident. Perplexity's product is a retrieval engine first, so showing a clickable source is close to the whole point. ChatGPT's default conversational mode is built on trained knowledge and only pulls in live citations when a query triggers web browsing. Claude's citation behavior depends heavily on whether retrieval or a connected source is part of the request at all.
That means the question "does AI link back to sources" does not have one answer. It has at least four answers, one per major platform, and they are not converging toward a shared standard.
Citation behavior is only half the story. The other half is what humans actually do when a link is put in front of them inside an AI-generated answer.
Here the clearest large-sample data comes from outside the chatbot category entirely: Pew Research Center's 2025 analysis of Google Search behavior when an AI summary appears in the results. Using browsing data from a panel of desktop and mobile users during March 2025, Pew found that users clicked through to a traditional web result in just 8% of searches that displayed an AI summary, compared with 15% of searches without one. Clicking a link inside the AI summary itself was rarer still, happening in only about 1% of all visits to a search results page with a summary present.
You can read the full methodology and findings in Pew's report on AI summaries and click behavior. It is worth being precise about scope here: this is Google Search's AI Overviews, not a standalone chatbot session in ChatGPT or Claude. But it is the most methodologically solid public data we could find on what happens after a link is shown, and directionally it lines up with what most GEO practitioners already suspect. Even when a source link exists, most users never act on it.
That is the uncomfortable core of this whole topic. It is not just that AI answers sometimes skip the link. It is that even when the link is there, the click-through habit itself appears to be weakening.
There is a second, subtler problem sitting underneath link-back rates: the gap between being named and being cited.
Kevin Indig's Growth Memo newsletter ran an analysis of this exact gap, examining citations for nearly 4,000 domains across 115 prompts in 14 countries, spread across four AI search engines. The framing is useful: a brand can be mentioned by name in an AI answer's prose without ever being formally cited as a source, and a brand can be cited as a source URL without its name ever appearing in the visible text. These are two different events, tracked differently, and conflating them is one of the more common mistakes we see marketing teams make when they try to gauge their AI visibility.
Read the original analysis in Growth Memo's piece on the ghost citation problem. The practical implication is that "were we cited" is the wrong first question. The better first question is "were we mentioned at all," because a named-but-unlinked mention is often the more common outcome, and it is invisible to referral traffic reports by design.
Everything above points at the same structural problem. Classic SEO measurement assumes visibility converts to a session, a session gets attributed, and attribution justifies the spend. Take away the guaranteed link and the whole chain has a broken link in it, literally.
This is exactly why we think share of voice, not click volume, is the right north star for AI visibility. We laid out the mechanics of that shift in how to track share of voice in AI answers, and the short version is that you have to measure presence and framing inside the answer itself, not just what leaks through to your web analytics afterward.
A brand that appears in a large share of relevant AI answers but is rarely linked is still building preference with every buyer who reads that answer. A competitor obsessing over their AI-driven click-through rate, while ignoring how often they are simply never mentioned, is optimizing for the wrong layer of the funnel entirely.
This is not an argument that clicks stop mattering. Paid conversion, demo requests, and signups still need a session to happen somewhere. It is an argument that click volume can no longer be the leading indicator for whether your GEO work is succeeding, because too much of the value now accrues before, or instead of, a click.
Given the research above, a few adjustments seem warranted for teams that are used to classic SEO reporting.
This is the gap Bold GEO exists to close. Rather than waiting for referral traffic that may never arrive, Bold GEO checks how your brand is actually described, mentioned, and cited across ChatGPT, Perplexity, Gemini, Claude, and Copilot on a daily refresh, so you can see the mention before you ever see (or fail to see) the click.
We want to be direct about the limits of what is publicly known here. There is no single, universally agreed "AI answers link back X% of the time" statistic, and anyone who hands you one precise cross-platform number should be treated with some skepticism. The Ahrefs study measures citation overlap with search rankings, not raw link-back frequency. Pew's data covers Google's AI Overviews specifically, not standalone chatbot sessions. Growth Memo's analysis focuses on the mention-citation gap across a defined, real but bounded sample. Put together, these data points do not give us one tidy percentage. What they give us is a consistent direction: links inside AI answers are inconsistent by platform, citations that do appear rarely get clicked, and being named is a separate, more common event than being formally cited. That direction is enough to act on, even without a single magic number.
If a future study lands on a clean, universal figure, we will update this post to reflect it. Until then, the more useful move for a marketing team is to stop waiting for a definitive answer and start measuring your own brand's mention and citation rate directly, platform by platform, the same way you would track rankings before a single industry-wide SEO benchmark existed.
Bold GEO monitors how your brand is cited across ChatGPT, Perplexity, Gemini, Claude, and Copilot on a daily refresh. 7-day free trial, no credit card.