Why AI Search Is Leaking Your Traffic — and How to Spot It
AI search isn't stealing your traffic dramatically; it's leaking it quietly. ChatGPT processes 2.5B prompts a day and sends 190× less traffic than Google. Here's how to spot the leak.
By Julian Hernandez ·
The short answer
AI search is leaking your traffic in 2026, not stealing it dramatically. ChatGPT processes about 2.5 billion prompts per day but sends roughly 190× less traffic to websites than Google does, according to recent Ahrefs research. That means every time a user asks ChatGPT the question they used to ask Google, the answer happens inside the chat window and your page never gets the visit. The leak shows up in your analytics as flat or declining clicks against rising impressions in Search Console, a small but growing trickle of referrals from chatgpt.com or perplexity.ai, and a quiet drop in branded search volume. Spotting it early is half the battle.
What does "leaking" actually mean here?
The traffic shift from traditional Google to AI engines is rarely a cliff. It's a slow leak that most analytics dashboards are not set up to detect.
The old behavior: a user types "what is GEO" into Google, lands on the top-ranking explainer post, reads for 90 seconds, and possibly clicks a related link. That visit is counted as one session in your analytics.
The new behavior: the same user asks ChatGPT "what is GEO." ChatGPT generates a 200-word answer naming three brands and possibly citing two sources. The user reads the answer, decides whether to dig deeper, and only sometimes clicks through. Most of the time, the question is answered inside the chat and the visit never happens.
You still own the same content, you may still rank #1 in Google for the query, but the actual user who needed the answer never reached your page. That is what we mean by leaking. The intent existed; the click did not.
The leak is uneven by query type. Definitional and informational queries leak the fastest because AI engines summarize them well. Comparison and decision queries leak more slowly because users often want to verify before deciding. Transactional queries barely leak at all because users still need to land on the actual product page.
How big is the leak in 2026?
The headline numbers, as of mid-2026, are striking enough to set the stakes.
ChatGPT now processes roughly 2.5 billion prompts per day, compared to Google's estimated 13.7 billion daily search queries. That puts ChatGPT at about 18% of Google's prompt volume on its own, a number that quietly puts ChatGPT ahead of Bing as a destination for search-like activity, per the Ahrefs analysis cited above.
But the per-query click-through rate tells a very different story. ChatGPT's click-through rate to external websites is roughly 96% lower than Google's. The aggregate effect: Google still sends about 190× more traffic to websites than ChatGPT does, despite ChatGPT's growing query volume. Add Perplexity, Gemini, Claude, and Google AI Overviews to the picture and the gap narrows somewhat, but the fundamental pattern holds — AI engines retain a much larger share of intent inside the answer, and send a much smaller fraction of that intent on to a click.
For an individual brand, this means a single number is the wrong way to think about the leak. Total website traffic could be flat or growing while the share of your category's intent that actually reaches you is collapsing. The brands that catch this early measure intent at the engine, not just sessions in their own analytics.
Where does the traffic actually go when AI engines answer?
When a query is answered inside an AI engine, the user either acts on the answer immediately or clicks one of the cited sources. There is no third option. The cited sources get the residual click traffic; everyone else gets nothing.
Who tends to win the residual click? Three patterns hold across the engines LiftRank monitors.
First, the brands actually named inside the answer. ChatGPT, Gemini, and Google AI Overviews often mention 2–5 brands per answer. Brands that get named pull a disproportionate share of clicks even when the engine does not produce a clickable citation — users open a new tab and search the brand name directly. This converts as branded search volume in Google, which is one of the cleanest signals that AI engines are working in your favor.
Second, the brands the engine explicitly cites. Perplexity always cites; Google AI Overviews usually does; ChatGPT sometimes does. When the engine produces a clickable citation, the cited source captures the visit. This shows up in analytics as referral traffic from chatgpt.com, perplexity.ai, or gemini.google.com.
Third, the brands whose content appears in third-party reviews and comparisons that the engine then cites. Roughly 85% of brand mentions in AI engine answers originate from third-party pages, not the brand's own domain. So the click that comes from an AI answer often goes to G2, Reddit, or a comparison post on a popular industry blog — and from there, secondarily, to the brand's own site.
The brands that do not show up in any of these three pathways get the residual of the residual: zero.
How do you spot the leak in your own data?
Four signals, looked at together, tell you whether you have a leak and how big it is.
Signal 1: Search Console impressions vs. clicks divergence. The clearest leading indicator. Pull your last 12 months in Google Search Console. If impressions are up materially (say, +20%) while clicks are flat or down, AI Overviews are appearing on your queries and answering the question for the user. The wider the gap, the bigger the leak. This signal has a name in the industry — sometimes called the "Crocodile Mouth Effect" — because the impressions line and the clicks line look like an opening jaw.
Signal 2: Branded search volume changes. Pull your branded search volume (the number of monthly searches for your brand name) from Google Search Console or a rank tracker. Steady or growing branded search alongside flat organic clicks suggests AI engines are mentioning your brand by name in answers; users then open a new tab and search you directly. This is the leak working in your favor. Flat branded search alongside flat organic clicks is the leak working against you.
Signal 3: AI referral traffic in GA4 or your analytics tool. Filter referrer traffic by source: chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai. Most brands now see a trickle of traffic from at least two of these. If you see traffic from one but not the others, that's a single-engine visibility profile and you're missing the rest. If you see traffic from none, you are invisible to all of them.
Signal 4: across the 11 engines. This is the only signal that measures the actual leak rather than its downstream effects. A LiftRank scan against your decision-intent prompt set tells you exactly how often you appear in AI answers across ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews, Copilot, Grok, DeepSeek, Mistral, Meta AI, and Qwen. If mention rate is low and dropping, your leak is widening even if the first three signals haven't fully caught up yet.
The four signals reinforce each other. Two or more pointing the same way is a confident read on whether you have a leak.
What should you do once you see it?
The leak is structural. You cannot reverse the shift from blue links to AI answers. What you can do is move into the answer.
Three moves, in order of leverage.
Move 1 — be one of the brands the engines name. This is the single highest-leverage action because it captures the residual click traffic and the branded-search backflow at once. Restructure your top 10 commercial pages for answer-first extraction (self-contained 150-word answers, FAQPage schema, question-form H2s) and audit your third-party citation graph for missing coverage on Reddit, G2, and the comparison sites your category uses.
Move 2 — earn the explicit citation, not just the mention. Once you are being mentioned consistently, the next move is to be the cited link. Perplexity and Google AI Overviews are the easiest engines to earn citations on because they cite aggressively; ChatGPT is harder. The mechanic is the same: deeply structured content, FAQPage schema, and authoritative third-party signals all push citation probability up.
Move 3 — instrument the measurement so you can tell the moves are working. This is where most teams skip a step. Without a baseline mention rate per engine and a weekly cadence, you cannot tell whether your content changes moved anything. A LiftRank Free plan gives you a monthly baseline across 3 engines with no credit card; Starter ($29/mo) covers all 11 engines weekly with 50 prompts.
The leak does not stop. The question is whether you're capturing the answer or being skipped over inside it.