The Short Answer
AI leakage is the revenue you lose when buyers research with AI, your brand does not come up, and they buy from whoever the machine named instead. It never appears as a line item in your analytics, because the buyer who checked you with AI and walked away never reached your site. What you see instead are seven indirect symptoms: rising blended CAC, unexplained growth in direct traffic, flat branded search in a growing category, falling conversion on unchanged pages, competitors named in AI answers, prospects arriving with objections you never gave them, and intact rankings that no longer produce traffic. Any three of these together mean it is already happening to you.
The hardest problems to fix are the ones that do not generate an alert. AI leakage is one of those. There is no dashboard row labeled “customers lost to ChatGPT,” no drop you can point at in a weekly report, no campaign to pause. The loss happens before anything you own gets touched.
That is not a small blind spot anymore. 58% of consumers now use AI to research major purchases, up from 41% a year earlier (Invoca, B2C Buyer Experience Report, 2026). Every one of those research sessions is a moment where your brand is either present or absent, and you will never see the ones where you were absent.
So you diagnose it the way a doctor diagnoses something that does not show on the surface: by reading the symptoms. Here are the seven that matter, in the order they usually surface, and how to check each one.
1. Blended CAC is climbing while your campaigns look healthy
This is the first thing most operators notice, and the easiest to misdiagnose. Channel-level metrics look normal. CPMs are stable, click costs are in range, your media buyer has a reasonable explanation for every account. But blended CAC keeps creeping up, and the math at the bottom keeps getting worse.
Rising acquisition cost has plenty of legitimate causes, and CAC has been climbing industry-wide for years, up roughly 60% over five years across an analysis of around 700 subscription companies (Paddle). Scale itself pushes CPA up, which we covered in our piece on why CAC rises as you scale spend. So this signal alone proves nothing.
What makes it an AI leakage symptom is the pattern: the cost increase is not concentrated in one channel or one campaign. It is spread evenly, which is what happens when the loss is not occurring inside any channel at all. You are paying the same to generate attention and converting less of it, because a share of the people you paid to reach are now stopping to verify you with an AI before they act.
How to check it: pull blended CAC by month for 18 months against media spend and channel-level CPA. If channel CPAs are flat but blended CAC is up, the leak is happening between the click and the conversion, somewhere you are not measuring.
2. Your direct traffic is growing and nobody can explain why
This is the most overlooked signal on the list, and the most technically useful, because it is the one place where AI leakage leaves a real fingerprint in your own analytics.
When someone clicks a link inside ChatGPT, Perplexity, or Gemini, the referrer header that would normally tell your analytics where they came from often does not survive the trip. Many AI applications, particularly mobile apps, strip that header, and the visit lands in your “Direct” bucket instead (Matomo, 2026). Users also copy and paste links out of AI answers rather than clicking them, which strips the same data.
So a growing slice of your AI-referred traffic is already in your reports, mislabeled as people who supposedly typed your URL from memory. That is worth finding, because these are good visitors. AI referral traffic converts at 14.2% on average, roughly five times traditional organic search (Opollo, 2026 AI Search Benchmark Report). They arrive already decided.
The leakage implication cuts the other way. If AI is sending you meaningful hidden traffic, AI is also being asked about your category constantly. Every one of those conversations you are not part of is a session where a competitor got named.
How to check it: segment direct traffic by landing page. Real direct traffic lands on your homepage. Someone arriving from an AI answer lands deep, on a specific product or article page they could not have guessed the URL for. New users landing directly on deep pages are almost certainly AI referrals in disguise.
3. Branded search is flat while your category is growing
Category demand is up. Competitors are hiring, spending, and expanding. Every market indicator says more people are shopping for what you sell. And your branded search volume has not moved.
Branded search is the cleanest proxy that exists for whether people are hearing about you. When category demand rises and your branded volume does not, someone else is being recommended into that growth. In a market where the recommendation increasingly comes from a machine, that is what losing the AI conversation looks like from the outside.
How to check it: compare your branded query volume in Search Console against category-level demand trends over the same 12 months. Flat branded volume in a rising category is a share-of-voice problem, and share of voice now includes the machines.
4. Conversion is falling on pages you have not changed
Same landing page. Same offer. Same creative. Same traffic sources. Lower conversion rate.
When nothing on your side changed, something on the buyer’s side did. The most common version of this in 2026 is a buyer who lands on your page, gets interested enough to want a second opinion, opens an AI assistant in another tab, asks whether your brand is any good, and does not come back. Nothing about your page failed. The verification step failed, and it happened somewhere you have no visibility.
The trust environment makes this worse. 71% of consumers trust companies less than they did a year ago (Salesforce, The Top Marketing Statistics to Know in 2026). More skepticism means more verification, and more verification means more chances for an AI to answer the question about you badly, or not answer it at all.
How to check it: find a page with stable traffic sources and unchanged creative, and chart its conversion rate over 12 to 18 months. A steady decline with no corresponding change on your end points to something happening off-page, mid-funnel.
5. AI engines name your competitors on your category’s defining prompts
This is the most direct signal available, and the only one you can check in the next five minutes.
Ask ChatGPT, Gemini, and Perplexity the question a buyer in your category would actually ask. Best options. Most trusted providers. Who to use for a specific need you serve. Then read who gets named.
The bar here is lower than most brands assume. On average, only 16.3% of AI responses mention any given brand on discovery prompts, while leading brands in a category reach as high as 56.5% (AthenaHQ, State of AI Search 2026). That gap between average and leader is the entire opportunity. Somebody in your category is going to occupy the top of that range.
Watch for the quieter version of this failure too: the AI does mention you, but with stale or wrong information. An old complaint, a discontinued product, outdated pricing, a former executive. A human researcher would contextualize that. A machine may hand it back flatly as current fact.
How to check it: run four or five real category prompts across all three engines, in a fresh session with no personalization, and write down who gets named and what is said about you. If competitors appear consistently and you do not, that is the leak, confirmed.
6. Prospects arrive with objections you never gave them
This one comes from your sales floor, not your dashboard, which is why it usually gets noticed and then dismissed as anecdote.
A prospect opens with a concern about your pricing model that is not on your site. They reference a limitation you resolved two years ago. They compare you to a competitor you have never positioned against. Ask where they heard it and the answer, increasingly, is that an AI told them.
This matters more than it looks. The prospects raising these objections are the ones who still showed up. They are the survivors of a filter you cannot see. For every one who arrives with a machine-supplied objection and gives you the chance to answer it, some larger number got the same answer and never made contact at all.
How to check it: ask your sales team to log, for one month, every objection that did not originate from your own marketing. Then run those exact claims as prompts against the AI engines and see whether they come back. Objections that reproduce are not anecdotes. They are a systematic problem with what the machines believe about you.
7. Your rankings held and your traffic fell anyway
You still rank where you used to rank. Traffic is down regardless.
Ranking and traffic have come apart, because a search result page increasingly answers the question itself. In the first four months of 2026, 68.01% of US Google searches ended without a click, up from 60.45% in 2024 (SparkToro, 2026). Rand Fishkin’s analysis puts AI Overviews among the likely drivers.
Research on the click behavior behind that shift found users clicked a traditional result in 8% of visits when an AI summary was present, versus 15% when it was not, and clicked a link inside the AI summary itself in just 1% of visits (Pew Research Center, from browsing data collected in March 2025, so treat it as directional for where 2026 sits rather than current measurement).
The strategic point holds either way. Position one on a page nobody clicks is not a win. The job has shifted from ranking in a list to being the source the answer is built from.
How many of these do you need before it is real?
Three. Any single one of these has innocent explanations, and treating one symptom as proof will send you chasing the wrong fix. Three or more at the same time is a pattern, and the pattern is consistent with exactly one cause: buyers are checking you with machines, and the machines are not making your case.
The reason this compounds rather than plateaus is that AI recommendation is self-reinforcing. Every time a model names a competitor and that answer goes unchallenged, the association gets a little stronger. Meanwhile the audience of machines keeps growing. Bots passed humans in web traffic for the first time in the internet’s history in June 2026, at 57.5% of requests versus 42.5% human (Cloudflare Radar, via Forbes, 2026), which we wrote about in our piece on marketing to bots.
What actually closes the gap?
Three things working together, in this order.
AI visibility earns the citation, so the machines name you when a buyer asks. That means answering the real questions buyers ask in a structure a machine can extract, keeping your entity information consistent everywhere you appear, and cleaning up the stale material a model might surface as fact.
For the full mechanics behind how AI engines decide who to cite, and the five levers that move it most, see our complete guide to AI visibility.
Authority distribution builds the third-party corroboration that makes the citation hold. Models weight sources they already trust, so being mentioned and confirmed outside your own website is what turns your claim into something a machine will repeat.
Paid media scales it, the way it always has. This is the engine, and it works harder once the first two are in place, because a buyer who pauses to verify you now gets an answer that supports the ad they just saw instead of undercutting it. The full argument for why this window is closing is in our piece on the AI extinction event.
If three or more of these signs are true for you, the leak already has a number attached to it. Run your revenue through our estimator to find out what AI leakage is costing you, then see if you qualify to have us plug the leak.
Frequently Asked Questions
What is AI leakage?
AI leakage is the revenue you lose when a buyer researches with an AI assistant like ChatGPT, Gemini, or Perplexity, your brand does not come up or comes up badly, and they buy from a competitor the machine named instead. It is invisible in analytics because the buyer never reached your site, so there is no session, no bounce, and no lost-conversion event to find.
How do I know if AI leakage is affecting my business specifically?
Check the seven signs above and count how many are true at once. One has innocent explanations. Three or more at the same time is a pattern. The fastest single check is the fifth sign: ask ChatGPT, Gemini, and Perplexity who the best options in your category are, and see whether you get named alongside your competitors.
Why does my direct traffic go up when AI sends me visitors?
Because many AI applications strip the referrer header that tells your analytics where a visitor came from, particularly on mobile, so those sessions land in the Direct bucket instead of a labeled AI channel. Copying and pasting a link out of an AI answer does the same thing. To find it, segment direct traffic by landing page: genuine direct traffic arrives on your homepage, while AI referrals land deep on specific pages nobody would type from memory.
Is this just SEO with extra steps?
No. SEO wins a position in a ranked list that a human then chooses from, so the human is still the decision-maker. AI answers collapse that into a single recommendation, and being ranked well no longer guarantees a click: 68.01% of US Google searches ended without any click in early 2026 (SparkToro). The job changes from ranking in the list to being the source the answer is assembled from, which is what AEO, Answer Engine Optimization, is built for.
Should I fix AI visibility before I keep spending on paid media?
No, and treating it as a choice is the most expensive mistake available here. Paid media is the engine that buys attention at scale and it keeps working. AI visibility makes that spend convert better, because the buyer who pauses to verify you gets an answer that supports your ad instead of contradicting it. Run them together.
How long does it take to close an AI visibility gap?
Organic AI authority compounds over months, because models need repeated, corroborated signals before treating a brand as a default answer. Paid placement inside AI platforms is available now and does not require that wait, which is why the two work best as a pair: buy presence in the conversation this quarter while the authority work builds underneath it.
Drew Kossoff is the Founder and CEO of Rainmaker Ad Ventures, a performance marketing agency with $300M+ in managed ad spend and more than $1B in client revenue across finance, health, DTC, publishing, and lead generation. Connect with Drew on LinkedIn.
Sources cited: Invoca, B2C Buyer Experience Report, 2026. Paddle, How Is CAC Changing Over Time. Matomo, AI Chatbot Traffic Guide for Web Analytics, July 2026. Opollo, 2026 AI Search Benchmark Report. Salesforce, The Top Marketing Statistics to Know in 2026. AthenaHQ, State of AI Search 2026. SparkToro, In 2026 Less than One Third of Google Searches Still Send a Click. Pew Research Center, July 2025 (March 2025 browsing data). Cloudflare Radar, via Forbes, June 2026.