AI is shaping your buyers, and your analytics can’t see it
The structure and markup of this piece were refreshed for current answer engines; the original analysis is preserved as written.
Long before a buyer ever fills in a form, they have been forming an opinion of you — and in 2024 a growing part of that happens inside an AI chatbot, in a private conversation that leaves no click, no cookie, no referral. The channel is real, and your dashboard cannot see it at all.
the short answer
In 2024, LLMs have become private research advisors for B2B buyers — and a chatbot conversation produces no click, cookie or referral. The moment AI shapes a buyer is structurally invisible to analytics; when the buyer arrives later, it is mislabelled as direct or branded search. You cannot measure this channel by clicks. You measure it by proxy — self-reported attribution, brand search, direct-traffic behaviour — or you value it at a false zero.
key takeaways
- The “dark funnel” is not new — B2B buyers have always moved through private, untrackable channels — and in 2024 LLMs join it as private research advisors.
- A chatbot conversation leaves no click, cookie or referral: the moment AI shapes a buyer is structurally invisible, and in 2024 there is not even an AI tagging convention to rely on.
- When the buyer does arrive later, it is usually via brand or direct search, which analytics mislabels — crediting AI’s influence to something else, or to nothing.
- You cannot measure this channel with click precision; you measure it by proxy: self-reported attribution, brand search volume, and the quality of your direct traffic.
- The real risk is not imprecision but a false zero: lacking a tidy metric, teams value AI at nothing and underinvest in a channel already shaping decisions they cannot see.
where the influence happens vs where it shows up
The split is the whole point. The decisions that matter — comparing you, shortlisting you — happen in the left half, where no instrument reaches. What your analytics records is the right half: a branded or direct visit that arrives already convinced. Read the dashboard alone and you credit the visible step, missing that the work was done in the dark.
Why a false zero is the real danger
The instinct, faced with a channel you cannot track, is to set it aside until you can — to wait for the referral report, the clean attribution, the number that justifies the budget. The problem is that the number is not coming, and the waiting has a cost. Every quarter you treat AI influence as unmeasurable, you are in practice treating it as zero: nothing in the dashboard, nothing in the model, nothing in the case for investment. Meanwhile the channel keeps working, shaping which vendors buyers trust before they ever appear in your funnel. A false zero is more dangerous than a rough estimate, because it does not just misstate the value — it removes the channel from the conversation entirely, and you cannot manage what you have decided does not exist.
The alternative is not to invent precision you do not have; it is to accept directional truth and act on it. You will not be able to say AI drove exactly this many deals. You will be able to say, from self-reported sources and brand search and the shape of your direct traffic, whether the channel is growing and whether the buyers it touches convert better — and that is enough to decide whether to invest in being the answer AI gives. For a measurement discipline, the hard part is psychological: giving up the comfort of a single clean metric in exchange for a defensible read across several imperfect ones. But that trade is the honest one, and it is the one that keeps you from underfunding a channel precisely because it is the hardest to see.
The dark funnel, in three parts
Why AI conversations leave no trace, how the influence resurfaces mislabelled, and the proxies that let you measure it anyway. Open each layer for the part that changes how you track AI’s effect.
01 Why AI conversations leave no trace
A search click is a trackable event: the browser carries a referrer, the visit lands with a source attached, and your analytics can say where it came from. A conversation inside a chatbot is not that. It is a closed environment — no referrer header, no cookie set on your domain, no pixel firing — because the interaction was never designed to hand a website an attributable signal. When a buyer asks an AI to compare you against a competitor, weighs the answer, and decides you are worth a look, none of that produces a record you can see. In 2024 this is especially stark: there is not even a settled convention for tagging AI-driven visits, so the conversations that shape opinion happen entirely off your instruments. The influence is real and the data simply does not exist — not because your tools are weak, but because the channel does not emit the kind of signal those tools were built to catch.
02 How the influence resurfaces, mislabelled
The buyer persuaded inside an AI conversation does not vanish; they resurface — just not in a way that credits the AI. Convinced you are worth considering, they later search your brand name, or type your URL directly, or click a result they were already primed to choose. To your analytics, that looks like branded search or direct traffic, sources it will happily attribute to "SEO" or to nothing in particular. The actual moment of influence, the AI conversation that created the intent, is invisible, and the visible touch that follows takes all the credit. This is how a channel can be quietly powerful and apparently absent at the same time: the work happens in the dark, and a later, trackable step inherits the result. Read your dashboard literally and you will conclude the branded search did it, when the branded search was only the echo of a decision made somewhere you cannot look.
03 Proxies for an unmeasurable channel
If you cannot measure the channel directly, you measure its shadows, and several are reliable enough to act on. Self-reported attribution is the most direct: an open "how did you hear about us?" on your forms lets buyers name AI as a source the moment they start doing so, ahead of any tool. Brand search volume is the next signal — a rise in people searching your name specifically is often the downstream trace of being recommended in places you cannot see. The behaviour of your direct traffic is a third: growth in direct visits that land on deep, specific pages rather than the homepage fits the pattern of people sent by an AI to a particular answer. And win/loss interviews close the loop, letting buyers tell you in their own words what shaped the decision. None is precise alone; together they triangulate a defensible read. The craft is to accept directional truth from several proxies instead of demanding a certainty the channel will never give.
A starter loop you can run this quarter
None of the proxies is hard to set up, and you do not need all of them at once. The fastest place to begin is a single open question on your demo and contact forms: "how did you hear about us?", written as a blank field rather than a dropdown, because a fixed menu only captures sources you already thought to list, and the point is to catch the one you cannot see. Read those answers weekly. The first time a buyer writes "ChatGPT" or "I asked an AI," you have direct evidence — earlier than any dashboard would have given it. Tag the answers into rough buckets so the pattern becomes legible over a quarter rather than a pile of anecdotes.
Alongside that, watch two numbers you already have. Brand search volume — how many people search your name specifically — is the cheapest external signal, and a steady climb in it, especially without a campaign to explain it, is consistent with being recommended somewhere you cannot observe. And segment your direct traffic by landing page: a growing share arriving on deep, specific pages rather than the homepage fits people sent by an AI to a particular answer, not people typing your address from memory. Put those three together — self-reported sources, brand search, direct-traffic shape — and review them on the same cadence you review anything else. You will not get a clean attributed number, but within a quarter you will have a defensible read on whether AI is shaping your pipeline, which is the honest version of measuring a channel that refuses to leave a click. Start the loop now, while the signal is small, so the baseline exists when the channel grows.
The dark funnel: quick answers
Can’t I just track AI referral traffic?
Barely, and in 2024 almost not at all — which is the heart of the problem. A referral only exists if the AI sends a click to your site carrying a trace of where it came from, and most AI interaction does not work that way. A buyer asks a chatbot to compare options, gets an answer, and either acts on it later or not; no click leaves the conversation, so there is nothing to log. Even when a click does happen, the chatbot environment often passes no referrer, so your analytics records it as direct traffic with no idea it began in an AI tool. As we write, there is not even a consistent tagging convention to lean on. So the honest answer is that referral tracking captures only the thin slice of AI influence that ends in an attributed click, and misses the much larger part where the buyer is persuaded inside the conversation and never clicks at all. Treating that thin slice as the whole channel is how you conclude AI does nothing, when it is quietly doing a lot.
Isn’t “dark funnel” just an excuse for unattributable spend?
It can be abused that way, so the discipline is to treat it as a measurement challenge rather than a free pass. The dark funnel is not a place to hide results; it is an acknowledgement that real influence happens in channels last-click attribution was never built to see — private conversations, communities, and now AI tools. The wrong response is to shrug and stop measuring. The right one is to measure differently: with proxies that give directional signal even when precise attribution is impossible. If your brand searches are climbing, if your direct traffic is growing and behaving like people who already know what they want, if buyers on sales calls keep mentioning they "asked ChatGPT," those are evidence the dark funnel is working, not excuses for the absence of a clean number. The danger is not that the dark funnel is unmeasurable; it is that, lacking a tidy metric, teams quietly value it at zero and underinvest in something that is moving deals.
How do I start measuring it?
Start with the cheapest, most direct proxy: ask. A self-reported attribution question on your demo or contact form — an open "how did you hear about us?" rather than a fixed menu — surfaces AI as a source the moment buyers start naming it, long before any software could. Pair that with brand search tracking, because a rising tide of people searching your name specifically is often the downstream echo of being recommended somewhere you cannot see. Then look at the quality of your direct traffic: if it is growing and lands on deep, specific pages rather than your homepage, that pattern is consistent with people sent by an AI to a particular answer. None of these is precise on its own, but together they triangulate. Read across self-reported sources, brand search, and direct-traffic behaviour and you get a defensible read on whether AI is shaping your pipeline — which is far better than the false zero you get from waiting for a referral report that will never come.
Does this mean attribution is hopeless?
No — it means precise, click-level attribution is the wrong standard for this channel, and clinging to it is what makes the situation feel hopeless. Marketing has always had influence it could not trace to a click: word of mouth, a mention at a conference, a recommendation in a private thread. The dark funnel, AI included, is more of that, not a new species of unmeasurable. The mature response is to hold two truths at once — that you cannot perfectly attribute AI’s influence, and that you can still know, with reasonable confidence, whether it is growing and helping. That is enough to make decisions with. The teams that do well here are not the ones who crack a perfect attribution model; there isn’t one. They are the ones who stop demanding click-level certainty from a channel that does not produce it, and start managing it on the honest, directional signals it does.
A note on sources and certainty
This is written in September 2024. The dark funnel itself is an established idea — buyers have always made decisions in private channels that last-click attribution cannot see — and what this note adds is that AI chatbots have joined those channels as research advisors. We have described the mechanics that were already clear: closed conversations that emit no referrer or cookie, influence that resurfaces as mislabelled direct or branded traffic, and the proxy methods marketers already used for dark social. We have not quoted the large-scale studies that put numbers on this later — what share of buyers use AI, what percentage of traffic it becomes — because those were measured afterward, and importing them backward would lend a false precision to a September 2024 view. Notably, there is not yet even a consistent way to tag AI-driven visits as we write. What is solid is the shape of the problem, and the AC Group’s working stance on it for 27 years: measure influence honestly with proxies rather than pretend an untrackable channel simply does not exist.