The Dark Funnel in the AI Era: How AI Search Creates a New One, and How to Get Into It
Oct 9, 2026 · 10 min read
Built on Gartner's March 2026 survey of 646 B2B buyers, the 2012 Atlantic essay that named dark social, and MDN's documentation of the Referer header. Written for the marketing leader whose pipeline keeps arriving from sources the dashboard cannot name.
The dark funnel is the research, conversations, and recommendations that shape a B2B purchase but never reach the seller's analytics or CRM. Buyers compare vendors in peer groups and private chats, then arrive as "direct" traffic or a branded search. The steps that decided the deal leave no record.
AI assistants have added a new layer to it. A buyer can ask ChatGPT for a shortlist and get one, so the first cut of vendors happens inside an answer the seller never sees. Earning a place in that answer is the job of answer engine optimization, and the second half of this article covers how to do it.
Key Takeaways
- Dark means unmeasured: the dark funnel covers every buying step that happens where analytics cannot follow, from a Slack recommendation to a podcast mention.
- Buyers prefer to research alone: 67% of B2B buyers say they prefer a rep-free experience, according to a Gartner survey published in March 2026.
- AI assistants added a new layer: the shortlist can now form inside an answer, before the buyer visits a single vendor site.
- More tracking will not light it up: an answer that names you and sends nobody produces no event in your analytics.
- Presence in answers is the lever: engines name the brands they can find described plainly, on their own pages and on the sources the engines draw on.
What Is the Dark Funnel?
Asked what is the dark funnel, most B2B marketers describe the part of the buying journey that happens out of view. The working definition: every step a buyer takes toward a purchase that never produces a record the seller can attribute. "Dark" means unmeasured, in the sense an analyst would use, rather than secret or illicit.
Some leaders hear “dark funnel” and assume it is an excuse for marketing that cannot prove results. It is not. The concept draws a line instead. Some things you can measure: presence in answers, branded demand, and the source a buyer states. The rest you cannot, and you should stop counting it as zero.
Where the Term Came From
Dark social came first. In October 2012, Alexis Madrigal of The Atlantic coined “dark social” in an essay on untracked sharing. He showed that links passed through email programs, instant messages, and some mobile apps arrived with no referrer data, so analytics filed them as "direct" or "typed/bookmarked" traffic.
The mechanism is a missing header. A browser tells a site where a visitor came from through the HTTP Referer header, which MDN says “allows a server to identify referring pages that people are visiting from.” When the referring app sends nothing, the visit can look as if someone typed your address. Madrigal’s explanation was blunter: “Most of the time, someone Gchatted someone a link.”
The funnel version came later. "Dark funnel" spread through B2B marketing in the years after, as marketers applied the same logic to a whole buying journey rather than a single shared link.
Dark Funnel vs. Dark Social
One is a channel problem, the other a journey problem. Dark social is the untracked sharing of links. The dark funnel is every untracked buying activity, and dark social is one part of it.
| Dark social | Dark funnel | |
|---|---|---|
| What goes untracked | A shared link | Any step in a purchase decision |
| Typical sources | Email, chat, and messaging apps | Peer groups, communities, podcasts, events, and AI assistants |
| How it shows up | "Direct" traffic | Direct visits, branded searches, and "heard about you" leads |
| Who it frustrates | Publishers counting referrals | B2B teams explaining pipeline |
Dark Funnel Marketing: Where B2B Buyers Research Without Leaving a Trace
B2B buyers do much of their research before a seller hears from them, and they say they prefer it. In a Gartner survey of 646 B2B buyers, run from August through September 2025 and published in March 2026, 67% said they prefer a rep-free experience. Gartner's Alyssa Cruz describes buyers who "search, vet suppliers, and shape solution needs with fewer early interactions."
Read the figure for what it is: a stated preference, not a measurement of untracked research. It tells you buyers want to work alone. It cannot tell you what share of that work your analytics misses.
The Channels Nobody Tags
Peer channels carry the most weight and the least data. A B2B buyer forms a view in these places without ever identifying themselves:
- Private groups: Slack and WhatsApp communities where practitioners ask each other which vendor they use.
- Direct messages: a one-line recommendation from a former colleague.
- Practitioner communities and forums: the threads behind community-led growth as a motion.
- Podcasts and events: a vendor named on stage or in an episode, then searched weeks later.
- Analyst conversations: inquiry calls and briefings that never touch your site.
- The buying group's own threads: internal channels where the committee trades links and opinions.
What Dark Funnel Marketing Asks Teams to Do Differently
Dark funnel marketing means creating material worth passing around in those places, rather than forcing every touch through a form. The test changes: a benchmark, a template, or a plain explanation earns its keep when a buyer forwards it, even if your analytics never sees the forward.
The trade-off is control. You give up a lead record for each download in exchange for influence that travels without one. Teams that accept the trade publish more of their thinking in the open.
How AI Search Creates a New Dark Funnel
The dark funnel B2B teams now face has a layer the classic version lacked. A buyer opens an assistant, describes the problem, and asks which vendors fit. The assistant summarizes the category and names a few companies, and the buyer's shortlist starts there.
The Shortlist Now Forms Inside an Answer
The decisive moment moves ahead of every touchpoint you track. In the classic dark funnel, a buyer still found vendors somewhere visible: a results page, a review site, a peer's message. Now the assistant does the finding, and the buyer may never hear of the vendors it left out.
Many of these sessions end without a visit, the pattern covered in zero-click search. Some happen in private assistants behind the buyer's firewall, which pass no query to you. HubSpot now reports AI as a traffic source of its own, the change behind HubSpot's AI referrals signal.
Three Ways the Assistant Layer Hides Influence
The three routes below are this article's own reasoning about where assistant influence goes missing.
- An answer that names you and sends nobody. The buyer reads your name and closes the tab. With no visit, there is no event.
- A later visit that arrives as branded search or direct. Days later the buyer types your name into Google or your URL into the browser. Search or direct takes the credit for a decision the assistant shaped.
- A click with a thin or missing referrer. When a buyer clicks a link inside an assistant, your analytics can only use what arrives with the visit: the Referer header, which the referrer policy can trim, or a tag on the link itself. If neither names a source, the visit lands in the same "direct" bucket Madrigal described in 2012.
Why This Layer Is Different From Dark Social
A model writes the recommendation now. In classic dark social, a person chose your link and passed it to someone else, the way Madrigal described links traveling by email and chat. In the assistant layer, the model composes the recommendation from what it has read about your category.
That changes where your influence sits. What you can shape is the reading list: whether the sources a model draws on describe you accurately, and whether you appear on enough of them to be named.
Dark Funnel Examples: A Buyer Journey Through ChatGPT
The journey below is illustrative, not a reported case. Dark funnel examples are easiest to read as a single journey, and this one follows a buyer through several channels, starting with an AI assistant.
The Journey, Step by Step
The buyer: a VP of Finance at a mid-size B2B SaaS company wants to replace a spreadsheet close process with spend-management software that handles several legal entities. Vendor names are placeholders.
- Asks ChatGPT for options: "Which spend-management tools fit a multi-entity SaaS company?" The answer names four vendors. Yours is missing, or it appears with an outdated description.
- Narrows the field: a follow-up asks the assistant to compare Vendor A and Vendor B on multi-entity support.
- Checks with peers: the VP reads a thread in a finance community and asks a CFO Slack group for opinions. This step is classic dark social.
- Searches the finalists: the VP searches both vendor names on Google and reads their pricing and integration pages.
- Books a demo: the VP requests a demo from Vendor A, and in the form's "how did you hear about us" field, types "Google".
What the Vendor's Analytics Recorded
| Step | What happened | What Vendor A's analytics recorded |
|---|---|---|
| 1. First question | The assistant names four vendors | Nothing |
| 2. Comparison | The assistant compares Vendor A and Vendor B | Nothing |
| 3. Peer check | A community thread and Slack opinions | Nothing |
| 4. Searches | Pricing and integration pages read | Branded organic sessions |
| 5. Demo request | Form submitted | One lead, source: Google |
The assistant created the shortlist, and the analytics recorded none of it. Vendor A's report credits Google for a deal the assistant started. The brand left out of that first answer never learned there was a deal to lose.
Why Your Analytics Misread It (and Where to Go for Attribution)
The dashboard files the dark funnel under other names. Unattributed visits can appear as direct traffic, as Madrigal documented in 2012. Assistant-driven discovery can appear later as branded search when a buyer searches for a name the assistant supplied. Buyers then fill in the "how did you hear about us" field with the last thing they remember, which in the example above was Google.
For a CMO, the rule is to stop reading flat traffic as absent demand. Report how often the brand appears in answers and how branded demand changes alongside sessions. Do not report assistant influence as a percentage of pipeline. You cannot observe the total influence, so there is no denominator to calculate. Recovering what can be recovered, from referrer rules to CRM fields, is the subject of attributing pipeline when an assistant sits in the middle.
How to Get Into the Conversation: Earning Citations Where Buyers Ask
The assistant layer stays untracked, so the work moves to what the assistant says. The goal is to be named, and described accurately, when a buyer asks the question. The sequence below is this article's recommended order of work.
1. Start From the Questions Buyers Ask
- Write down the questions in your buyers’ words. Pull them from sales-call notes and first-call conversations. “Which spend-management tools handle multiple entities?” tells you more than “best spend management software.”
- Run them across the engines. Ask each question in ChatGPT, Claude, Perplexity, Gemini, and any other assistant your buyers use. Record who gets named and which sources the answers cite.
2. Be Present Where the Answers Are Sourced
- State your category facts plainly on your own pages. Write what you do, who it fits, and what it integrates with, in sentences an engine can quote without the surrounding page.
- Earn presence on the third-party sources the engines draw on. Communities, publications, and review sites carry weight your own domain cannot, which is why off-site beats on-site for this work. The community presence buyers already trust is also what produces the citations a monitor measures.
3. Read Presence, Not Traffic
- Re-run the same questions on a fixed schedule. Compare who gets named, how you are described, and which sources appear. A change in the answer is the signal to watch, since a change in sessions may arrive much later or not at all.
Where Klarivo Fits in the AI Dark Funnel
Klarivo works on the presence side of the new dark funnel. In the homepage's words, "Klarivo is the managed AEO and Reddit service that gets B2B brands named in ChatGPT, Perplexity, Gemini, and Google AI." The work happens off-site: Klarivo "builds the Reddit, review, and publication footprint models retrieve when buyers ask who to buy."
Klarivo Monitor tracks ChatGPT, Claude, Perplexity, Gemini, and Grok, each provider tracked independently. It reports the top citation sources as domains, updated on a schedule you control. That tells you whether engines name you and which domains they drew on. The buyer's path before a demo request stays untracked, which is why presence in answers is the reading to watch.
The fintech case study also states its limit: attribution stops at the domain. Use Klarivo Monitor to track which answers name you, and Klarivo’s Reddit marketing program to build the footprint behind them. Book a Klarivo discovery call. The call takes 15 minutes, and you choose the time.
FAQ
Can intent data tools see into the dark funnel?
Partly. They infer account-level interest from activity on the sites and networks they can observe. None of them can see a private conversation between a buyer and an AI assistant, so the newest layer stays outside their view.
Is the dark funnel only a B2B problem?
No. Consumers research in private too. B2B feels it more because several people research the same purchase, often before anyone contacts a vendor. In Gartner's March 2026 survey, 67% of B2B buyers said they prefer a rep-free experience.
Should we stop gating content because of the dark funnel?
Not all of it. In this article's view, gate what a buyer wants after deciding, such as an implementation plan, and open the material a buyer would pass to a colleague while still deciding. A form-gated PDF is hard to forward to a Slack group, and forwarding is how material travels through the dark funnel.
How can sales teams spot dark funnel influence in a deal?
Ask in the first call: "What did you read, or ask, before you reached out?" Their response may reveal a peer thread, a podcast, or an AI recommendation that no form captured. Note it in the opportunity record so marketing can see patterns across deals.
Will AI assistants ever make the dark funnel trackable?
Not the part that matters most. An answer that sends no visit will never produce an analytics event. Search platforms now report on their own AI features: Google Search Console shows impressions in AI Overviews and AI Mode, while Bing Webmaster Tools reports aggregated citations across Copilot, Bing’s AI-generated summaries, and select partner integrations. Neither shows the conversations themselves, so the answers that sent nobody stay dark.
Start accelerating your Reddit presence
See how Klarivo can shift your visibility across AI, search and buyer communities.
Book a Demo
