Dark funnel and AI chatbot brand discovery: what's really happening

12 min readJuly 11, 2026By Spawned Team

AI chatbots now drive brand discovery before any Google search. Learn how the dark funnel works, what signals matter, and how to get cited. 2026 guide.

TL;DR: The dark funnel is buyer research that never shows up in your analytics: no referrer, no click, no session. AI chatbots like ChatGPT, Claude, Gemini, and Perplexity have become the dark funnel's biggest channel. When someone asks an AI assistant which tools to consider, your brand either gets named or it doesn't, and you have almost no visibility into when or why that happens.

What is the dark funnel and why do AI chatbots make it worse?

The dark funnel is all the buyer research and brand discovery that happens before a prospect visits your site or fills out a form. Word of mouth. Slack conversations. Private LinkedIn messages, podcast recommendations, offline events. None of it shows up in GA4. None of it has a UTM.

AI chatbots made the dark funnel a lot bigger. A buyer opens ChatGPT or Perplexity and types "what's the best project management tool for a 50-person remote team." They get a confident answer with three named tools. Then they go straight to one of those tools' websites, type the URL into their browser, and land in your analytics as direct traffic. You have no idea the AI was involved.

Forrester's 2023 B2B buying research found that 68% of B2B buyers prefer to research independently before engaging a vendor [1]. That was before ChatGPT became a mainstream research tool. The share doing AI-assisted research has only grown since. Nobody has precise figures on what percentage of dark funnel activity now runs through AI assistants specifically, but the usage curve tells part of the story: ChatGPT hit 100 million weekly active users by early 2023, and by early 2025 OpenAI reported over 400 million weekly active users [2].

The real problem is attribution. Someone discovers your brand through an AI chatbot and converts three weeks later. That conversion looks like direct, or branded search, or a sales touch. The AI's role is invisible to your stack. That's the dark funnel, amplified.

How do AI chatbots actually decide which brands to mention?

The honest answer is that it's partly opaque and actively being studied. But the mechanics are knowable enough to act on.

Large language models are trained on huge text corpora. Brands that appear often, in authoritative contexts, across many independent sources get encoded into the model's weights. When the model writes a recommendation, it pulls from those associations. High textual presence in training data means more mentions.

Training data isn't the whole story. Perplexity and Bing's AI mode do real-time retrieval, pulling live web results into the context window before generating an answer [3]. In those systems, your current SEO footprint matters directly. Pages that rank for relevant queries are the pages that get retrieved and cited. ChatGPT with browsing works the same way. Even in purely generative responses, the model's "knowledge" skews toward well-linked, often-cited content, because that's what dominated the training corpus.

A 2024 preprint from researchers at Columbia and Cornell studying citation behavior in AI search engines found that cited sources tended to have higher domain authority and more backlinks than uncited sources on the same topic [4]. That matches what practitioners see every day: traditional SEO signals still matter, they just work through a different mechanism now.

There's a subtler factor too. If your brand gets described in specific, definitive language across many independent sources, the model builds a clear concept of you. Vague brands, brands with conflicting descriptions, brands that only appear in their own marketing copy, they don't give the model much to grab. The AI isn't reading your website when it names you. It's drawing on what the rest of the web said about you.

What percentage of brand discovery now happens in AI chatbots?

Honest answer: nobody has clean numbers yet. This space moves fast, and most of the data is proprietary, based on small samples, or self-reported by parties with something to sell.

Here's what we do have. SparkToro's 2024 zero-click research found roughly 60% of Google searches end without a click [5]. That was before Google's AI Overviews rolled out widely. As AI Overviews answer more queries inside the results page itself, the share of search traffic reaching brand websites drops further.

On chatbot usage specifically, a 2024 Search Engine Land survey found 34% of respondents had used an AI chatbot to research a product or service before buying [6]. Among respondents under 35, that number was closer to 52%. These are self-selected survey figures, so treat them as directional, not precise.

The cleanest signal is your own analytics: a rising share of direct traffic with no campaign explanation, plus flat or declining referral and organic traffic, often tracks with growth in AI assistant usage. Some teams confirm it by surveying new customers about how they first heard of the brand.

The table below shows a rough discovery channel mix for a mid-market B2B SaaS brand, built from aggregated patterns practitioners have reported publicly. This is not a single study.

| Discovery channel | Estimated share of first touches (2023) | Estimated share (2025) | |---|---|---| | Organic search (tracked) | 38% | 28% | | Direct / dark funnel (incl. AI) | 31% | 41% | | Paid search & social | 18% | 17% | | Referral (tracked) | 8% | 7% | | Other | 5% | 7% |

These are rough practitioner estimates, not peer-reviewed data. The direction, tracked organic shrinking and dark/direct growing, is consistent with multiple sources including Similarweb traffic reports for major publishers [7].

AI chatbot use for product research by age group

| | | |---|---| | All respondents | 34% | | Under 35 | 52% |

Source: Search Engine Land, AI Chatbot Consumer Survey, 2024

Why can't I see AI chatbot traffic in Google Analytics?

ChatGPT, Claude, and most AI assistants don't send referrer headers when a user clicks a link from a response. The browser gets the URL, but the HTTP referrer field is empty or set to the chatbot's own domain in a way GA4 often buckets as direct.

Perplexity is the partial exception. It does send referrer data for links clicked in its citations panel, and some analytics setups see perplexity.ai as a referral source. But the bigger issue is that most AI chatbot interactions never produce a click at all. The user reads the brand mention, closes the chat, and either Googles the brand name or types the URL directly. That journey leaves zero AI-attributable signal.

The gap between AI-driven awareness and AI-driven tracked sessions is huge. Someone could see your brand five times in AI responses over three months before ever visiting your site. When they finally convert, your CRM shows "organic branded search" as the first touch. The AI's work stays invisible.

One practical workaround: add a "how did you first hear about us" field to your onboarding flow or demo request form, with AI assistant as an explicit option. Several marketing teams have reported that 10 to 20% of new customers self-report AI chatbots as their first discovery channel, which is almost certainly undercounted in their analytics. The true share runs higher, because people don't always remember or report accurately.

What signals actually make AI chatbots more likely to mention your brand?

Think of it as building brand surface area across the web. The more places your brand shows up, in credible and specific contexts, the more likely an AI's training or retrieval process surfaces you.

High-quality backlinks still matter. A 2024 Ahrefs analysis of which domains appear most in AI-generated responses found a strong correlation between referring domain count and AI citation frequency [8]. Same signal that drives traditional Google rankings, which makes sense: AI search systems either retrieve from ranked pages or trained on a corpus dominated by well-linked content.

Third-party mentions beat your own content. An AI doesn't weight your "about us" page heavily. It does weight a detailed comparison on G2, a review in a respected industry newsletter, a mention in a journalist's roundup. Getting your brand accurately described by independent sources is worth more than any amount of on-site optimization.

Specificity helps. If multiple independent sources describe you as "an inventory management tool for mid-sized e-commerce brands with Shopify integration," that consistent framing gives the model a clear concept to retrieve. Vague descriptors like "an innovative platform for modern businesses" are close to worthless.

Schema markup helps retrieval-based systems. Pages using proper Organization, Product, or Review schema give systems like Perplexity and Google's AI Overviews cleaner structured signals to pull from [9].

The generative engine optimization framework covers these signals in more depth, including the content formats that get cited most.

Speed and uptime matter indirectly. Retrieval-based systems crawl and index the web. Slow or frequently unavailable pages get deprioritized. Standard technical SEO hygiene applies here too.

How is the dark funnel different from traditional brand awareness?

Traditional brand awareness is broadcast. A TV spot, a display ad, a sponsorship. You spend money, reach an audience, and hope some of it sticks. The funnel is visible: impressions, reach, aided recall in brand studies.

The dark funnel is conversational and contextual. When a buyer asks an AI "what should I use for X," the answer is personalized to their query, delivered at the exact moment of intent, and it feels authoritative even when it shouldn't be. That's a different kind of exposure than a banner ad. A recommendation inside a trusted conversation converts at higher rates than passive awareness ever did.

The other difference is scale. One well-placed, accurate description of your brand in training data or high-authority sources can generate thousands of AI recommendations a month. You can't buy that placement directly, at least not yet the way you buy a Google ad. You earn it through content, PR, community presence, and third-party validation.

This changes how you should think about content investment. Content that explains your category and describes your brand's position in it accurately, even if it only ranks on page two of Google, can generate more dark funnel value than content ranking first for a low-intent keyword. AI systems retrieve semantically relevant content, not only highly ranked content.

Can you measure dark funnel AI activity at all?

Not perfectly. But you can get much closer than most teams are today.

The most direct method is prompt auditing: regularly querying AI chatbots with the questions your buyers actually ask, and recording whether and how your brand shows up. Tools built for AI search visibility metrics and KPIs automate this at scale. Manual auditing with a spreadsheet works fine for smaller brands or for a baseline.

The questions should mirror real buyer queries. Not "tell me about [your brand]" but "what are the best tools for [use case]" and "compare [your category] options for [buyer type]." That's where you find out whether you're in the consideration set.

Brand survey data helps. Ask new customers how they first heard of you, with AI assistant as an explicit option. Run it for 90 days and you'll have a real, if self-reported, baseline.

Direct traffic trends can flag AI activity. If direct traffic is climbing while your offline spend held flat, and branded search volume is climbing too, AI-driven discovery is a plausible explanation. Not proof. A signal worth tracking.

Some teams experiment with unique tracking URLs in structured data or AI-indexable content, hoping to catch clicks that come through citation links in Perplexity or Bing. This works for retrieval-based systems and does nothing for purely generative responses.

Spawned's AI visibility tool and similar platforms build dashboards to track share of voice across AI assistants over time, which gets you closer to a systematic answer than manual auditing alone.

How does this affect B2B versus B2C brands differently?

B2B brands have more to gain and more to lose from AI chatbot discovery.

B2B buyers do extensive independent research before engaging vendors. They ask peers, read comparison sites, and increasingly ask AI assistants. The consideration set for a $50,000 software purchase often gets set through dark funnel research, and an AI recommendation early in that process can put you on a shortlist before you've had any direct contact with the buyer. Getting into that initial response matters a lot.

B2C cycles are shorter and more impulse-driven, so dark funnel AI activity less often decides an individual purchase. But for considered B2C decisions, a car, a financial product, a health device, AI recommendations are increasingly influential. And for loyalty and advocacy, having your brand correctly described and positively framed in AI responses shapes perception even when no immediate purchase is at stake.

Content formats differ too. B2B brands benefit most from technical comparisons, integration guides, use-case documentation, and third-party analyst coverage. Those are the sources AI systems draw on for complex product recommendations. B2C brands benefit more from review site presence, editorial coverage, and social proof signals that get indexed and retrieved.

Either way, the underlying principle holds: you need the wider web to describe your brand accurately, in specific terms, across many independent credible sources. Your own website is necessary but not enough.

What content actually gets cited by AI search engines?

There's real data here now. A 2024 Bain & Company study found that AI search results disproportionately cited pages with structured, scannable formats: numbered lists, comparison tables, defined terms, and clear headings that match the user's question [10]. Long-form prose without clear structure got cited less, even when the content quality was high.

FAQ sections get retrieved disproportionately often. When a page has explicit question-and-answer blocks that mirror how buyers phrase their queries, AI retrieval systems extract and use that content more reliably. This isn't a trick. It's matching your content's structure to how the AI reads and retrieves.

Original data and specific claims carry weight. A page that says "our customers report a 34% reduction in onboarding time," with context and methodology, is more citable than a page that says "our customers see dramatic improvements." AI systems that include citations tend to prefer sources with concrete, verifiable-sounding claims.

Brand mentions in third-party reviews, roundups, and comparison articles matter a lot. G2, Capterra, TrustRadius, and independent analyst blogs appear often in AI citation sources because they have high domain authority and structured content. Getting accurate, positive presence on those platforms is high-value dark funnel work.

For a full breakdown of which content formats and technical signals affect AI citation rates, the AI SEO guide covers the research in more detail.

How should you change your marketing strategy to address dark funnel AI discovery?

Most marketing teams still optimize for a world where buyers arrive through tracked channels. That world hasn't disappeared. It's just shrinking relative to dark funnel activity.

The practical shift is investing more in what you can't directly attribute. Third-party editorial coverage, community presence, podcast appearances, review site management, analyst relations. All of it builds the distributed brand presence that feeds AI training data and retrieval systems. These investments are hard to report in a weekly dashboard, which is exactly why most performance marketers underinvest in them.

Audit your AI visibility regularly. At least monthly, run the 10 to 15 queries your buyers are most likely to type into an AI assistant. Record whether you appear, where in the response, how you're described, and whether competitors show up more prominently. This takes 30 minutes with a spreadsheet or a few minutes with a dedicated tool. The data shapes your content and PR roadmap.

Fix your description problem if you have one. If AI chatbots describe your brand vaguely, incorrectly, or not at all, the fix is creating more specific, consistent, publicly available descriptions and getting them replicated across independent sources. Your Wikipedia article (if you have one), your Crunchbase profile, your G2 listing, your press coverage all need to agree on what you do and for whom.

Don't abandon SEO. Because many AI systems retrieve from ranked pages, traditional SEO and AI search visibility work together more than they compete. Pages that rank on Google for relevant queries are the pages Perplexity and Bing's AI mode retrieve. The AI search and AI mode SEO tool articles cover how to optimize for both at once.

If you want a systematic starting point, an AI visibility audit is the fastest way to see where you actually stand across the major AI assistants before deciding where to focus.

What are the biggest mistakes brands make with dark funnel AI strategy?

First mistake: treating AI search like a new version of keyword SEO. Stuffing pages with AI-related keywords or publishing thin content optimized for AI detection tools does nothing. The signals that drive AI citation are earned credibility signals, not on-page keyword density.

Second mistake: ignoring the category definition problem. If AI chatbots don't have a clear concept of your category, they can't recommend you for the right queries. Brands that invented a new category, or operate in a fragmented space, often get mentioned inconsistently because the model has no clear framework for when to surface them. Creating and distributing clear, consistent category definitions, what you do, for whom, and why it's different, is foundational work that many brands skip.

Third mistake: measuring only what's easy. Teams tracking only sessions and conversions systematically undercount AI's contribution to pipeline. When AI looks small in the data, it gets defunded. Building even rough dark funnel measurement into your reporting changes the investment decision.

Fourth: over-rotating to AI optimization at the expense of channels that still work. Organic search, email, paid social, and content marketing still generate real, measurable returns for most brands. AI dark funnel strategy is additive, not a replacement. The brands getting this right keep their existing channel discipline while building AI-specific visibility in parallel.

Fifth: not updating your brand description assets when you change your product or positioning. Models trained on old data may describe your brand the way you positioned it two years ago. If your messaging shifted, the public assets that encode your brand description (press releases, review profiles, directory listings) need updating too. The model can't know you changed unless the web does.

Sources

Frequently Asked Questions

Does the dark funnel actually affect small brands or just big enterprise names?

Small brands are affected, often more than they realize. AI chatbots will mention niche or smaller brands if they have strong third-party presence in specific category queries. A small tool with 50 detailed G2 reviews and solid coverage in a niche newsletter can outperform a bigger competitor in AI responses for a specific use case. Being small doesn't disqualify you. Having low third-party visibility does.

How is AI chatbot brand discovery different from what Perplexity does versus ChatGPT?

Perplexity does real-time retrieval and cites its sources in the response. Your current SEO and content quality directly affect whether Perplexity cites you. ChatGPT in its default mode draws primarily on training data without live retrieval, so your brand presence in pre-training web data matters more. ChatGPT with browsing enabled behaves more like Perplexity. For most marketing purposes, you need both training-data presence and current web rankings.

Can I pay to get my brand recommended by AI chatbots?

Not directly, in any transparent or guaranteed way, as of mid-2026. OpenAI, Anthropic, and Google have not released ad products that place brands in organic AI responses. Perplexity has tested sponsored follow-up questions, which is adjacent but not the same as earned recommendations. Any vendor promising guaranteed AI chatbot placement in organic responses is overpromising. The current path is earned through content, backlinks, and third-party mentions.

How often should I audit my brand's visibility in AI chatbots?

Monthly is the practical minimum for most brands. Quarterly is too infrequent given how quickly AI systems update, especially retrieval-based ones like Perplexity. If you're in a category with fast-moving competitive dynamics, weekly audits of your top 10 buyer queries make sense. The audit is simple: run your target queries in each major AI assistant and record your mentions, your competitors' mentions, and how you're described.

What is 'share of voice' in AI search and how do I measure it?

AI share of voice is the percentage of relevant AI responses in your category that mention your brand, compared to how often competitors get mentioned. You measure it by running a consistent set of category and use-case queries across AI platforms, recording every brand mentioned per response, and calculating your mention rate. Tools purpose-built for AI visibility tracking automate this. Without automation, a spreadsheet with 15 queries run weekly gives a workable baseline.

Does having a Wikipedia page actually help AI chatbots know about my brand?

Yes, meaningfully. Wikipedia is heavily represented in LLM training data, and multiple AI companies have confirmed they used Wikipedia as a training source. A Wikipedia page that accurately describes your brand, category, and key product details gives models a high-quality reference to draw from. Not every brand qualifies under Wikipedia's notability guidelines, but for those that do, maintaining an accurate entry is high-value AI visibility work.

How do AI chatbots handle brand recommendations for sensitive categories like finance or health?

AI assistants apply extra caution in what Google calls 'Your Money or Your Life' categories. Health and financial brand recommendations often come with disclaimers, and the AI may be less specific in naming brands to limit liability. That said, brands with strong third-party clinical or regulatory evidence, FDA clearance, registered investment advisor status, and similar credentials are more likely to appear in those contexts. Authoritative third-party signals matter even more in regulated categories.

What's the relationship between zero-click search and AI dark funnel activity?

They're related but distinct. Zero-click search means users got their answer on the Google results page without clicking a link. SparkToro found this applied to roughly 60% of Google searches in 2024. AI dark funnel activity extends this: buyers ask AI chatbots instead of Google at all, never generating a search impression you could even see. Zero-click reduces traffic from search; AI dark funnel reduces the share of research that touches Google at all.

Is ChatGPT's training data updated frequently enough for it to know recent brand news?

No, not in real-time. ChatGPT's base model has a training cutoff date. Major product launches, acquisitions, or positioning changes after that cutoff won't be reflected in the model's knowledge unless it's using retrieval. OpenAI updates the base model periodically, not continuously. For recent developments to appear, they need to be indexed and retrievable via ChatGPT's browsing feature, or wait for the next model training cycle.

How does the dark funnel affect B2B pipeline attribution?

It creates what practitioners call the 'phantom pipeline' problem. Deals that close with 'unknown' or 'direct' as their first touch often had AI-assisted discovery somewhere in the journey. Multi-touch attribution models miss this entirely. The practical fix is a mix of survey data at sign-up, post-sale customer interviews about their research journey, and tracking growth in branded search volume as a proxy signal for rising AI-driven awareness.

What types of third-party content does AI use most when recommending brands?

High-authority review platforms like G2, Capterra, and TrustRadius are heavily represented in AI citation sources. Industry analyst reports, editorial roundups in respected trade publications, and comparison articles from sites with strong domain authority also appear frequently. Academic citations of your research, if you produce original data, carry weight too. The common thread is external, structured, credible sources that describe your brand specifically and accurately.

Can negative reviews or misinformation in the dark funnel hurt my brand in AI responses?

Yes. If your brand has a pattern of negative reviews on indexed platforms, or if misinformation about your product spreads across credible-looking sources, AI models can encode and replicate those negative associations. This is one reason review management and proactive PR matter for AI visibility more than for human readers alone. Correcting factual errors in Wikipedia, responding to reviews, and publishing clear factual content all help counteract negative signal.

How do I know if my brand is being recommended incorrectly by AI chatbots?

Run a prompt audit. Ask AI assistants open-ended questions in your category and read every mention of your brand carefully. Common problems include wrong use case attribution, outdated pricing or feature descriptions, confusion with a similarly named competitor, or correct mentions with weak or inaccurate descriptions. Document the discrepancies and trace them to source content that needs updating. This audit also reveals which competitors are being recommended instead of you, and why.

dark funnel and ai chatbot

Ready to try it?

Build your first app in a few minutes.

Start Building