The dark search funnel: why AEO is difficult to measure with attribution software

How do you measure AEO and know if your strategy is working? That's the question I hear dozens of times per month when speaking with marketing leaders at SaaS companies.

There's no shortage of AI visibility software in the market right now but nobody truly knows what buyers are searching in LLMs like ChatGPT because they don't expose impression data. You could be influencing buyer decision making - or be completely absent from AI answers - and be none the wiser.

This is the dark search funnel, the biggest attribution gap your marketing stack has never accounted for. This article explains why traditional attribution software is blind to AI-created demand and gives you a concrete three-layer framework to measure and capture the pipeline your CRM cannot currently see.

How AI search created a new blind spot in B2B marketing

The dark funnel concept was Chris Walker at Refine Labs popularized to describe buyer touchpoints attribution software cannot track. As Cognism explains, the dark funnel covers "the places that buyers are engaging and making decisions that no attribution software or tracking can account for," representing 75% or more of the path to purchase.

The original dark funnel channels were social media, private communities, word of mouth, and podcasts. AI search has now added an entirely new and significantly larger layer on top of all of them.

According to Forrester research, 89% of B2B buyers now use generative AI at some point in their buying process. Responsive reports that one in four B2B buyers use AI more often than conventional search when researching suppliers, and two-thirds rely on AI chatbots as much or more than Google when evaluating vendors. B2B buying trends data also shows a measurable decline in traffic on B2B vendor websites, consistent with buyers stopping their research inside LLMs rather than clicking through to individual sites.

When buyers arrive at your site after AI has already shortlisted competitors they have been actively comparing, your MQLs convert at lower rates because they are further along in a process you were never part of. This is why MQL-to-opportunity conversion can drop even when traffic stays flat.

Why traditional attribution software misses AI referrals

The technical gap is straightforward, but its commercial impact is substantial. As RSA Creative Studio documents, AI-enabled browsers often do not reliably pass referrer data to GA4, causing AI-driven visits to appear as direct or "(not set)." GA4 assigns traffic sources at the session level, and when referral information is absent, the entire session gets labeled as direct traffic with source "(direct)" and medium "(none)."

The user journey that creates this gap is straightforward: a buyer queries an LLM, reads an AI-generated summary with brand mentions, and forms a shortlist. Some time later, the buyer searches your brand name directly. GA4 records the session as "direct" with zero connection to the AI touchpoint that created the demand.

The demand creation moment has moved from Google into private LLM chat windows, and no standard attribution tool follows it there. For a deeper look at how AI Overviews specifically affect click behavior, see our guide on how Google AI Overviews works.

The three-layer framework for measuring AI search visibility

You cannot fix what you cannot measure. Since standard attribution tools are blind to AI-created demand, you need a framework built from three complementary layers that together give you a complete picture of AI-influenced pipeline.

Track leading indicators through LLM testing

Leading indicators track your brand's presence in AI answers before that presence translates into pipeline. You measure three metrics:

• Mention rate: The percentage of AI responses to category-relevant queries that mention your brand.
• Citation rate: The percentage of AI responses that link directly to your content or cite it explicitly.
• Share of voice by topical cluster: Your citation presence relative to competitors, expressed as a percentage of total category citations across tracked platforms and query groups.

To track these, run structured query testing across the major platforms. Map buyer-intent questions covering your core use cases and run them consistently across ChatGPT, Claude, Perplexity, and Gemini. Record which brands appear, how they are described, and whether your content is cited. For example, if you sell sales enablement software, test "What is the best sales enablement platform for Salesforce users?" across all four platforms. If your brand appears in 1 out of 10 responses while competitors appear in the other 9, you have roughly a 10% share of voice on that query cluster and a concrete gap to close.

At Discovered Labs, we use internal technology to automate this at scale to build a knowledge graph of citation patterns by cluster, content format, and query type. This allows us to improve citation win rates across client programs rather than guessing from anecdotal tests. (For a detailed comparison of AI citation tracking platforms, see our Discovered Labs vs. SE Ranking analysis.)

Monitor traffic signals and organic growth patterns

LLM testing shows where you stand in AI answers. Your website analytics confirm whether that visibility is driving demand. Two signals are particularly useful.

Branded organic growth. When AI mentions increase your brand's visibility, buyers who were not previously aware of you start searching your brand name. If your branded search volume grows while non-branded organic traffic stays flat, this pattern may indicate AI-created demand is working. It will not appear as AI referral traffic, but the growth pattern is distinctive and worth tracking monthly.

AI referral traffic reports. GA4 records some AI-referred sessions, particularly from platforms like Perplexity that pass referrer data more consistently than others. Build a custom segment filtering for known AI referral domains and monitor this alongside your direct traffic trend. Our guide on measuring Google AI Overviews pipeline impact covers how to structure this reporting inside GA4. AI Overviews more than doubled their query coverage from 6.49% to 13.14% between January and March 2025, and when an AI Overview appears, organic click-through rates drop from 1.41% to 0.64%, a 54.6% decline. If your traffic appears stable while AI Overview coverage expands, you are likely losing clicks you are not measuring.

Capture conversion data with self-reported attribution

The highest-signal layer is also the simplest: ask buyers directly how they found you.

Add a required "How did you hear about us?" field to every demo request and contact form, with specific options for ChatGPT, Perplexity, Claude, Gemini, and a general "AI search" choice. This single field recovers attribution data that no tracking script can collect. As Cognism notes, Refine Labs pioneered this approach for dark funnel measurement and it consistently outperforms software-based attribution for revealing the true demand source.

For deals already in your pipeline, Birdeye's AI attribution guide recommends creating CRM fields for "AI platform used" and "AI discovery: yes/no," then training your sales team to ask about research methods during discovery calls. Set your conversation intelligence platform (Gong, Chorus, or Outreach) to flag keywords like "ChatGPT," "Perplexity," "AI recommended," and "I asked AI." Map flagged calls to closed-won deals in Salesforce to build a revenue dataset your CFO can read. Authority Tech's attribution playbook reinforces that this composite approach is currently the most reliable way to quantify AI-influenced pipeline.

How to illuminate the dark funnel and capture AI-driven pipeline

Measuring the gap is the first step. Closing it requires structural changes to how you produce and distribute content.

Audit your current AI search visibility against competitors

Before you build a content plan, you need a baseline. Run your 30 most important buyer-intent queries through ChatGPT, Claude, Perplexity, and Gemini and record three things for each: which competitors are cited and how often, whether your brand appears at all, and what specific content or sources the AI references. Our competitive AEO audit guide walks through how to structure this benchmarking systematically.

AI search visibility audit checklist:

• Select 30 to 50 buyer-intent queries across your core use cases and ICPs
• Test each query across ChatGPT, Claude, Perplexity, and Gemini
• Record mention rate and citation rate per platform for your brand and top three competitors
• Calculate share of voice by topical cluster
• Identify the content types and sources the AI platforms prefer to cite
• Flag the 10 queries with the largest competitive citation gap
• Build a content roadmap targeting those 10 gaps first

Structure content for LLM retrieval using the CITABLE framework

Most content fails to earn AI citations not because it is low quality but because it is not structured for how LLMs retrieve and extract information. The CITABLE framework (Clear entity, Intent architecture, Third-party validation, Answer grounding, Block-structured, Latest & consistent, Entity graph) we developed addresses this directly while keeping the content readable for humans. For a full breakdown of how it compares to other AEO approaches, see our CITABLE vs. Growthx methodology comparison.

Each piece of content should follow all seven elements:

• C - Clear entity & structure: Open with a 2 to 3 sentence BLUF (bottom line up front) that names your company, product category, and primary use case, giving the LLM an unambiguous entity signal.
• I - Intent architecture: Answer the main question in the first paragraph, then address adjacent questions in subsequent H3 sections, covering the full range of queries a buyer might ask about this topic.
• T - Third-party validation: Embed references to reviews, user-generated content, community mentions, and news citations within the content body as described evidence, not just as links.
• A - Answer grounding: Back every claim with a verifiable, linked fact. LLMs prioritize content with explicit sourcing over unsubstantiated assertions.
• B - Block-structured for RAG: Write in 200 to 400 word sections with clear headings, tables, FAQ blocks, and ordered lists, matching how retrieval-augmented generation (RAG) systems chunk and index content.
• L - Latest & consistent: Include timestamps and update dates. Ensure every factual claim about your company, pricing, or capabilities is consistent across your site, G2 profile, LinkedIn, and third-party mentions.
• E - Entity graph & schema: Make relationships between your company, product category, use cases, and integrations explicit in the copy and reinforce them with Organization, Product, and FAQ schema markup.

For practical guidance on FAQ schema and how it improves AEO rankings, see our FAQ optimization guide. Our 15 AEO best practices guide covers the full checklist across Google AI Overviews and ChatGPT.

Table: Traditional SEO vs. AEO content structure

Build third-party validation to shape the AI consensus

AI models treat the consensus of external sources as more reliable than your website claims. Your citation rate depends heavily on what others say about you in places LLMs crawl and trust.

The three highest-impact third-party validation channels for B2B SaaS are:

1. Reddit: LLMs train heavily on Reddit data and frequently surface Reddit discussions in responses. Publishing in the right subreddits with genuine, non-promotional content shapes the narrative around your category. If you need support with Reddit strategy, Discovered Labs offers Reddit marketing using aged, high-karma accounts to publish in target subreddits without triggering spam filters. For content tactics that specifically influence LLM retrieval, see our guide on Reddit comments that LLMs reuse.
2. G2 and Capterra: Review platforms with structured, third-party content carry high trust signals for AI systems. Consistent, up-to-date profiles with detailed customer reviews improve your likelihood of citation in competitive category queries.
3. Industry publications and PR: Mentions in authoritative tech and marketing publications provide the external consensus AI models use to confirm your brand's positioning and expertise.

The key operational requirement is consistency. AI models skip citing brands with conflicting data across sources. If your pricing on your website differs from your G2 profile, or your ICP description on LinkedIn contradicts your homepage, you are actively suppressing your own citation rate. For a full breakdown of how the major platforms select and prioritize sources, see our guide on AI citation patterns.

Tying AI visibility to closed-won revenue in your CRM

The measurement framework above gives you leading indicators and a richer picture of AI influence. The board wants pipeline numbers, and the CFO wants ROI.

Here is how AI-influenced revenue flows into your existing CRM when you set up attribution correctly. Start UTM tagging on day one. AI referral traffic that does carry referrer data, typically from Perplexity and some Gemini touchpoints, should be captured with a source parameter (for example, utm_source=perplexity as a manual tag on tracked links) so it flows into HubSpot or Salesforce as a distinct source. Layer this alongside your "How did you hear about us?" field data and your Gong call flags to build a composite picture of AI-influenced opportunities.

For the pipeline math, the conversion data across multiple independent studies is consistent:

The reason for this premium is structural. By the time a buyer clicks through from an AI recommendation, the LLM has already done the research, shortlisted the vendors, and in many cases framed your product favorably for the buyer's specific context. Demand capture is almost complete before the first visit.

This is consistent with what we see directly with clients. One B2B SaaS company went from 550 AI-referred trials per month to over 2,300 in approximately four weeks after implementing a structured AEO content program. Another improved ChatGPT referrals by 29% and closed five new paying customers in the first month.

Traditional SEO typically captures bottom-of-funnel intent on Google. AEO often captures the research and shortlisting phase happening inside LLMs. For a full breakdown of how both channels work together rather than competing, see our AEO definition and mechanics guide and our SEO agency service page. You can also review our AEO pricing options if you want to understand the investment and scope before a conversation.

Ready to see exactly where your brand appears versus competitors across 20 to 30 buyer-intent queries? Request a custom AI Search Visibility Audit from Discovered Labs and we will show you the citation gaps costing you pipeline and the content roadmap to close them.

Specific FAQs

How long does it take to see pipeline impact from AEO?

Initial AI citations typically appear within 1 to 2 weeks of publishing CITABLE-optimized content. Measurable pipeline impact (AI-referred MQLs tracked in Salesforce with attribution) typically requires 90 to 120 days of consistent publishing and ongoing third-party validation building.

What percentage of AI-driven visits are currently invisible to GA4?

RSA Creative Studio's analysis confirms that AI-enabled browsers frequently fail to pass referrer data, causing a large proportion of AI-driven sessions to appear as direct or "(not set)" in GA4. Perplexity passes some referrer data, while ChatGPT in-app browsing generally does not.

How many buyer-intent queries should I test in an AI visibility audit?

Start with 20 to 30 queries covering your top use cases, buyer personas, and competitive comparisons. Our competitive AEO audit guide suggests expanding your query set once you have your initial baseline and topical cluster map.

Can self-reported attribution data be used to build a board-level ROI case?

Yes. Combine "How did you hear about us?" form responses with Gong call transcript flags and map flagged opportunities to closed-won revenue in Salesforce to calculate AI-sourced pipeline. This is the same methodology Refine Labs uses to justify dark funnel investment to CFOs and boards.

Does AEO replace traditional SEO?

No. SEO tends to capture bottom-of-funnel demand on Google, while AEO tends to capture the research and shortlisting phase happening inside LLMs before buyers reach Google. B2B SaaS companies with strong Google SEO but no AEO program are visible at the point of demand capture but invisible during demand creation.

Key terms glossary

Answer Engine Optimization (AEO)
The practice of structuring content so that AI-powered platforms including Google AI Overviews, ChatGPT, Perplexity, Claude, and Gemini can extract, cite, and attribute your brand as a trusted source. Unlike traditional SEO's focus on ranking individual pages, AEO targets extractable answers and brand citations across AI-generated responses.

Dark funnel
The buyer touchpoints that attribution software cannot track, including social media, peer conversations and community discussions, podcasts, and now AI search sessions inside LLMs. These touchpoints make up 75% or more of the path to purchase for most B2B buyers.

Self-reported attribution
A method of collecting attribution data by directly asking prospects how they discovered your company through "How did you hear about us?" form fields, post-demo surveys, or sales discovery call questions. It is a valuable method for capturing AI-influenced demand that tracking scripts cannot see, though it should be used alongside other attribution data sources to account for recall bias and incomplete responses.

Share of voice (AEO)
Your brand's citation presence in AI-generated responses compared to competitors, expressed as a percentage of total category citations across tracked platforms and query clusters. A share of voice of 30% means your brand appears in 30% of AI responses to your tracked buyer-intent queries.

Retrieval-Augmented Generation (RAG)
A technique where an LLM retrieves information from an external knowledge source to ground its answers in current, factual data rather than relying solely on training data. Content structured in 200 to 400 word blocks with clear headings, tables, and lists aligns with how RAG systems chunk and index content for retrieval.