How-To Guide

Measuring the Dark Funnel From AI Search

A practical playbook for surfacing the AI-driven portion of the dark funnel. Survey methods, MMM specification, incrementality calibration, and the metrics that hold up in finance review.

By Ramanath, CTO & Co-Founder at Presenc AI · Last updated: May 17, 2026

Why the Dark Funnel Grew When AI Search Arrived

The dark funnel, the portion of the buyer journey that happens outside trackable channels, has always included offline conversation, podcast listening, and private community discussion. AI search added a new and rapidly growing layer: a research phase that previously happened on Google and now happens on ChatGPT, Claude, and Perplexity, with no observable touchpoint until the buyer arrives at the brand site as "direct."

The result is that direct traffic has become both more material to revenue and less interpretable as a metric. Measuring the dark funnel from AI search is the discipline of breaking that direct-traffic black box apart.

Step 1: Run Continuous Post-Conversion Surveys

The simplest and most underused dark-funnel measurement is asking buyers how they heard about the brand. Add a "how did you first hear about us" question to your post-conversion survey, with AI assistants as a named option, and the option to specify which assistant.

Run continuously, not as a one-off. Report weekly aggregates of the "AI assistant" share of converted revenue. Even with 20 to 40 percent response rates, the directional signal is reliable enough to anchor executive conversations and to validate or challenge model-based attribution.

Step 2: Decompose Direct Traffic

Look at direct traffic share over time. Most brands see direct rising as a percentage of total traffic over the 2024 to 2026 window. If your direct share has grown by 10 to 30 percentage points and paid acquisition is flat, the dark funnel has expanded, and AI search is the most likely cause.

Cross-reference direct traffic growth with the survey "how did you hear" data. If "AI assistant" responses are growing in proportion to direct traffic, you have triangulated the AI portion of the dark funnel's growth.

Step 3: Add AI Visibility to MMM

MMM is the only attribution framework that can value dark-funnel channels. Add a weekly AI visibility series (LLM share of voice) as a media-equivalent variable, refit the model, and inspect the new decomposition. The expected pattern is that the base demand contribution falls (AI was hiding there) and the new AI variable picks up a meaningful share of revenue.

For brands without MMM, the interim alternative is a before-and-after analysis of branded search and direct traffic around step-changes in AI visibility (a new Wikipedia article, a major PR feature). Noisier but directionally useful.

Step 4: Look for Halo Effects

The halo effect of AI search shows up as branded search and direct traffic lift in the one- to four-week window after AI visibility movement. Plot weekly AI visibility against branded search volume; if the two series move together with a short lag, halo is present, and the magnitude of the relationship is the cross-channel coefficient that MMM should be capturing.

Step 5: Run Geographic Lift Tests for Causal Anchoring

The triangulation step. A geographic lift test on AI visibility inputs (pause PR and content syndication in matched regions for eight to twelve weeks) produces a causal estimate of the AI dark funnel's contribution. Use this to calibrate the MMM and to anchor the survey-based estimates in causal evidence.

Step 6: Build a Composite Dark-Funnel KPI

Combine the four signals (server log floor where available, survey self-attribution, MMM contribution, lift-test calibration) into a single composite estimate of AI dark-funnel contribution. The composite is wider in confidence than any single signal but is far more defensible in board reporting because it triangulates across independent methods.

Report the composite as a range with point estimate, not a single number. "AI search contributed an estimated 12 to 18 percent of converted revenue, with point estimate 14 percent, supported by MMM, lift test, and survey triangulation" lands credibly. "AI search contributed exactly $X" sounds suspicious because it is.

How Presenc AI Helps

Presenc AI provides the weekly AI visibility series that anchors the MMM specification, the halo analysis, and the lift test design. The platform also tracks where in the buyer journey AI assistants are surfacing the brand (top-of-funnel category queries vs decision-stage comparison queries), which sharpens the dark-funnel measurement by distinguishing AI-driven demand creation from AI-driven demand capture.

Frequently Asked Questions

In B2B SaaS, surveys typically attribute 8 to 18 percent of converted revenue to AI assistant influence as of 2026, with the share growing. Consumer categories vary more widely depending on category and AI assistant adoption. The absolute number is less important than the trajectory; AI dark funnel share has been rising in every category that measures it.
No. Direct traffic is one symptom. The dark funnel is the upstream consideration activity that produces direct traffic, including AI assistant research, podcast listening, private community discussion, and offline word of mouth. Direct traffic is the visible end of an invisible process; the dark funnel is the process itself.
Partially. Surveys and direct traffic decomposition give directional estimates. Full causal measurement requires either MMM with an AI visibility variable or geographic lift testing on AI visibility inputs. The interim methods are useful while building toward the rigorous methods, not replacements for them.
Triangulation. Combine server log evidence (lower bound), survey self-attribution (directional), and MMM contribution (statistically integrated). Present as a range with the components, not as a single point estimate. CFOs and boards are more receptive to honest triangulation than to confident-sounding single-method numbers that fall apart under questioning.

Track Your AI Visibility

See how your brand appears across ChatGPT, Claude, Perplexity, and other AI platforms. Start monitoring today.