The attribution gap: B2B buyers increasingly discover vendors through AI tools, but that discovery rarely shows up in analytics. The traffic arrives as direct. The buyer signs a deal months later and nobody knows where the relationship started. This guide covers how to close that attribution gap across five layers: form fields, dark traffic analysis, Perplexity referrers, CRM tagging, and AI share of voice correlation.
Pipeline attribution for AI search is harder than organic search attribution and easier than most teams think. The hard part is that AI tools do not pass referrer headers in most cases, so the standard UTM-based attribution that works for Google and LinkedIn does not work for ChatGPT or Claude. The easy part is that buyers who discovered you through AI have distinctive behavioral characteristics you can use as a proxy, and many of them will tell you directly if you ask.
For context on how B2B buyers use AI tools during vendor research, see the B2B buyer AI search behavior guide. For the CMO-level framework for reporting on AI-influenced pipeline, see the CMO operational playbook. This guide is the implementation layer: exactly what to set up and how.
Why AI Traffic Is Invisible in Most Analytics
When a user clicks a citation link in ChatGPT, Claude, or Gemini, the browser typically sends no HTTP referrer header to your server. Your analytics platform sees an unattributed session and files it as direct. This is the dark traffic problem that makes AI-sourced traffic conversion data so difficult to isolate: the sessions are converting at higher rates than organic, but they are invisible in source attribution.
Perplexity is the exception. Perplexity consistently passes a referrer of perplexity.ai, so Perplexity-sourced traffic is detectable in analytics under the referral channel. You can verify this right now by checking your analytics for sessions with referrer containing "perplexity".
| AI Platform | Referrer Behavior | Analytics Visibility |
|---|---|---|
| ChatGPT / OpenAI | No referrer passed | Shows as direct traffic |
| Claude / Anthropic | No referrer passed | Shows as direct traffic |
| Perplexity | Passes perplexity.ai referrer | Visible in referral channel |
| Google AI Overview | Google referrer (varies) | May appear as organic/Google |
| Gemini (standalone) | No referrer passed | Shows as direct traffic |
Part 1: Form Attribution Fields
Self-report is the most direct attribution method and more reliable than most marketers assume. Buyers who discovered you through an AI tool often have a clear memory of it, especially since AI search is still a relatively new behavior that stands out in their mental model.
Add a "How did you hear about us?" dropdown to every demo and contact form
Make AI tools explicit named options, not a catch-all "other" bucket. Buyers who discovered you through ChatGPT will not select "social media" or "other" if the form does not name AI tools specifically.
Effective option list for the HTDYHAU dropdown:
options = [
"ChatGPT or Claude",
"Perplexity",
"Google AI Overview",
"AI search tool (other)",
"Google / organic search",
"LinkedIn",
"Industry publication or blog",
"Event or webinar",
"Colleague or peer referral",
"Customer referral",
"Other"
]For shorter forms where a dropdown is too heavy, add a single checkbox: "Did an AI tool help you find or learn about us?" with a text box for details. This captures the binary signal without adding form friction. Map the responses to a dedicated CRM field, not a notes field, so you can query it.
Part 2: Dark Traffic Behavioral Signals
Not every buyer will fill in the attribution field accurately. For sessions that arrive as direct traffic but show ICP-characteristic behavior, you can use behavioral signals as a high-confidence proxy for AI referral.
- Entry page is a deep solution page, not the homepage. Organic search lands on any indexed page. AI citations link to specific resources. If direct traffic is landing on your pricing page or a specific use-case page, AI is a likely source.
- Session engagement is high from the first visit. AI-referred buyers have already done research before arriving. They exhibit shorter time-to-contact and lower bounce rates than cold direct traffic.
- ICP firmographic match. If your ICP is Series A-C SaaS in the US and the direct traffic is from that firmographic profile, it is disproportionately likely to be AI-referred versus truly direct.
- Multiple visits in a short window. Buyers doing AI-assisted research often revisit after prompting the AI again. A pattern of 3-5 visits in 72 hours with no search attribution is consistent with AI-assisted consideration.
- Form or demo request on first or second visit. AI-referred buyers arrive with more context and convert faster. First-visit demo requests from direct traffic are a strong AI-referral proxy.
Segment your direct traffic in analytics by entry page, session depth, and engagement rate. The "high-quality direct" segment will correlate strongly with your AI SoV movement over time. This pattern is documented in the AI search conversion behavior data.
Part 3: Perplexity Referrer Tracking
Because Perplexity passes a referrer, you can track it explicitly. Set up a dedicated analytics segment or UTM override to capture all sessions where the referrer matches "perplexity.ai". This gives you a clean, direct view of Perplexity-attributed traffic and its conversion behavior.
For a more precise view, check your Perplexity traffic against your AI share of voice score on Perplexity specifically. A higher Perplexity SoV score should correspond to higher Perplexity referral traffic. When the two move together over 60-90 days, you have a causal signal rather than just a correlation. The competitive AI brand audit framework covers how to benchmark your Perplexity visibility specifically.
Part 4: CRM Setup for AI Attribution
Your form attribution data only becomes useful if it reaches the CRM in a queryable way. Three things to configure:
Add a dedicated "AI Discovery Source" field to your Lead and Contact objects
Do not rely on the existing Lead Source or Campaign field. Those are already mapped to other attribution models. A dedicated field keeps AI attribution clean and queryable without disturbing historical data.
Train sales to ask and log it on discovery calls
Add "How did you first hear about us, was AI involved in your research?" as a standard discovery question. Log the answer in the AI Discovery Source field. Sales confirmation correlates the form self-report with a human verification, making the attribution more credible to finance.
Build a quarterly AI-influenced pipeline report
A simple CRM report filtered to records where AI Discovery Source is non-null, showing total pipeline value and closed-won revenue. This is the number you take to the CMO and board. Keep it a separate line item from total pipeline, not merged into it.
Part 5: AI SoV Correlation
The final layer is correlating your AI share of voice movement with pipeline outcomes over time. This is the strategic proof layer that turns tactical attribution into a business case for continued investment.
Build a quarterly correlation model with three variables: monthly AI SoV score, monthly AI-attributed form submissions, and monthly dark traffic volume. When all three move in the same direction for two or more consecutive quarters, you have a credible causal claim rather than just a correlation.
This is the data behind the board presentation framework in the CMO playbook and the ROI model in the GEO ROI framework. Start with a 3-month baseline before any major GEO activities, so you can show a before/after comparison.
Frequently Asked Questions
Why does most AI search traffic show up as direct traffic in analytics?
ChatGPT, Claude, and Gemini do not pass referrer headers when users click citation links. The browser sends no source information, so analytics files the session as direct traffic. Perplexity is the exception and passes a perplexity.ai referrer consistently, so Perplexity-sourced traffic is visible under the referral channel.
What form field options should I add for AI attribution?
Add explicit named options: "ChatGPT or Claude," "Perplexity," "Google AI Overview," "AI search tool (other)" -- separate from organic search, LinkedIn, or event options. Buyers who discovered you through AI will not self-select into "other" if the form does not name AI tools specifically.
How do I correlate AI share of voice with pipeline in a way that's credible to finance?
Track three variables monthly: AI SoV score, AI-attributed form submissions, and dark traffic volume with high engagement. Run as a quarterly analysis to reduce noise. When all three move in the same direction for two consecutive quarters, you have a credible causal claim. Add deal-level CRM confirmation from sales discovery calls to strengthen it.
Jeevan AI connects AI citation tracking with attribution data so you can show revenue impact alongside visibility metrics.