LLM Revenue Attribution

Most B2B marketers can tell you how much traffic they get from Google organic. Almost none can tell you how much pipeline came from ChatGPT. That gap — the distance between where demand is now being created and where our measurement tools still look — is the category this page defines.

LLM revenue attribution and AEO pipeline attribution describe the same practice. This page is the broader category overview, written for a reader who arrived searching for "LLM attribution" and wants the full picture — what it is, why it's hard, and how Lantern solves it.

The category in one paragraph

Buyers increasingly discover B2B software inside LLM chat windows. They ask ChatGPT or Claude for a recommendation. The model answers. Some buyers click through; most copy URLs. Weeks later, the buyer converts — but the attribution trail is gone. Standard analytics route the session into Direct. CRM Original Source reports Organic Search or Direct. The CMO cannot report LLM-influenced pipeline without a dedicated measurement layer. LLM revenue attribution is that layer.

Why LLM attribution is categorically harder

Standard web attribution was designed for a world where buyers clicked. LLM-mediated demand breaks three of its core assumptions:

Assumption 1 (broken): every session has a referrer

ChatGPT and Claude answers are typically rendered inside chat interfaces that open links in new tabs without passing a reliable HTTP referrer. Mobile app citations route through the iOS/Android system browser — no referrer. Copy-paste is worse — no signal at all. Result: 70.6% of AI-referred sessions land in GA4 as Direct (Loamly, Q1 2026). Direct is the analytics dumpster — the channel Direct = We Don't Know.

Assumption 2 (broken): discovery channels produce clicks

Much of LLM influence happens without a click at all. A user asks ChatGPT "which tool is best for X?" The answer names three tools. The user remembers the names, types them into a new Google search later, and clicks a branded result. Your site sees the branded organic click; the LLM touchpoint that shortlisted you leaves no trace in your analytics.

Assumption 3 (broken): last-click is a reasonable approximation

For discovery channels like LLMs, last-click attribution is structurally wrong. The buyer who first heard about you on Claude returns via branded search. Last-click credits the branded search. The Claude touchpoint that created the demand gets zero credit, looks unprofitable, budget gets cut, pipeline disappears over the next 2–3 quarters.

These three breaks compound. By the time a CMO asks "is LLM optimization driving pipeline?" their analytics tell them "Direct is up and you have no idea why." That is the category-defining measurement problem.

The CRM-native solution

LLM revenue attribution solves the problem not at the analytics layer, but at the CRM layer. Because CRM contacts survive longer than analytics sessions, and because the CRM can combine multiple signals across weeks or months, CRM-native instrumentation is the only place the full attribution chain survives.

The four-signal pipeline:

Signal 1. Self-reported source (highest confidence)

A free-text field on every inbound form: "How did you first hear about us?" Responses containing ChatGPT, GPT, Perplexity, Claude, Gemini, or "AI search" are LLM touchpoints. This is the cheapest, highest-signal input in any B2B SaaS stack. Under-used, critical.

Signal 2. UTM-tagged landing sessions (medium-high confidence)

UTMs on links you can influence — your own comparison pages, Reddit answers, LinkedIn replies that LLMs cite. utm_source=ai-search, utm_medium=citation, utm_campaign=<engine>-<content>. Produces a clean first-touch record when the buyer clicks through rather than copy-pastes.

Signal 3. Citation-proximity (medium confidence)

Direct sessions that land on a cited URL within minutes of a known citation serve. Lower-confidence (typically 40% credit) because it's probability, not proof. Useful for cases where the buyer skipped both the UTM and the self-report.

Signal 4. Prompt-level enrichment (inferred)

When an AEO monitoring system is running in parallel, individual citations can be tagged to specific prompts. Over time, the pattern of "which prompts produce pipeline" becomes a first-party dataset — the most valuable output of a mature LLM attribution program.

Each signal writes a custom property on the CRM contact record. Properties propagate to the primary company and the deal. Monthly CFO memo reports LLM-influenced pipeline (stage-weighted) and LLM-influenced closed-won (hard number). CAC payback math uses the hard number.

The HubSpot path

HubSpot is the V1 CRM integration for Lantern. The HubSpot implementation uses five custom contact properties (aeo_first_touch_engine, aeo_first_touch_prompt, aeo_first_touch_timestamp, aeo_self_reported_source, aeo_touchpoint_count), three workflows (parse self-reported text into engine; propagate properties from contact to company to deal; calculate stage-weighted pipeline on deal-stage change), and a standard HubSpot list filter for the monthly report.

Full implementation details, including JSON property definitions and workflow specs, are on the AI Search Attribution in HubSpot page.

The Salesforce path

Salesforce is Lantern's V1.5 integration, shipping months 4–6 from V1 launch. The Salesforce implementation follows the same pattern:

• Custom fields on Lead and Contact: AEO_First_Touch_Engine__c, AEO_First_Touch_Prompt__c, AEO_First_Touch_Timestamp__c, AEO_Self_Reported_Source__c, AEO_Touchpoint_Count__c.
• Custom fields on Opportunity: AEO_Influenced__c (checkbox), AEO_First_Touch_Engine__c, AEO_First_Touch_Prompt__c, AEO_Attributed_Pipeline__c.
• Process Builder or Flow: parse self-reported source into engine; copy from Lead to Contact at conversion; copy from Contact to Opportunity on opp creation; calculate stage-weighted pipeline on stage change.
• Reports and Dashboards: a SOQL report filtered on AEO_Influenced__c = TRUE with a summary on AEO_Attributed_Pipeline__c.

Salesforce-native customers typically have a longer sales cycle (enterprise) and benefit from 180-day lookback windows rather than the 90-day default.

What makes a number CFO-safe

The discipline exists because CMOs need CFO-safe numbers. A number survives a CFO review when it has four properties:

1. It is in dollars, not citations. "$X of LLM-influenced pipeline this quarter" > "share of voice up 12%."
2. It exposes its assumptions. Stage-weighted probability disclosed. Lookback window disclosed. Self-reported match rules disclosed. Proximity-signal weights disclosed.
3. It lives in the CRM, not a vendor dashboard. The CFO's Ops team can pull the underlying list and see the same number the CMO is reporting.
4. It benchmarks against the rest of the marketing mix on the same denominator. LLM CAC payback vs paid CAC payback vs SEO CAC payback, in months, on the same ACV assumption.

Tools that don't produce CFO-safe numbers tend to lose renewal reviews regardless of how good their dashboards look.

Why monitoring alone isn't enough

A standalone monitoring tool — the category occupied by Profound, Scrunch, AthenaHQ, Peec AI — tells you where you're cited. It doesn't connect those citations to deals in your CRM. The difference is categorical, not quality:

• Monitoring output: share of voice (%), citation count (#), prompt coverage. Useful for the SEO lead running the program.
• Attribution output: LLM-influenced pipeline ($), LLM CAC payback (months), share of marketing-sourced pipeline from LLMs (%). Useful for the CMO signing the renewal.

The audience brief captures this in a single phrase: monitoring, not attribution. It is the one-line takedown of every AEO dashboard that can't close the loop to revenue.

The role of Lantern

Lantern is LLM revenue attribution for B2B SaaS. We ship the full chain: real-time citation monitoring across ChatGPT, Claude, Perplexity, Gemini, and AI Overviews; session and contact resolution; CRM-native property writes on HubSpot (V1) and Salesforce (V1.5); stage-weighted roll-up using the customer's own deal stage configuration; first-touch CFO PDF and multi-touch operational dashboard.

Two tiers: $99/mo for the self-serve install on HubSpot, or Enterprise for Salesforce integration, custom schemas, field-level encryption, and co-built CFO memo templates. No per-prompt math. No per-seat math.

Where to go next

• AEO Pipeline Attribution — the franchise definition and the six-step citation-to-closed-won chain.
• AEO Attribution Methodology — the math: stage-weighted probability, first-touch vs multi-touch, lookback windows.
• AI Search Attribution in HubSpot — the implementation guide with JSON property definitions.
• ChatGPT Citation to HubSpot Deal — one deal traced end to end.
• The CFO's Guide to AEO Budget Defense — the renewal-review memo template.
• AEO ROI Calculator — run your own numbers.