AI-attributed pipeline is the dollar value of opportunities whose earliest touch traces back to an AI answer engine mention. It is the metric that turns AEO from a marketing curiosity into a revenue line item, and it is the one number that survives a CFO conversation about whether AI search is worth funding.
What AI-attributed pipeline actually measures
AI-attributed pipeline is not a claim that a Perplexity thread caused a closed-won deal. It is a claim that at the time a specific contact first entered your CRM, one or more AI engines were mentioning your brand for the questions that contact was most likely asking. The strength of that claim depends on the join. The join depends on your data model. And the data model is where most attribution programs quietly die.
A useful mental model: AEO attribution is a two-hop join. The first hop connects a prompt run to a mention. The second hop connects a mention window to a CRM first-touch. If both hops are clean, you have a defensible influenced-pipeline number. If either hop is missing, you have a story.
Why classic attribution tools cannot see AI search
GA4 lumps most AI-answer traffic into Direct. Chat products send referrer headers that browsers often strip. Perplexity sometimes attaches source parameters, sometimes does not. ChatGPT web frequently sends users through an intermediate redirect. The result is that even when a buyer does click through from an AI answer, your analytics stack usually cannot tell you it happened.
This is why single-touch, click-based attribution breaks for AEO. You need a model that treats AI mentions as an off-site influence layer, not as a click source. Every serious AEO attribution program leans on the mention-window join rather than the referrer signal. Our Attribute module describes the mechanics in more detail.
The mention-window join, explained
Here is the mechanic that actually works.
- Your measurement layer runs a fixed prompt battery every week. For each engine and prompt, it records whether your brand was mentioned, and how prominently.
- Your CRM records every new contact with a first-touch timestamp, a source, and a set of attributes (industry, company size, role) that let you infer which prompts they were likely asking.
- For each new contact, you look back N days (typically 30 to 60). Any prompt-mention pairs that overlap that window and match the contact profile count as an AI-influenced touch.
- You mark the contact, and any opportunity that contact is part of, as AI-influenced. Sum the opportunity ARR to get AI-influenced pipeline.
This is a correlation model, not a causal one. Nobody can prove that a specific Perplexity thread produced a specific meeting request. But if 40 percent of your new pipeline every quarter comes from accounts whose first-touch window overlapped with AI mentions of your brand, that is a signal your CFO will care about.
Sourced pipeline vs influenced pipeline
Two metrics, not one, is the pragmatic answer.
- AI-influenced pipeline is the union of all opportunities whose first-touch window overlapped an AI mention. This is the big, defensible number.
- AI-sourced pipeline is the subset where AI was the only inbound signal in the window, or where the contact self-reported "found you via AI search" in a form field. This is the small, conservative number.
Report both. Influenced tells the growth story. Sourced tells the incremental story. If you report only influenced, your CFO will assume you are triple-counting. If you report only sourced, you will understate the program by an order of magnitude.
The prerequisite: clean first-touch capture
None of this works if your marketing automation platform does not have a reliable concept of first touch across every inbound source. This is the least glamorous part of an AEO program, and the one that most often gets skipped in the excitement of picking a measurement tool. Do the plumbing first.
Concretely: every form on every page should stamp UTM parameters into hidden fields, your MAP should persist those into the contact record on creation and never overwrite them on subsequent submissions, and your CRM sync should carry them into the account and opportunity objects. If any of those three links is broken, your attribution stops at the front door.
Handling multi-buyer accounts
In B2B, deals close on accounts, not contacts. A single opportunity often has three to seven contacts, each with their own first-touch. That means an "AI-influenced" flag needs to be an account or opportunity attribute, not a contact attribute. The rule that works in practice: if any contact on the account had an AI-mention overlap in their first-touch window, the account counts as influenced.
This will overcount slightly. That is fine, because you are also reporting the more conservative sourced number alongside it. What matters is that the definition is stable across quarters so the trend is credible.
Reporting AI-attributed pipeline to executives
The one-slide report is: AI-influenced pipeline this quarter, versus last quarter, versus the same quarter last year, alongside AI-sourced pipeline in the same three cuts. Add a fourth line for weighted share of voice so the leading indicator sits next to the lagging one.
If share of voice is up and influenced pipeline is up, the program is working. If share of voice is up but pipeline is flat, your content is winning citations but not the accounts that matter. If share of voice is flat but pipeline is up, you have a lucky quarter and should not budget on it. The chart tells the story faster than any narrative.
Common mistakes
- Reporting a single "AI-sourced" number. Under-reports the program by 5 to 10 times and dies at budget review.
- Attributing on referrer only. Misses most AI traffic because referrer headers are unreliable.
- Using inconsistent first-touch windows. Some quarters 30 days, some quarters 90. The trend line becomes meaningless.
- Ignoring account-level rollup. Contact-level attribution understates enterprise deals.
- Attributing without a share of voice reference. Pipeline moves for many reasons. Without the leading indicator, you cannot tell if the pipeline change is AEO or something else.
A 60-day rollout for the attribution loop
If your measurement layer is already running, wiring attribution is a 60-day project.
- Days 1 to 15. Audit first-touch capture end to end. Fix any broken UTM or MAP handoff.
- Days 16 to 30. Define influenced and sourced pipeline in the CRM. Stand up the mention-window join.
- Days 31 to 45. Backfill the last two quarters so you have a trend, not a snapshot.
- Days 46 to 60. Ship the executive report. Weekly cadence for the ops team, quarterly cadence for the board.
Where to go next
Attribution is the loop that unlocks the budget. Once your CFO sees a defensible AI-influenced pipeline number moving in the same direction as your share of voice number, the program is funded for the year. If you want a working model for the measurement side that feeds this loop, start with our share of voice guide.