By maxaeo.ai ｜ Published 2026-10-13 ｜ Updated 2026-10-13
An AI search attribution window for B2B defines how long an AI-related touchpoint remains eligible for credit before a target event, such as a demo request, opportunity creation, or closed-won deal. The window should come from observed conversion lag—not a platform default or a generic SaaS benchmark.
For long-cycle purchases, the best approach is to set one operating window for routine reporting and a longer shadow window for sensitivity analysis.
What Is an AI Search Attribution Window?
An AI search attribution window is the lookback period during which a ChatGPT, Perplexity, Gemini, or other answer-engine interaction may be associated with a later B2B outcome. It determines which AI referrals, self-reported discoveries, and account-level signals qualify as potential influence.
The window is not the attribution model itself. A model decides how much credit each touch receives; the window decides which touches are eligible to receive any credit.
For example, if an account receives an identifiable Perplexity referral 140 days before opportunity creation, a 90-day window excludes it. A 180-day window includes it, but the reporting model must still determine whether the visit was sourced, assisted, or merely correlated with the opportunity.
Why Do Default Attribution Windows Fail for B2B AI Search?
Default windows often end before a B2B buying committee completes research, internal validation, procurement, and security review. AI discovery can also be separated from conversion by branded searches, direct visits, colleague sharing, or a different stakeholder entering the CRM.
Google Ads applies a 30-day default to many click-through conversions and supports windows of up to 90 days for relevant Search and Display conversion sources. GA4 documentation also describes a 90-day default key-event lookback setting for “all other conversion events.” These configurations are useful platform settings, but they are not evidence that 30 or 90 days fits an enterprise sales journey. (support.google.com)
This mismatch creates two risks:
- Too short: Early AI discovery disappears, and later direct or branded traffic receives excessive credit.
- Too long: Weak, unrelated touches accumulate around every opportunity, inflating influenced pipeline.
Which Conversion Event Should the Window End At?
The correct endpoint depends on the business question. Do not use one window interchangeably for lead generation, pipeline creation, and revenue reporting.
| Reporting question | Window endpoint | Appropriate interpretation |
|---|---|---|
| Did AI search generate an inquiry? | Demo, trial, or contact event | AI-sourced or AI-assisted lead |
| Did AI discovery contribute to pipeline? | Opportunity creation | Account-level pipeline influence |
| Did AI help advance an active deal? | Stage change or sales milestone | Deal acceleration evidence |
| Was AI present in a winning journey? | Closed-won date | Revenue influence, not automatic causation |
Opportunity creation is usually the cleanest primary endpoint because it separates pre-pipeline demand creation from sales-cycle execution. Closed-won analysis should remain a secondary view, especially when contracts take months to progress after qualification.
The broader AI search attribution model for enterprise SaaS explains how to keep sourced, assisted, self-reported, and correlated outcomes separate.
How Should B2B Teams Calculate the Window?
Use a lag distribution rather than an average. Averages can be distorted by a few unusually long deals, while a median alone may exclude a large share of legitimate journeys.
- Choose one endpoint. Start with opportunity creation or a qualified demo rather than mixing multiple conversions.
- Collect eligible evidence. Include identifiable AI referrals, buyer-declared AI discovery, and dated sales notes. Keep aggregate brand visibility separate.
- Calculate elapsed days. Measure the interval from the earliest credible AI touch to the selected endpoint.
- Segment the data. Separate product-led, SMB, mid-market, and enterprise journeys where their sales motions differ.
- Calculate P50, P80, and P95. These percentiles show where half, 80%, and 95% of observed conversion lags fall.
- Backtest candidate windows. Compare the opportunities retained and the ambiguous touches added at 90, 180, 270, and 365 days.
A practical calculation is:
Eligible lag = target-event date − earliest credible AI-touch date
Only records with a credible account or contact connection should enter this calculation. A monitored prompt showing that an AI engine mentioned the brand is market-level evidence, not proof that a particular buyer saw the answer.
The 80/95 Dual-Window Framework
The 80/95 framework uses the 80th-percentile lag as the operating window and the 95th-percentile lag as a shadow window. This preserves most observable journeys in routine reporting while exposing long-tail influence without automatically adding it to attributed pipeline.
Consider an illustrative cohort of 60 opportunities:
| Lag metric | Days from AI touch to opportunity |
|---|---|
| P50 | 74 days |
| P80 | 143 days |
| P95 | 272 days |
The team could round P80 to a 150-day operating window and P95 to a 270-day shadow window. Opportunities between days 151 and 270 would be reviewed as long-tail influence rather than included automatically.
These figures are a worked example, not a benchmark. The information gain comes from comparing the marginal value of a longer window: if extending 150 days to 270 days adds three credible opportunities but dozens of weak associations, the shorter operating window is more defensible.
How Should Buying Committees and Invisible Touches Be Handled?
B2B AI attribution should operate at the account level because the person who sees an AI recommendation may not be the person who books the demo. Contact-level records should roll up into an account timeline without overwriting the original acquisition source.
Use evidence tiers:
- Observed: An AI referrer or tagged link connects directly to an identified conversion.
- Declared: A buyer names an AI engine or describes the prompt used during research.
- Assisted: A known AI interaction precedes a later conversion through another channel.
- Contextual: Brand visibility changed for relevant prompts, but no person-level connection exists.
Observed and declared evidence can support pipeline reporting. Contextual evidence should guide hypotheses and content decisions, not receive direct revenue credit.
A practical implementation is detailed in the guide to tracking conversational search touchpoints in CRM.
How Can the Selected Window Be Validated?
A defensible window must survive backtesting. Recalculate the previous two or more quarters using several candidate windows, then compare retained opportunities, duplicate credit, evidence quality, and unexplained direct conversions.
Track four validation metrics:
- Percentage of credible AI-influenced opportunities retained
- Number of weak or context-only touches admitted
- Change in pipeline credit when the window expands
- Variation by segment, region, prompt cluster, and contract type
Freeze the selected window for a reporting period. Changing it after every campaign or executive review makes comparisons unreliable. Recalibrate only after enough new opportunities have matured to alter the lag distribution.
Prompt-level segmentation can be built with a B2B generative search prompt-cluster framework, while downstream performance can be checked through ChatGPT-to-demo conversion measurement.
Frequently Asked Questions
Is 90 Days Long Enough for B2B AI Attribution?
Only if the company’s observed lag data supports it. A 90-day window may fit a short trial-led journey but exclude early discovery in enterprise deals. Compare 90 days with longer windows before adopting it.
Should the Window Run Until Closed-Won Revenue?
Use closed-won as a secondary influence view. Opportunity creation is usually a cleaner endpoint for measuring demand creation because post-opportunity delays can reflect procurement and sales execution rather than marketing influence.
Can AI Brand Mentions Start an Attribution Window?
Not by themselves. A monitored mention shows that a brand was visible for a prompt, but it does not prove that a buyer saw the response. Start person-level attribution with observable or declared evidence.
How Often Should the Window Be Recalculated?
Review it after a meaningful volume of opportunities has completed the measured journey. Recalculation by quarter may be practical, but the correct cadence depends on deal volume and cycle length.
Final Recommendation
Set the operating window at the P80 lag for each major sales segment, maintain P95 as a shadow view, and report evidence strength alongside pipeline credit. This makes uncertainty visible while preventing both early-touch deletion and unlimited influence claims.
Use a free AI visibility audit to identify the prompts, engines, mentions, citations, and competitors that should feed the visibility layer of your attribution system.