AI Traffic Conversion Attribution: Models and Measurement

Standard attribution models were designed for linear customer journeys with predictable touchpoints. AI-driven traffic does not follow those patterns. A user who discovers your content through a ChatGPT citation, reads it, leaves, returns three days later via a branded Google search, and converts on the second visit has an attribution path that standard models handle poorly.

Last-click attribution credits the branded search entirely. First-click credits ChatGPT. Linear attribution spreads credit across both but also across any intermediate touchpoints (social, email, direct) that may have been incidental. None of these models reflect the actual causal role each channel played.

AI referral traffic introduces two specific attribution distortions:

Distortion 1: First-Touch Underweighting. AI platforms are research-starting channels. Users land on a page from ChatGPT or Perplexity, get a preliminary answer, and then leave. If they convert later through a direct visit or a branded search, the AI source receives zero credit in last-click models. Since AI traffic is frequently the first meaningful interaction with a brand, this systematically undercounts AI's contribution.

Data from publisher analytics in early 2025 showed that for content-driven conversion paths involving an AI referral, the AI touchpoint was the first interaction in over 70 percent of paths but the last interaction in under 15 percent. Last-click attribution would credit AI only 15 percent of the time.

Distortion 2: Attribution Window Mismatch. GA4's default attribution window is 30 days for last-click and 90 days for data-driven models. AI platform users tend to have longer research cycles because they are comparing information from multiple sources. A 30-day window may cut off valid conversion paths that began with an AI referral 40 days earlier. The result is artificially low attributed conversions for AI sources.

The data-driven attribution (DDA) model in GA4 is the best starting point for AI traffic. DDA uses algorithmic credit distribution based on the incremental impact of each touchpoint. Unlike rules-based models, DDA does not over-rely on the first or last touch.

However, DDA has limitations for AI traffic:

• DDA requires 400+ conversions per channel within 30 days to generate a stable model. For smaller sites where AI traffic is new, the conversion volume may not meet this threshold. GA4 falls back to last-click behavior below the threshold.
• DDA does not distinguish between AI platforms. ChatGPT and Perplexity get pooled into the same referral or unassigned bucket unless custom channel grouping is already in place.

If DDA is not viable due to volume constraints, a position-based model (40/20/40 split across first, middle, and last touch) provides a reasonable approximation for AI conversion paths. This model gives appropriate weight to the AI first touch while still crediting the closing action.

Step 1: Define conversion paths with AI touchpoints. Create a custom channel group (as described in prior posts) so that AI sources appear as distinct channels. Without this grouping, AI touchpoints are invisible to the attribution model.

Step 2: Extend the attribution window. In GA4 admin, navigate to Attribution Settings and extend the conversion lookback window to 90 days for all model comparisons. This ensures that AI-initiated paths are not truncated.

Step 3: Run model comparison reports. GA4's Model Comparison tool lets you compare how different attribution models distribute credit across channels. Run a comparison between last-click, first-click, linear, position-based, and data-driven models. Focus on the delta for your AI channel. If the difference between first-click and last-click credit for AI sources exceeds 5x, you have a strong signal that your current default model is misattributing AI conversions.

Step 4: Select and freeze a model. Choose the model that best represents AI's role in your conversion paths. Document the rationale. Do not change models mid-quarter, as model shifts distort trend comparisons.

Run the GA4 Model Comparison report for the last 90 days. Select last-click as the baseline model and data-driven as the comparison model. Export the channel-by-channel conversion credit for both models. For each AI source channel, calculate the percentage difference in attributed conversions. If the difference exceeds 50 percent for any AI source, your current attribution model is distorting AI performance. Switch to a position-based model or DDA (if volume permits) and document the change in your analytics governance log.