## AI Assistant Tracking Requires New Metrics **Fewer clicks can still mean more revenue—if you’re cited.** Hold **average order value** ($AOV$) constant and let - $S_0$ = baseline sessions - $S_{\mathrm{ai}}$ = *incremental* AI-referred sessions (net of cannibalization) - $\mathrm{CVR}_{\mathrm{avg}}$ = baseline conversion rate - $\mathrm{CVR}_{\mathrm{ai}}$ = conversion rate of AI-referred sessions If $S_{\mathrm{ai}}/S_0 \approx 0.10$ (10 %) and $\mathrm{CVR}_{\mathrm{ai}}/\mathrm{CVR}_{\mathrm{avg}} \approx 4.4$ (4.4 ×), then the expected lift in revenue is $ \frac{\Delta R}{R_0} = \frac{S_{\mathrm{ai}}}{S_0}\; \frac{\mathrm{CVR}_{\mathrm{ai}}}{\mathrm{CVR}_{\mathrm{avg}}} \approx 0.44 \quad (44\%). $ This shows why measuring the **right** assistant-specific metrics is critical: traditional $CTR$ alone can’t explain the upside. --- ### Business-level Metrics (track in GA4) | Metric | Why it matters | Implementation hints | | --- | --- | --- | | **AI-referred Sessions** | Size of the new channel | Standardised UTMs, referrer allow-list, server-side tagging | | **AI Session Outcomes** | Profitability of the channel | $\mathrm{CVR}$, $\mathrm{RPS}$, $AOV$, $LTV$, refunds/returns | --- ### Assistant/content-level Metrics (benchmark loop) 1. **Build** a representative query set $Q$ (100–200 long-tail prompts across intents, buyer stages, user profiles; branded & unbranded; clustered by topic). 2. **Generate** 3–10 assistant-style sub-queries per prompt (captures “query fan-out”). 3. **Sample** responses across target assistants and record: * **AI Citations** – raw count of your brand mentions/links. * **Share of Answer** ($SOA$) – your slice of citations vs competitors. * **Sub-query Rank** ($SQR$) – how often you appear in the top $k$ organic results for the generated sub-queries. 4. **Re-sample** at fixed intervals (e.g., weekly) to monitor lift or decay. #### Share of Answer (SOA) Per-query: $ \mathrm{SOA}(q)= \frac{\text{your brand citations in the response to }q} {\text{all brand citations in that response}}. $ Aggregate over a query set $Q$: $ \mathrm{SOA}(Q)= \frac{1}{|Q|} \sum_{q\in Q}\mathrm{SOA}(q). $ *Notes:* Weight citations by prominence if desired (e.g.\ primary answer gt;$ collapsed source gt;$ footnote). Average $\mathrm{SOA}_a(Q)$ across assistants $a$—raw or impression-weighted—to spot where to invest. #### Sub-query Rank (SQR) For each assistant-generated sub-query $q\in Q$, let $r_q$ be your best organic rank (set $r_q=\infty$ if you don’t rank in the window). Then $ \mathrm{SQR}@k= \frac{\lvert\{\,q\in Q:\ r_q\le k\,\}\rvert}{|Q|}, $ which is recall@ $k$. Choose $k$ to match funnel depth (e.g.\ $k=3$ or 10). Compute $\mathrm{SQR}_e@k$ per engine/assistant $e$ and average or weight by impressions. > **Key question:** Are you growing $SOA$ fast enough—and improving $\mathrm{SQR}@k$ on the sub-questions assistants actually generate—to turn declining $CTR$ into higher $RPS$, $LTV$, and total revenue? --- ### Implementation Gotchas * Recover missing referrers with **server-side tagging** plus **allow-listed referrers**. * Maintain a **brand dictionary** (name variants, product lines, tickers) for reliable citation matching. * Log **assistant model/version, locale, timestamp** to control for response variance. * Track **citation position** so you can weight $SOA$. * Deduplicate hosts when computing $r_q$ if domain coverage (not page count) is what you care about. * Measure **cannibalisation** explicitly (uplift vs holdout) when estimating $\Delta R/R_0$.