How to prove AI search ROI to leadership.

The slow way: three data sources and a prayer

The manual version stitches three disconnected sources. Shopify analytics for traffic by landing page. A handful of ChatGPT and Perplexity screenshots where the brand appears. A paragraph in the deck arguing that AI is driving traffic. Then someone asks "how much," and you have two bad options: invent a percentage (the deck dies under scrutiny) or argue that AI attribution is impossible because chat referrers are scrubbed (true, but it sounds like excuse-making).

The structural problem: every AI-visibility story needs two anchors — a visibility metric (citations) and a commercial metric (sessions, revenue) — joined at the same grain. Brand-level mentions joined to site-level traffic proves nothing; the correlation has to live at the SKU, where a specific product's citations rose and that product's sessions rose with them. Without per-SKU joins, the narrative stays impressionistic no matter how many screenshots it carries.

The eCommerce Insights way

1. Establish the dated baseline. The moment optimization begins: catalog scores, per-SKU citation counts, Shopify session volumes by landing page. Recorded automatically; export the dated PDF so the "before" is on the record.
2. Track per-SKU lift. For each optimized SKU, citation-count deltas across ChatGPT, Perplexity, Google AI Overviews, Gemini, Claude, and Copilot, plotted per SKU. The measurement methodology behind this is in measuring AI visibility: what actually matters.
3. Join visibility to sessions. The Shopify integration pairs each SKU's visibility delta with its session and revenue delta. Where both rose in the same window, the lift is flagged attributable. This is the closest honest proxy — perfect attribution does not exist, and the report says so.
4. Convert to a labeled range. Session deltas times each SKU's historical session-to-revenue rate, reported as a range: "$35K–$85K this quarter" is credible; "$62,341" is not. Labels: attributable where correlation is tight, contributing where causation is murkier, illustrative for the model-based pieces. Never "certain."
5. Build the two slides. Slide one: score before/after, cited-SKU count before/after, the revenue range. Slide two: three to five named SKUs — what changed on each PDP, what moved. Branded PDF export. A framing that reads well in a boardroom: "Cited SKUs grew from 12 to 34 this quarter; based on session deltas on those SKUs we attribute $35K–$85K of revenue to AI visibility lift. AI attribution science is still developing; we refine as the data matures."

The spend side of the equation is public: plans run $99–$349/mo (pricing), which makes even the conservative end of a defensible range an easy multiple. The complementary GA4 work — making AI referrals visible at all — is its own job: track AI referral traffic in GA4.

What "good" looks like

The counterintuitive rule: honest hedging strengthens the deck. "Attributable vs. contributing vs. illustrative" labeling reads as rigor, and leadership funds rigor. The confident single number gets one meeting; the cautious range gets a standing line item.

Related jobs