Scorecard · No. 12 / 17
How should a supplier connect AI visibility to commercial outcomes?
Last updated
Demand a measurement chain, not a claim that citations equal pipeline. Keep the following evidence streams separate:
| Evidence stream | Question it can help answer | Proposed buyer safeguard |
|---|---|---|
| Controlled prompt observations | Are we appearing in the selected buying situations? | Preserve prompts, surfaces, repetitions and definitions |
| First-party platform reporting | Where is our site receiving recorded exposure or citations within that platform’s reporting scope? | Record metric definitions and reporting limitations |
| Website referrals and conversions | What identifiable visits and downstream actions occurred? | Audit channel rules, landing pages and duplicate handling |
| CRM and customer interviews | Did prospects report using AI during research? | Retain the original answer and avoid forcing a single-source explanation |
| Intervention comparison | Did changed areas improve differently from an appropriate comparison? | Record pre-existing trends, concurrent changes and possible spillovers |
OpenAI states that ChatGPT search referral URLs include utm_source=chatgpt.com, providing one useful analytics signal. Google’s current generative AI report documents impressions, while Microsoft explicitly separates citation share from traffic share. These are complementary observations, not interchangeable conversion measures. (help.openai.com)
Where feasible, compare changed and unchanged page or topic groups over comparable periods. Treat this as a design to scrutinise, not an automatic causal proof: shared brand changes, unequal starting trends and platform changes may undermine the comparison.
A June 2026 preprint on one website illustrates the issue. It separated referral growth from a broader platform-growth baseline, yet reported that a conservative placebo test left the intervention effect suggestive rather than conclusive. Do not transfer its effect estimates to another business. (arxiv.org)
For renewal, ask: what decision became clearer, what useful work was deployed, what changed in comparable evidence, and what remains unproven?
Sources cited on this page
- help.openai.com help.openai.com/en/articles/12627856?utm_source=openai
- arxiv.org arxiv.org/abs/2606.04362