Measure AI search visibility at several points: whether your business appears in an answer, whether a page is cited, whether someone visits, and whether that visit leads to a useful inquiry. Keep those measures separate so the report shows what you actually know.
An answer can mention your business without linking to you. A citation can appear without producing a visit. A buyer can also discover you in an AI tool and return later through another channel. One number cannot describe all of those journeys.
Choose questions that reflect a buying decision
Start with a manageable set of questions from customer conversations, sales calls, and search research. Include discovery questions and more specific questions about fit, implementation, and comparisons.
For a Webflow agency, hypothetical examples might include “What should a B2B team check before moving from WordPress to Webflow?” and “When should we hire an in-house Webflow developer?” These are example prompts to refine with actual buyer language, not measured demand data.
Keep branded and unbranded prompts separate. Asking whether Beqi offers a service tests how accurately the system describes Beqi. Asking which agencies fit a particular project tests a different form of discovery.
Use a repeatable prompt log
Run the chosen questions across the platforms that matter to your audience. Record the prompt, date, platform, visible model or mode, search setting, and any location or account context that could affect the result. Keep the answer or a reference to it so someone else can review the classification.
Repeat the observations. The research preprint “Don’t Measure Once” examines variation in AI search visibility and argues for repeated measurement. It supports treating a single response as an observation rather than a dependable position.
| Field | What to capture |
|---|---|
| Prompt and buyer task | The exact wording and the decision it supports |
| Run context | Date, platform, model or mode, search setting, and relevant location or account context |
| Brand mention | Whether the answer names the business, including an agreed alias |
| Owned-domain citation | Whether the response links to your domain and the exact URL |
| Accuracy | Any misleading service, pricing, capability, or company detail |
| Other sources | Which sources the answer uses and what they contribute |
Keep the tracked set stable for a reporting period. If you add prompts, report the new set separately or mark the change. Otherwise, a higher score may reflect easier questions rather than improved visibility.
Use native reports for the signals they provide
Google Search Console’s Generative AI performance report shows impressions from AI Overviews and AI Mode, with page, country, date, and device views. Google says the worldwide rollout was completed on August 31, 2026. A property may not show the report when it has insufficient impressions. Use it for Google’s reported exposure; it is not a cross-platform lead report.
Bing’s AI Performance report, introduced in public preview in February 2026, reports citations across supported Microsoft AI experiences and selected partners. It includes cited pages and a sample of grounding queries. Microsoft explicitly distinguishes citation counts from ranking or placement.
Webflow AEO analytics combines tracked-prompt results, bot activity, and identifiable AI-referred visits. Its visibility score concerns mentions; its citation rate uses responses containing citations as the denominator. Access requires the supported plan and Analyze setup. Check the definition behind each metric before comparing it with another tool.
Keep those reports side by side. Their populations and measurement methods differ, so adding them into one “AI visibility total” would be difficult to interpret.
Define your own reporting terms
For a manual prompt log, use simple definitions and write them beside the results. One useful choice is:
- Mention rate: successful recorded responses that mention your business, divided by all successful recorded responses in the selected set.
- Owned citation rate: successful recorded responses citing your domain, divided by all successful recorded responses in the selected set.
- Accuracy issues: the count and description of material mistakes in those responses.
These are suggested internal definitions. In particular, the citation-rate denominator above differs from Webflow’s documented metric. Label it clearly, and keep failed or unavailable runs separate from valid responses that omit your business.
Always show the sample size. In a hypothetical set of ten responses, moving from one mention to two doubles the count but still represents a change of one observation. A percentage alone makes that look more certain than it is.
Connect identifiable visits to qualified inquiries
Review the AI referrers your analytics can identify and confirm that primary inquiry events work. Follow the path into the CRM where your setup and consent choices support it. Define a qualified inquiry with the sales team before reporting success.
Add an optional “How did you hear about us?” question or capture that context during a sales conversation. Keep the original answer alongside the channel classification. A buyer who names ChatGPT can provide context that a referral report misses.
Report observed AI referrals, self-reported AI discovery, and other supporting evidence as separate fields. The same lead may appear in more than one of them, so deduplicate before counting people or opportunities.
Attribution will still be incomplete. Missing referral information, cross-device journeys, consent choices, and later visits can obscure how someone first discovered you. Rising branded search or direct traffic can provide context, but it does not prove that AI search caused the increase.
Make the report useful to the next month’s work
Use the report to choose a specific action. An inaccurate answer may point to outdated company information. A competitor citation may reveal an unanswered implementation question. Visits that fail to become inquiries may justify reviewing the landing page and form journey.
Keep a dated log of content and technical changes, plus other activity that could affect results. That gives you a better basis for reviewing progress than comparing two screenshots.
Beqi can help define the baseline, implement website changes, and report on the signals your team can actually observe. Start with the questions that matter to a purchase decision and the inquiry journey you want them to support.
Related reading
Written by
.jpeg)