- Keep AI visibility, referral visits and qualified enquiries separate; they are different events, not interchangeable results.
- Use native search reports alongside existing website analytics, recording each report’s scope and date range.
- Do not add Google generative AI impressions to Web totals: the reporting overlaps.
- Keep unknown attribution visible and judge changes over comparable periods, with enquiry quality as the commercial check.
1. Define what a useful result means
Before opening a dashboard, define a qualified enquiry for your business. It could mean a genuine request for a service you supply, from a customer you can serve, with enough information to discuss the next step. Keep spam, recruitment requests and unrelated sales pitches outside that count.
Choose a small group of commercially important pages, such as a core service and its supporting guide. Record their preferred URLs, intended audience and purpose. Track both these pages and the whole site so an improvement on one page does not hide a wider decline.
2. Read visibility reports in context
Google’s Generative AI performance report (opens in a new tab) reports impressions in supported Search experiences, including AI Overviews and AI Mode. Review pages, countries, devices and dates. Its data overlaps with Web reporting, so adding these impressions to your existing Web totals would double-count visibility.
An absent report or sparse data does not establish that your website is technically blocked. Check the report’s availability guidance and your property settings. Google’s Search generative AI control (opens in a new tab) includes sites by default; an authorised owner can review whether that setting has been changed.
Bing’s AI Performance announcement (opens in a new tab) describes citations across supported Microsoft and partner experiences. A citation records a reference to content, not a website visit or a sale. Read the report’s coverage before comparing platforms; similarly named metrics may measure different things.
3. Identify the visits you can actually observe
If your site already has suitable analytics, review recorded traffic sources and landing pages. OpenAI’s publisher FAQ (opens in a new tab) says ChatGPT adds utm_source=chatgpt.com to referral URLs. Check how your analytics handles that source rather than relying only on a broad referral-channel total.
Keep each platform’s observed source separate. Record sessions and their landing pages using the same report definition each month. Avoid combining users, sessions and page views into one traffic figure. A person can return several times or visit several pages.
Treat these as recorded referrals, not complete AI influence. People can read an answer and later search your brand, use another device or share a link. Consent choices and measurement settings also affect what is recorded. A direct or unknown visit should remain unassigned unless you have supporting evidence.
4. Keep a repeatable reporting worksheet
Copy this structure into your existing reporting document. Use complete, comparable periods and record the timezone and any filters. Enter “not available” when a measure is missing; do not turn missing data into a zero. Retain the underlying exports so someone else can reproduce the comparison.
| Measure | Evidence to record | Interpretation |
|---|---|---|
| Google AI impressions | Search Console export, period, page and filters | Visibility within supported features; overlaps Web totals. |
| Bing AI citations | AI Performance export, period and cited URLs | References within report coverage; not visits. |
| Recorded AI referral sessions | Analytics source, landing page and period | Observed visits; attribution is incomplete. |
| Qualified enquiries | Deduplicated enquiry count and qualification rule | Commercial opportunities, not every form submission. |
| Customer-reported discovery | Optional answer to how the customer found you | Supporting evidence; may involve several channels. |
| Changes and next action | URL, publication date, change made and review owner | Context for investigating movement; not proof of causation. |
When counts are small, show the actual numbers before percentage changes. One additional enquiry can create an impressive percentage without demonstrating a reliable trend. If you calculate a referral conversion rate, use qualified enquiries attributable to that same measured visitor group and period, not all enquiries received by the business.
5. Let the evidence choose your next improvement
- Visibility without recorded visits: review whether the page offers a useful reason to continue reading or evaluate your service.
- Visits without suitable enquiries: check service fit, geographic availability, mobile usability and the clarity of the next step.
- Missing visibility on priority pages: inspect indexing, crawl access and content usefulness before publishing more pages.
- A sudden reporting change: check filters, tracking changes and platform coverage before attributing it to a content edit.
Keep a dated change log and review several comparable periods. Seasonal demand, campaigns and platform changes can move results too. Google’s AI optimisation guidance (opens in a new tab) emphasises useful content and sound technical foundations. Use the evidence to improve those foundations through SEO and AI search and GEO.