AI Citations but No Visits? Measure Visibility, Traffic and Enquiries

Start with the question the business needs answered

“Are we visible in AI answers?” and “Did those appearances produce useful enquiries?” are related questions, but they need different evidence. A search report may show that a URL appeared, while the website can observe only activity that reaches the site. Sales records then describe what happened after contact.

Write down the question before opening the dashboards. If the goal is to check whether a newly corrected product page is being used as a reference, page-level visibility may be relevant. If the goal is to assess customer acquisition, you also need evidence about visits, received enquiries and their fit with the business.

Keep four layers of measurement separate

Use this worksheet to record the source and meaning of each number. Do not add impressions and citations together, or divide one system’s enquiry count by another system’s impressions without a defensible connection between them.

Read Google and Bing reports according to their definitions

Google announced dedicated generative AI performance reports in Search Console in June 2026 and states that the rollout reached all websites worldwide by 31 August 2026. The announced views report impressions and dimensions such as pages, countries and dates; do not infer a dedicated AI click figure from an impressions-only field. The data is also included in the overall performance reporting described by Google.

Bing’s AI Performance documentation describes citations and cited pages across supported Microsoft and partner experiences. Its grounding-query information represents a sample, and citation counts do not establish a page’s rank or position inside an answer. These platform reports are useful evidence, but they are not a universal count of everything said about your business by every AI product.

Treat identifiable referral traffic as partial evidence

When analytics records a recognisable referring source, review the landing page and subsequent actions. Also document the cases you cannot reliably attribute. People may use applications, copy links, switch devices or return later through a different route, so visible referrals should not be described as a complete record of AI influence.

Keep a consistent source-classification rule and save it with the report. If the rule changes, annotate the change before comparing months. Do not collect sensitive enquiry text in analytics simply to improve attribution; the business can classify enquiry quality in its appropriate sales workflow.

Use a small, dated monthly review

Choose a defined set of important pages and a consistent review window. Record each platform separately, along with page changes, reporting filters and any missing data. In an illustrative review, a page might show more reported citations without a corresponding increase in identifiable visits. That is a finding to investigate, not proof that the content failed or that the business gained customers.

Look at whether the cited material matches the offer and whether visitors who do arrive can take a useful next step. Track received and qualified enquiries using the same definitions each month. Small samples should be reported as counts with context rather than dramatic percentage changes.

Use manual answer checks as examples, not a score for the whole market

A manual check can reveal an outdated price, a mistaken company description or a missing source link. Record the exact question, product, date, language, relevant settings and answer evidence. Repeating a few prompts does not reproduce every user’s experience, and personalised or changing answers limit comparisons.

When you find an error, correct the underlying company information you control and document the source. Continue to observe it without promising that the answer will change on a specific date. The useful outcome of measurement is a clearer next action and a more accurate record of what is known.