1. Check whether the question named the clinic
A clinic name appearing in the answer to ‘What does [clinic name] treat?’ means something different from the clinic appearing as an option for ‘I’m looking for a [specialty] in [area].’ To interpret a recommendation rate, you need the exact prompts and their branded or non-branded classification.
2. Look at the denominator and how many runs were unmeasured
If the rate is defined as ‘valid responses that recommended the clinic ÷ all valid responses,’ report how many times the prompt was run, how many responses were received and assessed, and how many could not be assessed because of errors or skipped runs. If valid responses exist and none recommend the clinic, the rate is 0%. If no valid responses exist, mark the rate as not measured (null). Rates with different denominators are difficult to read as changes under the same conditions.
3. Distinguish mentions, recommendations, sources, and facts
Count a name in the text or a list as a mention. Count it as a recommendation when presented as an option that meets the patient’s criteria. Distinguish a link to a Place listing from a citation that links to a page on the clinic’s own domain. Compare specialty, hours, and address with official information. A citation and a recommendation can occur independently in the same answer.
4. Keep enough detail to repeat the test
Record the service and model names when displayed. Note the search mode, region and language, date, account and conversation conditions, and number of runs for each prompt. Use the same questions and scoring rules for before-and-after comparisons. Label separately any result for which you cannot confirm search ran. Links and answers in Google’s AI search features can also vary by the model and method used.
5. Check visits and inquiries in the relevant records
Use site analytics to see whether visits followed the appearance of a link. A phone button click, inquiry form submission, and completed booking are separate actions. Attribution may be missing due to consent settings or app and browser behavior, so document the tracking coverage in your report. You can count actions without sending contact details or inquiry text in analytics events.
TRY THIS TODAY
Try this today
- Check that the report includes exact prompts, the valid-response denominator, the number of unmeasured runs, and the rule used to identify a recommendation.
- Check whether citations increased and whether visits and inquiries increased using the relevant records for each.
Sources and verification date
Verified. This guide is for practical preparation and does not guarantee visibility.