Measuring AI Visibility for a Hong Kong Mental-Health Charity

A Hong Kong mental-health charity needed a repeatable way to measure whether it appeared in AI answers and whether visitors continued into a priority service journey. The work also had to distinguish visibility, engagement and action signals without treating any one metric as proof of completed applications.

Loketing measured the same 50 Cantonese questions across Perplexity, Microsoft Copilot and Google AI Overviews in matched seven-day windows. We connected that reporting with GA4 engagement and apply/referral click tracking. The verified delivery covered measurement, tracking and reporting; the available records do not establish that specific content or schema changes caused the observed results.

Pooled citation share rose from 11.60% in 15-21 May to 17.48% in 21-27 June 2026, an increase of 5.88 percentage points or 50.7% relative. Separately, priority service landing-session engagement across all channels rose from 51.0% in April to 68.6% during 1-18 August. GA4 recorded 868 apply/referral click events from June through 18 August.

The denominator is every source citation recorded in the prompts export. The numerator is every citation to the charity's whole domain, rather than to one service alone. The matched comparison used the same 50 Cantonese questions and equal seven-day windows: 731 domain citations out of 6,303 total citations from 15-21 May, compared with 1,094 out of 6,260 from 21-27 June 2026.

Perplexity rose from 11.1% (344/3,094) to 20.7% (650/3,137). Microsoft Copilot rose from 15.0% (157/1,048) to 22.4% (242/1,078). Google AI Overviews fell from 10.7% (230/2,155) to 9.9% (202/2,042), so it did not improve. These engine rows come from citation-level exports. Their denominators differ slightly from the pooled prompt totals, so the two sets of denominators should not be combined.

The engagement comparison covers all traffic channels to the priority landing journey: 51.0% in April and 68.6% during 1-18 August, a 17.6 percentage-point increase. GA4 recorded 439 apply/referral click events in June, 236 in July and 193 from 1-18 August, for 868 events in total. These are click events, not unique applicants or completed applications.

These are observed changes during the engagement, not a controlled causal test. Campaign activity, audience changes and changes in AI retrieval may also have contributed. Generated citation figures for July and August were excluded. A separate August answer-capture panel provides a new baseline and should not be read as growth from the May-June comparison.