Define a useful enquiry before counting visibility
If AI visibility rises, the next question is whether the business is receiving more suitable enquiries. Define suitability before the test: the right problem, a plausible budget, a relevant service, and a realistic next step. More form submissions are not automatically better outcomes.
An 18 August preprint by Wang and colleagues reports an experiment with 1,100 participants in which AI search treatments affected publisher click-through. It gives a reason to separate exposure from visits. It does not forecast the enquiry rate of your service business or validate the specific pilot below.
Choose one change with a plausible mechanism
Use a hypothesis such as: “Clarifying who this service fits and showing one relevant proof example will improve the proportion of suitable enquiries.” That is more actionable than “optimise for AI.” It names the page, the intervention, and the expected buyer response.
Select an established service page with enough activity to observe. Avoid changing the headline, price, form, promotion, and proof all at once. If several changes are necessary, record them as a combined intervention and accept that their individual effects cannot be separated.
Record the baseline and the reporting boundaries
Collect a comparable prior period, the page's enquiry count, suitable enquiries, booked conversations, and available discovery information. Record the qualification rule so it stays consistent. Keep the original denominator: three suitable enquiries out of four is different evidence from seventy-five out of one hundred.
Use Search Console's documented AI report for its supported visibility measures. Keep analytics sessions and CRM outcomes in separate columns. Their definitions and date boundaries may differ; do not join them into a fictional person-level journey.
Run the month as a small learning project
- Before day one: write the hypothesis, baseline, qualification rule, and known confounders.
- Day one: publish the approved page change and record the exact version.
- Week one: check form operation and data collection. Avoid judging commercial success immediately.
- Weeks two and three: inspect data quality and record campaigns, outages, or sales-process changes.
- Day thirty: compare outcomes and decide whether to keep, revise, extend, or stop the test.
Read the result without manufacturing certainty
In a fictional example, suitable enquiries rise from three of ten to four of eight. The proportion improves, total volume falls, and the absolute improvement is one suitable enquiry. That is a reason to inspect the conversations, not announce a dramatic AI-driven growth result.
If the sample is too small, extend observation or test a more active page. If visibility rises while enquiries remain unsuitable, revisit offer clarity and targeting. If good enquiries increase without a traceable AI referral, preserve the uncertainty rather than assigning credit to your preferred channel.
A closing memo should include the version tested, raw counts, relevant context, interpretation, and next decision. The AI Problem-to-Solution Assessment can help define a bounded test before the team invests in a larger reporting or content system.
Link Map
Sources and editorial review
Sources reviewed on 14 September 2026. The practical workflows and illustrative examples are JQ AI SYSTEMS analysis unless explicitly attributed.
- Wang, Gleason, Bart, Wilson and Metaxa: AI in Search Reduces Publisher Referrals Without Improving User Experience (18 August 2026; research preprint).
- Google Search Console Help: Generative AI performance report (Current documentation, reviewed 14 September 2026).