Your website appears in more of the AI answers you checked this month. Did more people see it, click through, or ask for a quote? A manual answer, a reported impression and a website session measure different events. You cannot add them into one “AI visibility” total.
Here is a repeatable way to report what changed—and what remains unknown. First, see our guide to mentions, citations, impressions and visits if those terms are new to you.
Start With the Decision, Not a Visibility Score
Choose pages and customer questions worth watching. A hypothetical Malaysian shoe retailer might track its sizing guide and a product-category page: do they answer fitting and buying questions accurately, and do useful visits or enquiries follow?
Set a repeatable period and record the pages, country, language and device filters. Save a dated export. If the pages, filters or analytics setup change, start a new baseline.
Keep Each Evidence Source in Its Own Column
| Source | Record | Do not infer |
|---|---|---|
| Google Search Console | Impressions for links shown in supported Google Search AI features, by page and the same date, country or device filters. | That every AI answer mentioned your brand, or that the impression became a visit. |
| Bing Webmaster Tools AI Performance | Reported citations, cited pages and grounding-query groups on its supported surfaces. | Exact customer prompts, complete exposure, ranking or clicks. |
| Manual answer sample | The answer shown for a saved prompt, date, product, mode, language and location context. | How often all customers saw the same answer. |
| Website analytics and enquiries | Recorded sessions, landing pages and tested enquiry events or actual leads. | Which particular AI answer caused each visit or sale. |
Google’s Generative AI performance report covers listed Google Search features such as AI Overviews and AI Mode. Its impressions are already included in the Web total; adding both totals double-counts them. If the report is unavailable, write “not available”, not zero. See our Google AI features guide for the reporting boundaries.
Bing’s AI Performance public preview covers supported Microsoft experiences and selected partners. Its citation and grounding-query data are sampled or aggregated, not a complete log. A grounding-query group is not an exact customer prompt; use it to identify cited pages and themes.
Use Manual Checks as a Sample, Not a Market Estimate
Save a fixed set of customer questions. Check the same wording, product, mode, language and location context. Record the date, answer, named businesses, cited URLs and factual errors. Our prompt-tracking guide covers the method.
If one run cites the sizing guide and the next does not, report “one of two checks,” not “50% AI market share.” Answers vary by time and context. Use the sample to find information gaps, not to estimate every customer’s exposure.
Read Website Visits and Enquiries Separately
In GA4 Traffic acquisition, review session source/medium; compare landing pages using a suitable page report or exploration with the same date range. Keep a list of recognized AI referrers and note when that list changes. Count a qualified enquiry only after checking the actual form submission, call or lead record. A click on a button is not automatically a completed lead.
Not every AI-assisted visit carries an identifiable referrer. GA4 can place visits without clear referral information under Direct; do not relabel them “AI traffic.” A visitor may also return by another route later. See our SEO reporting guide for the wider context.
Compare Like With Like Before Explaining a Change
Compare the same pages and filters. Note site edits, tracking changes, seasonality and platform report changes beside the numbers. An edit and a rise in citations during the same month do not prove cause and effect.
The retailer’s sizing guide might appear in more manual checks while identified referral sessions stay flat. Neither observation proves what all customers saw or whether the guide created value. Check the page’s answer and repeat the same test.
Write a One-Page Monthly Report
- Scope. List the pages, country, language, device, period and fixed prompt set. Keep a dated export or screenshot of the source reports.
- Observed changes. Show each platform report, manual sample, site sessions and qualified enquiries in separate rows, with their source and filters.
- Known limits. State unavailable data, sampling, changed tracking and any missing referrer information. Do not turn unknown into zero.
- Possible explanations. Note site edits or market and platform changes as possibilities, not proven causes.
- Next check. Assign one page or claim to verify, plus the date and owner for the next comparable measurement.
Use the report to choose an action. If the sample quotes an outdated shoe size chart, verify and correct it. If citations rise but enquiries do not, inspect the linked page and enquiry route. See our AI search strategy guide for broader priorities.
Frequently Asked Questions
How Often Should I Measure AI Search Visibility?
Use a schedule you can repeat, such as a monthly report with the same page and prompt set. Check urgent factual errors sooner. More frequent checks do not make a small manual sample representative of all users.
Can GA4 Show Every Visit From an AI Answer?
No. GA4 can report recorded sessions with identifiable sources, but some referrals are absent and later visits can arrive by another route. Do not treat Direct traffic as a substitute for an AI-referral report.
Can I Add AI Citations, Impressions and Clicks Into One Total?
No. They describe different events and reporting scopes. Keep each source separate, and do not add Google AI impressions to Web impressions because those AI impressions are already part of the Web total.