How to Measure AI Search Visibility

Start with four measures

Record how often AI answers mention your brand, how often they cite your website, how many visits arrive from identifiable AI sources, and how many of those visits turn into enquiries. The first two come from sampled answers and publisher reports. The last two come from website analytics and lead records.

If you run a South African business, test the markets you actually serve. A Cape Town consultancy, a national ecommerce business and a South African exporter should expect different recommendations for similar questions.

1. Build a prompt set around customer decisions

Start with about 20 to 30 prompts that cover the services you want to sell. Use sales calls, customer emails, Search Console queries and enquiry forms to get the wording. Include unbranded questions for discovery and branded ones for accuracy. Track the groups independently so an answer to “What does our company do?” cannot inflate visibility on questions a new prospect would ask.

Store the exact prompt, language, location, platform, product mode and date. Where practical, use a fresh session and record whether web search was on. If you serve customers in English and Afrikaans, sample each language and report results independently. Adding new prompts is useful as priorities change. Keep a stable core set so the monthly comparison remains meaningful.

2. Track mentions and citations independently

A mention happens when the answer names your company. A citation happens when it links to a page as a source. Your brand can be mentioned with a citation to a directory, while your own page can be cited without your brand appearing in the answer text. Capture the cited URL and domain alongside the brands named.

For example, if a brand appears in 6 of 30 tested answers, its sample mention rate is 20%. If its website is cited in 3 of those 30 answers, its sample own-site citation rate is 10%. Use the number of valid answers as the denominator and record unsuccessful runs separately. For Google searches, record whether an AI Overview appeared as well as whether it included your site. This helps distinguish a change in feature frequency from a change in your coverage.

Open the sources that matter. Check whether the linked page supports the claim, whether the answer names a similarly titled company and whether the URL still leads to the right page. Save the relevant answer text and date so a colleague can see what you observed. Repeat a small subset of high-value prompts to see how much results vary between runs.

3. Use native reporting where it is available

Google Search Console offers a dedicated generative AI performance report for impressions from features such as AI Overviews and AI Mode. The overall Search performance report also includes that activity, so keep the two views distinct when totalling impressions. Google Search Central documentation

Bing Webmaster Tools’ AI Performance report shows how often URLs from your site are cited across supported Microsoft AI experiences. Review cited URLs and sampled grounding queries to understand what topics those URLs support. Citation counts do not tell you where a source appeared in an individual answer. Microsoft’s AI Performance documentation

Keep native reports next to your prompt sample. They cover activity those publishers can observe. Your sample gives you the wording of the answer, the competitors named, and the market and buying stage behind each prompt.

4. Connect identifiable visits to enquiries

In GA4, open Traffic acquisition and review session source and medium. OpenAI documents that links from ChatGPT search include utm_source=chatgpt.com. Check the landing pages and configured key events for that source, such as a confirmed contact form submission or appointment booking. Test the key event yourself and compare meaningful enquiries with your CRM or inbox. OpenAI’s publisher FAQ

A customer can read a recommendation, open a new tab and search for your brand. That journey can show up as organic search or direct traffic. A simple “How did you hear about us?” field can add context. Report identifiable visits and self-reported leads with their respective attribution methods so readers can understand what each number measures.

5. Turn the findings into one change you can verify

• Competitors appear for service questions: compare the sources supporting their inclusion and check whether your service page explains the same buying details, such as coverage, eligibility and pricing.
• An answer repeats outdated information: correct your own page and contact any relevant third-party publisher whose old listing supports the claim.
• A useful page attracts citations but few enquiries: check whether its next step fits the reader’s question and whether the landing page explains the relevant service.
• A priority page is hard to access: verify its indexability, rendering and internal links. Review the specific platform’s crawler documentation before changing robots rules.

Keep a dated record of the change and recheck the same prompt set after a reasonable interval. One answer can change because of location, model updates or variation between runs. Avoid buying placements that promise AI citations or flooding the web with repetitive brand mentions. Spend the effort on accurate facts, sources customers trust and content that answers real buying questions.

Frequently asked questions