To measure AI search visibility you’ll need two different sets of records: those showing the answers in which your company or website appeared, and the analytics which show what took place when someone arrived at your site. The rule should be applied each time a brand is referred to or a citation is made, and it must be applied on every visit and in every conversion.
A screenshot can confirm an appearance. A monthly report needs more context. Which questions produced it? Was the description right? An incorrect service detail, for example, needs checking before that appearance is treated as useful exposure. For the commercial result, you will need the platform reports and customer records too.
What should you count?
A brand mention names your company or product. A citation indicates which material is providing the answer; the assistant may mention your guide when explaining it without suggesting that you should be hired.
The website visits which come from AI can be recognised, the conversions covering both the enquiries that were received and the purchases that were made. Keep them separate from appearances in answers. Adding mentions, citations and sessions produces a total with no consistent unit, and some observations overlap.
Build a baseline you can repeat
Choose questions from sales calls, support enquiries, site search and keyword research. For an agency, one might be: “What should a US retailer ask an outsourced marketplace management partner?” That is an illustrative prompt, not evidence of search volume.
Twenty or thirty questions is a suggested workload for the first round. It will leave plenty of buying questions untested. When the question already contains your company name, getting it back in the answer tells you little about discovery. Record those questions separately. Group the new questions separately so that they won’t have an effect on the main comparison.
Whoever runs the checks in ChatGPT, Gemini, Perplexity or Microsoft Copilot should leave a record of:
- Exact prompt, platform and date.
- Language, target market and available model or search-mode details.
- New conversation or follow-up question.
- The saved response and cited URLs.
In the following round, put the same key questions to people in new conversations. Comparison is easier if you have a fixed schedule, and even if the circumstances are the same you will be able to notice the differences in the way the questions are answered when you ask them twice.
A timeout belongs in the missing-data log. An answer that completed without mentioning you belongs in the results. Disclose changes in coverage before comparing percentages.
Calculate mention and citation rates
A response should count once for each valid response, even if the brand name appears more than once within it. Regarding citations from your own site, you need to verify that the response includes a link to your domain which is qualified; only one such qualified link is necessary to count the response, further links will not increase that count.
Mention rate = responses naming your brand ÷ all valid responses in the defined set × 100.
Owned-site citation rate = responses citing your domain ÷ all valid responses in that set × 100.
With 25 prompts run twice on one platform, there would be 50 responses to examine if every run completed. Fifteen mentions out of those 50 would put your mention rate at 30%. If ten responses linked to your domain, the citation rate would be 20%. The numbers here are only an example. This calculation does not tell you how many buyers saw an answer.
If the assistant provides a link to an article on your business which has been published by a third party, you must give that link its own entry instead of considering it as a citation from your own website. Moreover, when working out a rate based solely on the answers which include sources, you should label the denominator being used and retain the measure which is based on all the responses.
For competitor share of voice, divide your appearances by all brand appearances in a fixed competitor set, counting each brand once per response. This is sample share of voice, not market share.
What do Google and Bing report?
Google Search Console
Google’s Search Generative AI reports show impressions within AI features in Search and Discover. Google says the rollout reached all websites worldwide by August 31, 2026. Pages, countries and dates are available, with device reporting for Search.
These impressions remain included in overall performance reporting. Adding the AI figures to that total would double count them. Impressions alone also cannot supply an AI click-through rate.
Bing Webmaster Tools
The report, which can be accessed via AI Performance in Bing Webmaster Tools, displays the number of citations that your pages have acquired over time and includes information regarding Microsoft Copilot, AI-generated Bing summaries, and a number of partner integrations; it also provides examples of grounding queries, which are the phrases used to retrieve the content.
Those phrases need not match a customer’s original question. Citation counts do not establish placement within an answer. Report them separately from your sampled citation rate because coverage and denominators differ.
Find identifiable AI visits in GA4
Google’s channel definitions include an AI Assistant channel for recognized sources such as ChatGPT, Gemini and Copilot. Google AI Overviews and AI Mode fall under Organic Search. The AI Assistant total therefore covers only part of AI discovery.
Review session source/medium with landing pages, sessions and engaged sessions. Compare equivalent periods and similar visitor intent. Custom source rules need a written record too, with a review when the sources change.
Referral data will likewise be missing and you could end up seeing a higher number of direct visits or branded searches, even though this doesn’t explain how those visitors came to find out about your company, so don’t assign them to AI for that reason. Before looking at the poor results, check the landing page tracking as well as the form or purchase events and make sure that consent preferences are respected.
Connect visits to business outcomes
A B2B team will usually want to know whether an enquiry was worth passing to sales. A newsletter signup answers a different question. Check what the key event actually records. A button click can happen without a successful submission, and spam or duplicate records need removing.
Report conversions from identifiable AI-source sessions separately from conversions where the recorded AI visit happened earlier. Customer-reported discovery adds another source of evidence. These groups can overlap, so deduplicate leads or transactions before combining them.
The reporting model affects credit when you use event-scoped dimensions. Changing it leaves session-scoped dimensions unaffected, as Google’s attribution guidance explains. Record the scope, model and lookback window used.
For B2B sales, retain the agreed source in your CRM and allow comparable time for leads to qualify and close. In that case you’re comparing the enquiries from last week with those leads which had three months to turn into sales.
When working out the session conversion rate, divide the number of sessions in which the action you have selected took place by the number of eligible source sessions and then multiply the result by 100; a session should only be counted once, although when you are calculating the number of leads per session, leads from the same session will have an effect on the figure.
Make the monthly report useful
Show sampled mention and citation rates, accuracy checks, Google AI impressions and Bing citations. Then add identifiable sessions, relevant landing pages and qualified business outcomes. Include completed checks, missing observations and method changes.
Put counts beside percentages. In a sample of 20 sessions, a change from two converting sessions to three takes the rate from 10% to 15%. That’s only one additional converting session. Match reporting dates and timezones, noting when content or tracking changed.
In month one, agree what to measure and gather the baseline while checking that tracking works. Review cited pages and lead quality before choosing changes to test. An old page being cited may need updating. If visits produce no enquiries, the landing page is one place to investigate. These are investigation priorities, with no promise of growth within 30 days.
How can Alacrity help?
At Alacrity E-Commerce Solutions provide services in the areas of SEO and digital marketing together with website design and e-commerce development. You need to contact the team in order to discuss the pages that require checking, and while work is underway you must decide on the results you want to be reported as well as decide whether or not analytics or CRM integration is necessary.
Make the report easy to check
Even a colleague who has not collected the data should still be able to understand the meaning of each figure and must be able to carry out the necessary calculations themselves. A note should each time a method is changed so that the next comparison can be properly checked.
FAQ‘s
Can you track brand mentions without clicks?
Repeated answer checks capture mentions even when nobody visits your website. They describe the monitored responses, rather than total audience exposure.
Does GA4 capture all AI search traffic?
Its AI Assistant channel covers recognized assistant sources. AI Overviews and AI Mode sit within Organic Search; missing referral information can leave other visits unidentified.
Does a citation mean a recommendation?
Read the answer around the link. An assistant might use your page to explain a topic and say nothing about choosing your business.
How often should checks run?
Look at the situation week by week and then examine the business results on a monthly basis. Even if you change the schedule to match your own working habits, you’ll still have to make a number of observations before you can assess the performance.
Can AI visibility be linked to revenue?
Where tracking and customer records support the connection, sales can be attributed under your chosen method. That does not prove AI alone caused the purchase.
Tagged ai search visibility, ai search, ai mentions, ai citations, referral traffic, ai traffic, ai conversions, chatgpt visibility, gemini traffic, copilot citations, ga4, search console, bing webmaster tools, ai analytics, generative ai search, aeo, geo, alacrity ecommerce solutions