Use GA4 to measure identifiable ChatGPT visits, NeuraCite to monitor relevant AI answers, and your CRM or enquiry records to capture what customers say influenced them. For a Cardiff business, these sources answer three connected questions: are you appearing in useful recommendations, are people arriving at the website, and are those enquiries becoming suitable opportunities? A shared reporting period and clear definitions let you assess the commercial value of AI visibility without losing the customer journey between dashboards.
Build a report around three forms of evidence
Saved AI answers show how your business appears for the questions you track.
GA4 shows measurable visits and configured actions on your website.
Customer feedback and lead records reveal discovery routes and enquiry quality.
Which tools should you put in place first?
1. GA4: measure identifiable visits and actions
GA4's Traffic acquisition report helps you inspect where new and returning sessions came from. Start with Session source / medium and review the sources actually present in your property. Then examine landing pages, engagement and the actions that matter to the business. Google documents the Traffic acquisition report.
2. NeuraCite: monitor the recommendation context
Use NeuraCite Answer Engine Insights to test relevant buyer questions and inspect saved answers, competitor appearances and sources. Its GA4 integration helps review observable AI referral activity alongside visibility work. This gives the team a place to investigate both discovery and recorded website engagement.
3. Your CRM or enquiry register: assess the lead
Use the system your sales team already maintains to record how the prospect found you, what they need and whether the enquiry fits. A spreadsheet can work initially if someone owns it. The useful feature is a consistent route from enquiry to qualified opportunity, booking or completed sale.
How do you identify ChatGPT traffic in GA4?
Inspect actual source values before building a permanent filter. Look for identifiable ChatGPT domains and campaign-source values in the selected period, then check the associated landing pages. Keep a documented list of the values your report includes so another person can reproduce it.
Use session-scoped dimensions when asking where particular visits came from. First user source answers a different question about initial acquisition, while event-scoped attribution can allocate credit differently. Google's guide to traffic-source scopes explains these distinctions.
Record the source, medium, sessions and relevant actions together. Review unfamiliar or suspicious patterns before accepting them as customer activity. Also check your own testing visits so repeated internal checks do not distort a small local business's report.
A customer may see your name in an AI answer, search for the business later or telephone without visiting the website. That journey may produce no identifiable ChatGPT referral. Keep the customer feedback beside the analytics evidence so useful discovery is not overlooked.
Which website actions should count as leads?
Count a completed business enquiry as a lead action, then distinguish it from browsing behaviour. For a Cardiff service company, suitable events might include a successful quote request or confirmed appointment request. A page view, button click and submitted form represent different levels of intent.
Configure the event against a successful outcome and test it. Check that a failed form submission does not count and that refreshing a confirmation page does not create avoidable duplicates. Mark the appropriate event as a GA4 key event once the collection is correct. Google explains how key events are configured and reported.
Keep lead quality in the sales record. A form submission can be spam, an unsuitable location or a request for a service you do not offer. Use consistent qualification criteria, such as a real contact, relevant need and serviceable area, before treating it as a commercial opportunity.
What should you ask customers who mention ChatGPT?
Add a short, optional discovery question to the enquiry process and let customers answer naturally. “How did you first hear about us?” works as an opening. When someone mentions an AI assistant, ask what they were looking for and what made them consider the business.
For example, a Cardiff interiors company could learn that a prospect asked for suppliers who could visit their home and provide an installation service. That detail is more useful for content planning than the platform name alone. It tells the team which buying criteria need a clear answer.
Record the platform the customer remembers, if any.
Capture the need or question in their own words where practical.
Note whether they clicked a link, searched separately or contacted you directly.
Keep “unsure” available instead of forcing a channel choice.
Offer the question in Welsh as well as English when that reflects your customer service. Have the Welsh wording reviewed by a fluent speaker. Avoid requesting a complete private conversation when a short account of the discovery route is enough.
How should a Cardiff business compare visibility and leads?
Report the measures side by side over the same dates, with their definitions visible. Use a stable set of relevant prompts for visibility, consistent source rules for referrals and clear qualification criteria for leads. Include changes to your website or tracking that could affect the comparison.
This is a hypothetical reporting example, not a client result:
Monitored answers: the business appeared in 8 of 20 successful tracked answers.
Website activity: GA4 recorded 30 sessions under the selected ChatGPT source values.
Enquiries: two qualifying enquiries occurred in those identifiable sessions.
Customer feedback: four prospects separately mentioned ChatGPT during enquiry handling.
The four self-reported prospects could overlap with the two recorded enquiries. Reconcile the records where you can rather than adding them together. The 8-of-20 result describes that monitored sample; it does not establish a market-wide share of all Cardiff conversations.
For local context, track the service areas that matter to fulfilment. A business covering Cardiff and the Vale should record whether enquiries come from places it can serve. Do not assume an analytics location field proves where the customer needs the work carried out.
What does a useful monthly review look like?
Start with the decisions the report should support. Review the questions where your business is missing, the pages receiving identifiable visits and the reasons enquiries qualify or fail. Then choose a specific improvement, such as clarifying coverage, explaining a service or making the enquiry route easier.
Use campaign tags consistently on links you control when they serve a defined measurement purpose. They do not reveal earlier AI exposure or retrospectively identify an untagged visit. Google's traffic-source documentation explains how source information and manual tags populate reports.
If tracked recommendations improve but enquiries remain weak, inspect the landing page and offer. If good enquiries mention ChatGPT despite few recorded referrals, retain that feedback and examine the journey. Both findings can guide useful work without forcing every measure into a single score.
To plan the visibility side, use our local GEO prompt research guide. Bring a service page, example buyer questions and your reporting requirements to a NeuraCite demo, then check the current plans against the work your team needs to do.