Once you start working on AI search, the next question is obvious: is it working? Unfortunately, measurement here is patchy. Some activity can be tracked, some can only be sampled, and some cannot be seen at all.
This guide separates the three groups so you can build a modest, honest reporting routine, and so you can spot vendors who promise more certainty than anyone can deliver.
What can be tracked directly
Visits that actually arrive on your site can be measured. If a person clicks a link inside an AI answer or a chat assistant and lands on your page, your analytics may record a session. That gives you real numbers on traffic, engagement and conversions from those visits.
What you cannot measure directly is the many times your business was mentioned without a click, or the times you were left out. Those impressions happen inside someone else's product.
Finding AI referral traffic in GA4
Some assistants pass a referrer, so their visits show up in GA4 under a source name such as a chat product's domain. Others strip the referrer or open links in ways that appear as direct traffic. This means any GA4 count will be a floor, not a full picture. Source names also change, so check them periodically.
- In GA4, open Reports, then Acquisition, then Traffic acquisition.
- Change the primary dimension to Session source or Session source / medium.
- Search the list for known assistant domains and note which appear.
- Build an Exploration or a custom channel group that bundles those sources together.
- Compare engagement and key events against other channels over the same dates.
If your key events are not set up, start there. A proper analytics setup that records form fills, calls and purchases makes the data worth reading.
What Search Console can tell you
Google has said that clicks, impressions and position data from its AI features in Search are counted within the Performance report for the Web search type. In practice, that means AI-related activity is generally blended into your totals rather than broken out in a tidy separate view. Reporting options in Search Console have been evolving, so check the current report and Google's documentation instead of assuming a filter exists.
Even so, Search Console remains useful. Watch queries and pages over time, look for long, conversational queries appearing more often, and confirm your key pages are indexed without errors. Treat it as a health check rather than an AI scoreboard.
Manual prompt checks
The most direct way to see how assistants describe your business is to ask them. Keep a list of 10 to 20 prompts that match real customer questions, such as who offers emergency electrical repair in your city, and run them on a schedule across the tools your customers use.
- Record the date, tool, exact prompt, and whether your business was named or linked.
- Note whether the details stated about you are correct.
- Use a logged-out or fresh session where possible, and remember location affects results.
- Repeat monthly and look for patterns, not single results.
Answers vary between runs and between users, so this is a sample, not a measurement. Its biggest value is catching wrong information you can fix at the source.
Why guarantees are not credible
No outside party controls what an AI system says. Results differ by wording, location, time, personalization and model version, and they can change overnight. A promise of a first-place AI ranking or guaranteed citations describes something that does not exist as a stable, fixed position. Ask any vendor what exactly they measure, how often, and how they handle variation. Vague answers tell you what you need to know.
Be careful with third-party visibility trackers too. They can be useful for trends, but they rely on sampled prompts and should be treated as estimates. Ask how they choose prompts, how often they refresh, and whether you can export the raw answers.
Setting realistic expectations
Imagine a home cleaning company that runs its prompt checks for three months. In month one it is not mentioned. In month two a chat assistant names it but lists an old service area. In month three it appears in one tool and not in another. That is a normal, noisy picture, and none of it proves cause and effect.
Instead of chasing each change, look for steady signals: more branded searches, more conversational queries in Search Console, a few additional AI-source sessions, and growth in calls or form submissions. Pair that with fixes you control, such as correcting facts on your site and profiles.
When a client or manager asks whether AI work is paying off, an honest answer is that it is hard to isolate, and that you are watching leading indicators alongside revenue.
A simple reporting routine
Keep it small and tied to business results. Each month, record AI-source sessions and conversions from GA4, review Search Console query and page trends, and log your prompt checks. Then compare against leads, calls and sales in your digital marketing reporting.
If those outcomes are stable or growing, your fundamentals are working. If they are not, look first at page quality, tracking accuracy and your paid and local channels before chasing a metric nobody can fully see.
Key takeaways
- AI referral visits are partly trackable in GA4, but counts are incomplete.
- Search Console reporting on AI features is blended and evolving, so verify what it shows.
- Manual prompt checks are samples that help you catch errors, not rankings.
- Treat any guarantee of AI rankings as a warning sign.