On this page
Here is an illustrative version of a message we get often: a brand’s AI visibility score climbs for the third month running, and in the same stretch their AI referral sessions in GA4 quietly drop. Nobody touched the tracking setup, the product did not change, and both numbers are being read correctly. The instinct is to assume one of them is wrong. Neither is. They are measuring two different moments in the same interaction, and once you separate those moments the divergence stops being confusing and starts being informative.
Two different questions, two different instruments
A presence score, the kind behind most AI visibility tools including ours, answers one question: across a fixed set of prompts run on a schedule, how often does an engine name your brand at all? It is built the way our share of voice methodology works, sampling answers rather than watching real users, and it counts a mention whether or not that mention carried a clickable source, and whether or not anyone reading the answer ever visited your site.
AI referral traffic answers a completely different question: of the people who read an AI answer, how many clicked a link inside it and landed on your site? That number lives in your analytics, not in a tracker, and it depends on a chain of conditions a presence score never touches, whether the engine attached a citation to that particular answer, whether the citation pointed at you specifically, whether the platform passes a referrer at all, and whether the reader needed to click through in the first place. Our guide to tracking AI referral traffic in GA4 walks through what that chain looks like end to end.
Once you hold both definitions next to each other, a rising score and falling traffic are not a contradiction. They can both be true because they are describing different halves of the same conversation: the engine is talking about you more, and the person on the other end is acting on that less.
The click is getting harder to earn everywhere
Some of the fall in referral traffic has nothing to do with your specific visibility and everything to do with what AI answers do to clicking in general. Pew Research tracked 68,879 real Google searches from 900 consenting US adults and found that when an AI summary appeared above the results, users clicked a traditional result only 8% of the time, against 15% when no summary appeared, and clicked a link inside the summary itself in just 1% of visits (Pew Research, 2025).
That gap has been widening, not narrowing. Ahrefs’ own click-tracking found that a top-ranking page’s click-through rate fell 34.5% when an AI Overview was present, in an April 2025 measurement, and that by December 2025 the same comparison showed a 58% drop (Ahrefs, 2026). None of this is specific to your brand or your content. It is the shape of the channel now, and a visibility score that only measures whether you are named will not reflect it, because it was never built to.
Being cited is not the same as being clicked
It helps to be more careful than “cited or not” when you look at the click side, because citation still matters, just less than intuition suggests. Seer Interactive’s tracking of AI Overview queries found that pages the Overview actually cited earned a 2.07% average organic click-through rate, more than double the 0.94% earned by uncited pages ranking on the same results page, but still well under the 3.35% baseline on comparable queries with no AI Overview at all. The same tracking shows the floor itself moving: organic CTR on AI Overview queries fell from 3.19% in January 2025 to a low of 1.31% in December 2025, before climbing back to 2.36% by February 2026 (Seer Interactive, 2026).
So a citation is worth having, it roughly doubles your odds of a click compared with an uncited mention next to it, but it still leaves most of the audience reading the answer and moving on. A presence score that counts the citation as a win is counting correctly. It is just not counting the same thing as a session in your analytics.
Chat assistants are stingier with citations than Google
The gap is starker still outside Google. Similarweb’s analysis of ChatGPT sessions found that fewer than 7 in 100 prompts included any citation link back to a source at all, as of May 2026, compared with Google’s AI Overviews, which attach a sources panel to nearly every result (Similarweb, 2026). A brand can be named constantly inside ChatGPT’s prose, the exact behavior a presence score is designed to catch, without a single one of those mentions ever offering the reader something to click.
This is why a score that is genuinely strong on ChatGPT and genuinely weak on referral traffic is not a red flag by itself. It may simply mean the engine where you are winning happens to be the one that links out the least. Our piece on why AI visibility tools report different scores covers a related confusion, two tools disagreeing about the same brand, and the same root cause applies here: the number is only meaningful once you know exactly what it counted.
What is actually moving when your score goes up
A presence score rises when more of your tracked prompts return your name, across more engines, more consistently over the run. That can happen because you earned a genuinely new mention, because a competitor’s product changed and freed up space in the answer, or because the specific engines you gained ground on happen to answer more of your prompt set that month. None of those reasons require a single click to have happened, and none of them are captured anywhere in GA4.
Referral traffic, meanwhile, moves on its own separate schedule, driven by how many of the answers that named you also carried a working citation, how many of those citations pointed specifically at your domain rather than a directory or review site about you, and how many readers needed the click badly enough to make it. A brand can improve steadily on the first axis for months before the second one notices, simply because most of what a presence score picks up never had a clickable link attached to begin with.
Read the two numbers as a pair, not a scoreboard
The practical fix is to stop expecting one line to confirm the other and start reading them as a pair that tells you where in the funnel something changed. If your score is rising and referral traffic is flat or down, check whether the gain came from engines or answer types that rarely cite at all, that is very often the whole explanation, and pull up your AI referral traffic report to see whether citations existed but simply weren’t clicked. If both numbers fall together, that is a real problem worth investigating, since it means you are losing presence, not just losing clicks on presence you kept. Our guide to tracking brand presence in AI search covers how to set up that kind of run so it is repeatable enough to trust month over month.
If you are not yet tracking presence separately from clicks, a free Visibility Check gives you a same-day baseline across engines, and the AI visibility tracker page covers what a full monitoring setup looks like once you are ready to watch the trend rather than take one reading. Treat the score as the answer to “does the engine know about us,” and treat referral traffic as the answer to “did that particular answer make someone act.” They will not always move together, and that is not a bug in either number.