Key takeaways
- Clicks from AI answers are rare. Pew Research found that Google users clicked a link inside an AI summary in about 1% of visits where one appeared.
- Referral traffic is the floor, not the total. Branded search, direct visits and what customers say about how they found you fill in the rest.
- No single signal proves the effect. Look for several signals moving together after your AI mentions change.
- Plan the comparison before you act: a baseline period, a clear change and something left unchanged to compare against.
Why clicks tell only part of the story
Answer engines are designed to resolve the question on the page. When the answer already names three suitable products and explains the difference, there is little reason to click. The reader leaves with a shortlist in mind, not a link in the browser history.
Public data supports this. In a study of the browsing activity of 900 US adults in March 2025, Pew Research Center found that users clicked a traditional search result in 8% of visits when a Google AI summary appeared, against 15% when it did not, and clicked a link inside the summary itself in just 1% of visits. Chat assistants are not the same as search summaries, but the pattern is the same: people read, and most do not click.
So a brand that judges AI only by referral visits will undervalue it. The effect is real but delayed, and it shows up under other names in your reports.
The signals worth tracking
Each signal below catches a different path from an AI answer to your business. Each also has a blind spot, which is why they work best together.
| Signal | What it shows | Blind spot |
|---|---|---|
| AI referral visits | People who clicked a link in an answer, identified by referrer or source tag | Misses everyone who did not click; some apps pass no referrer |
| Branded search | People who learned your name somewhere and looked it up | Also rises with ads, press and word of mouth |
| Direct traffic | People who typed the address or used a shared link | A catch-all bucket with many unrelated causes |
| Self-reported source | What customers say when asked how they heard about you | Memory is imperfect and answer options shape responses |
| Sales and support notes | Buyers who mention an AI assistant in calls or chats | Only captured if staff record it consistently |
Read together, the signals form a pattern. When AI mentions rise and branded search, unexplained direct visits and survey answers naming an AI assistant all rise in the following weeks, the case is strong. When only one of them moves, look for another cause first.
Setting up the measurement
- Separate AI referrals in analytics. Create a channel group for referrers such as chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com, and keep the list updated as new assistants appear.
- Track branded search every week. Use Google Search Console and Naver Search Advisor for impressions of queries that contain your brand name, and keep the same query list over time.
- Ask customers. Add a how-did-you-hear-about-us question to sign-up, checkout or the first sales call, with AI assistants as an explicit option next to search, social and referrals.
- Record your AI mentions on the same timeline. Measure how often you are mentioned for your key questions and plot it next to the traffic and survey data.
- Note everything else you change. Campaigns, price changes, press coverage and seasonal peaks all move the same signals, so they need to be on the record.
Designing a fair before-and-after comparison
The strongest evidence comes from a deliberate change. Suppose, for example, you publish clearer comparison pages for one product line and your mentions in answers to questions about it rise over the following weeks. If branded search and self-reported AI discovery rise for that product line but not for the others, the link is far more convincing than a rise across the board.
- Keep a comparison group. Leave some product lines, regions or question groups untouched so you can see what would have happened anyway.
- Use a long enough baseline. Several weeks before the change show normal movement and seasonal patterns.
- Allow for delay. People who read an answer may act days later, so compare weeks, not days.
- Avoid overlapping launches. A big campaign in the same weeks makes the result impossible to read.
Even a well-designed comparison gives a direction, not a precise multiplier. That is enough to decide whether to keep investing, and more honest than a single impressive number.
What changes in the Korean market
In Korea, much of the follow-up happens inside Naver and KakaoTalk. A reader who sees a brand in an AI answer often searches for it on Naver, reads blog reviews and then visits, which shows up as Naver search traffic rather than AI. Links shared in KakaoTalk chats usually arrive as direct traffic. Treat rises in Naver brand searches and unexplained direct visits as part of the evidence, not as separate stories.
How flowgen connects mentions to results
GEO Insight measures how often you are mentioned for the questions that matter. Analytics places those mentions next to AI referrals, branded search and your own releases on one timeline, so before-and-after comparisons are easy to read. AI Marketer then suggests which product lines or questions to work on next.