Brands can now see how often AI assistants mention them. The harder question is whether those mentions do anything. Flowgen built this study to measure downstream behaviour directly: we matched AI responses that mentioned a brand to the same user's subsequent web browsing, and compared it with what that user would have done anyway.
Key takeaways
- AI mentions generate 1.5–2.5x baseline brand-site traffic over the following 7 days.
- Gemini shows the largest relative lift (~2.5x); Google AI Overviews the largest absolute volume; ChatGPT the most consistent uplift (38–86% by industry).
- 20.5% of downstream visits occur within 1 hour of the mention and 42% within 24 hours.
- Only ~2.5% of visits carry trackable AI-referral parameters, even after ChatGPT's May 2026 link update.
The data behind this study
The analysis covers more than 2 million AI conversations from January to June 2026 across ChatGPT, Gemini and Google AI Overviews. Data comes from a double-opt-in, privacy-compliant US panel that captures both the AI interaction and the user's browsing stream, so we can see what happens after a brand is mentioned.
Definitions
AI-exposure: a brand mention appearing in an AI response where the brand was not in the user's prompt. Match: a subsequent visit to the mentioned brand's website by the same user within the window.
Experimental setup
We used a forecasted backward-placebo design. For every exposure we compared the 7-day post-exposure visit rate with the user's visit rate in the three preceding 7-day "placebo" windows, forecasting a baseline from those windows. Confidence intervals were estimated by bootstrap with 2,000 replicates.
Over 7 days, AI mentions drive up to 1.5–2.5x brand site visits
All three platforms show a measurable uplift. Gemini delivers the largest relative increase, Google AI Overviews the largest absolute volume, and ChatGPT a consistent lift across categories.
| Platform | Treated | Baseline | Lift (95% CI) |
|---|---|---|---|
| Gemini | 5.42% | 2.21% | +3.21 pp [2.34, 4.08] |
| Google AI Overviews | 7.79% | 4.83% | +2.96 pp [2.76, 3.17] |
| ChatGPT | 6.39% | 4.33% | +2.07 pp [1.67, 2.48] |
The uplift varies by industry
Gemini drives the strongest relative lifts in retail and financial services, while software and telecom show platform-specific variation. Percentage-point and relative uplift by industry and platform:
| Industry | Google AI Overviews | ChatGPT | Gemini |
|---|---|---|---|
| Financial Services | +3.7 pp (+59%) | +3.8 pp (+59%) | +5.7 pp (+132%) |
| Retail | +3.5 pp (+52%) | +2.6 pp (+38%) | +5.6 pp (+140%) |
| Software | +4.0 pp (+128%) | +3.0 pp (+59%) | +3.6 pp (+107%) |
| Telecom | +3.0 pp (+137%) | +1.5 pp (+85%) | +2.3 pp (+71%) |
Many downstream visits happen quickly
Across all platforms, 20.5% of first brand-site visits happen within one hour of the mention and 42% within 24 hours. Google AI Overviews is the most immediate (45.7% same-day), ChatGPT moderate (38.7%) and Gemini slowest (30%). Because the majority of visits still happen after day one, a 7-day window is the right measurement horizon.
Tracking click-throughs misses most of the traffic
ChatGPT's May 2026 link update increased the share of visits carrying visible referral parameters from about 1.0% to 2.47% in June. Even so, more than 97% of these brand-site visits still did not include a UTM. The uplift itself was stable on both sides of the change:
| Period | Treated | Baseline | Lift (95% CI) |
|---|---|---|---|
| Before May 7 | 6.26% | 4.26% | +2.00 pp [1.52, 2.52] |
| After May 7 | 6.86% | 4.71% | +2.16 pp [1.35, 3.01] |
Direct attribution captures a sliver of the traffic. AI works more like offline advertising: the impact surfaces hours or days later as an "organic" visit.
What this means for your brand
- Treat AI mentions as a demand driver, not just a vanity metric: the lift is measurable and persists for a week.
- Match strategy to platform effectiveness in your category: Gemini for relative lift, AI Overviews for volume, ChatGPT for consistency.
- Measure visibility, behaviour and context together. Referral parameters alone will undercount AI's impact by more than 30x.
Methodology
Double-opt-in US panel tracking AI interactions and page views, January–June 2026. Forecasted backward-placebo design comparing the 7-day post-exposure visit rate with three prior 7-day windows. Bootstrap confidence intervals (2,000 replicates). We excluded users with brand searches or visits in the prior week and removed generic domains. Caveat: this is an observational site-visit study, not a randomized experiment, so it does not fully eliminate selection bias.