Similarweb just released The Downstream Impact of AI Visibility, studying finance, travel, and beauty brands across AI-generated recommendations. The finding that matters: when AI recommended Capital One in credit card prompts, users were 14.2% more likely to visit Capital One's site within seven days. American Express saw a 7.2% lift. The mechanism isn't search clicks. It's direct visits, people typing the domain into their browser after an AI tool named the brand.
That's the same behavioral pattern outdoor advertising relied on for decades. Presence drives later consumption. The channel changed; the psychology didn't.
Why Mentions Beat Links in AI Visibility
Separate research on AI Overviews tells a sharper story about what drives these recommendations in the first place. Branded web mentions (including unlinked text references) show a 0.664 correlation with AI Overview visibility. Backlinks? Just 0.218. Roughly three times weaker.
The reason is structural. Language models learn from raw text across the web, not link graphs. A blog post that names your brand with real context carries weight even without a hyperlink. Brands in the top quartile for web mentions averaged 169 AI Overview appearances versus 14 for the next quartile down. Meanwhile, 26% of brands in the analysis registered zero mentions at all.
That gap is enormous. And it compounds: more mentions increase branded search volume, which strengthens authority signals, which feeds back into AI-generated answers. A flywheel, if you actually earn the initial mentions.
The Attribution Problem Nobody Has Solved
Here's the uncomfortable part. No platform currently provides defensible citations-to-revenue attribution. Not Similarweb, not SEMrush, not the newer AI visibility tools like Gumshoe or Siftly. The strongest proxy available is branded search lift, and that proxy is influenced by paid campaigns, PR, seasonality, and a dozen other variables.
Rand Fishkin flagged this directly in his commentary on the study: the data is aggregate, focused on large consumer brands (Sephora, American Express, Capital One). Whether the same effects hold for a mid-market B2B SaaS company with $15M ARR is genuinely unknown. The study also can't answer whether those users would have found those brands anyway, or whether AI presence outperforms ranking #1 in Google organic results.
Correlation isn't causation. Treat the 0.664 figure as directional signal, not proof. That said, directional signal backed by clickstream data from millions of devices is better than the nothing most teams are working with today.
What to Actually Do With This
If you're a CMO or VP Marketing at a B2B SaaS company, the play isn't to panic-buy an AI visibility tool. It's to run a structured experiment.
Step 1: Map where AI looks in your category. Identify the 3-5 sources AI systems cite most when answering prompts related to your product category. You can do this manually (run 20-30 prompts in ChatGPT, Perplexity, and Google AI Overviews, note which sources appear repeatedly) or use monitoring tools in the $25-$99/month range.
Step 2: Audit your named presence in those sources. Are you mentioned by name with real context? Being cited without being named educates the market while a competitor captures demand. Being named without credible sourcing looks popular but lacks proof. You need both: brand name plus substantive context in the same placement.
Step 3: Set up measurement. Track monthly: mention rate across AI responses, citation rate (are you sourced or just named?), and branded search volume as your primary proxy KPI. Gartner forecasts a 25% decline in traditional search traffic by 2026, so branded search lift is the metric that bridges the old measurement world and the new one.
The hypothesis (make it falsifiable): if we earn named mentions with real context in the top 3-5 sources AI cites in our category, then branded search volume will increase by 10-20% within 90 days, because AI systems will surface our brand more frequently in relevant prompts.
Success = branded search lift ≥10%. Guardrails = monitor that direct traffic quality (bounce rate, time on site) doesn't degrade. Stop-loss = if after 90 days branded search is flat and mention rate hasn't moved, the category may not have enough AI query volume to justify the effort. Redirect resources.
The Trade-Off Worth Naming
This approach requires shifting effort away from link-building toward content and PR programs that generate named mentions. For teams whose entire SEO strategy is built on backlink acquisition, that's a real reallocation. The risk: link-driven organic traffic dips before AI-driven branded demand picks up. Budget for the gap.
Ads, by the way, barely move the needle on AI visibility according to the research. Organic mentions in credible sources carry the weight. That's good news for teams with strong content operations and bad news for teams that rely on paid to compensate for weak brand awareness.
The Similarweb study is the first credible look at a question the industry has been hand-waving about for over a year. The answer, at least for large consumer brands, is yes: AI mentions drive real downstream behavior. Whether that holds for your specific market, your specific ACV, your specific buyer persona is something only your own data can tell you. The tools to measure it exist now, even if the attribution is imperfect. The teams that build the measurement muscle this quarter will have a 90-day head start on everyone who waits for perfect data that isn't coming.
