Similarweb this week published a study tracing what happens to website traffic after a user receives an AI recommendation - and the numbers challenge the assumption that AI visibility and measurable business outcomes belong in separate conversations.

The report, titled The Downstream Impact of AI Visibility, follows real user journeys across three industry verticals and six months of browsing data. It is the first study to measure the connection between a ChatGPT recommendation and the subsequent visit to a brand's website using panel-based clickstream data rather than modeled estimates or survey recall. The findings carry direct implications for how marketing teams set budgets, run attribution models, and split investment between organic search protection and paid brand campaigns.

The problem the study sets out to solve

The report frames its central question plainly: every brand is asking whether it appears in AI answers. The logical follow-up - what does being visible in AI answers actually do for a brand - has gone largely unanswered because the standard tools do not capture the answer.

According to Similarweb, AI platforms do not send traffic the way search engines do. The predominant journey is a user asking a question, reading the answer, and ending the session. No click is recorded. No referral is attributed. Standard analytics show almost nothing. The user visits the brand's website days later via a different channel entirely, with no technical thread connecting the two events.

The study documents this gap using a four-stage flow: a user asks an AI for a product recommendation, the AI responds and the session ends with no click recorded, the user visits the brand the AI recommended days later, and the traffic arrives through a channel - most often search - that attribution models have no way to connect to the original conversation.

What the study measured and how

The data foundation is Similarweb's proprietary real-user panel, which tracks the actual browsing behavior of opted-in users across desktop web in the United States. According to Similarweb, user journey data was collected between July 2025 and December 2025, covering a six-month window. A supplemental survey - focused on the AI-to-visit discovery stage - was conducted in January 2026.

The industries selected were Finance, Travel, and Beauty. Within each vertical, brand pairs were chosen based on competitive overlap: two brands operating in the same category, targeting similar audiences, and competing for the same search queries. The Finance pair was American Express and Capital One. Travel covered Skyscanner and Kayak. Beauty covered Sephora and Ulta. This pairing approach allows the study to compare outcomes for brands that received a ChatGPT recommendation against those that did not, within an identical market context.

The methodology for identifying AI recommendation exposure is precise. According to Similarweb, the qualifying condition was users who submitted queries to ChatGPT and received a response that included a recommendation for a specific brand. Users who mentioned the brand name in their prompt were excluded - removing instances where the user already had a brand in mind before the AI conversation began. That exclusion strengthens the causal reading: the AI recommendation itself, not prior brand awareness, is the variable being tested.

The attribution window is seven days. Downstream visits were tracked for seven days following the AI conversation. AI-influenced visits are defined as visits to a recommended brand's site within this window by a qualifying new user. "New user" carries a strict definition: only users with no recorded visit to the brand's domain in the prior four weeks were included, to ensure traffic uplift represents new acquisition rather than returning customers.

The study describes itself as the first to connect AI visibility to downstream behaviour using real user data.

The headline numbers

Three figures anchor the study's public findings.

Brands recommended by ChatGPT are 2.5 times more likely to receive a visit within the following seven days compared to brands not recommended. That multiplier holds across all three industries studied. The effect is also symmetrical: when the competitor brand received the recommendation instead, the traffic went to them. According to Similarweb, this zero-sum dynamic means AI visibility is not a soft brand metric - it is a traffic driver with a measurable impact on site visits that operates on a delayed timeline standard referral tracking cannot capture.

The second headline figure is that 55.9% of AI-influenced traffic arrives via search. Not as a direct referral from ChatGPT. Not via an AI citation link. Via a branded search query, typed into Google or another search engine, after the AI conversation has already ended. In the non-AI-influenced cohort, search accounts for 40.4% of visits. The gap - 55.9% versus 40.4% - quantifies the extent to which AI recommendations redirect user behavior toward branded search as an intermediary step.

The third figure is that AI-influenced visitors spend 2 times longer on site. The specific numbers from the report's chart: AI-influenced visitors average 12.0 pages per visit and 11.8 minutes on site. Non-AI-influenced visitors average 6.5 pages and 5.6 minutes. The engagement differential is consistent across both metrics.

The channel mix in full

The report publishes a complete channel breakdown for AI-influenced visits versus non-AI-influenced visits, covering five channels measured across US desktop from July 2025 to December 2025.

For AI-influenced visits, the channel distribution is: Search 55.9%, Direct 19.9%, Referrals 13.5%, AI 8.8%, Other not separately stated. For non-AI-influenced visits, the distribution is: Search 40.4%, Direct 38.8%, Referrals 10.5%, AI 5.0%, Other 5.3%.

Several aspects of this breakdown are worth examining separately. Direct traffic - users typing a URL or using a bookmark - accounts for 19.9% of AI-influenced visits compared to 38.8% of non-influenced visits. That inversion is significant. Non-influenced visitors are more likely to navigate directly, which typically indicates stronger prior brand familiarity. AI-influenced visitors, by contrast, are new users - defined as having no prior visit in the last four weeks - so they route through search to find the brand rather than navigating directly.

The AI channel itself accounts for only 8.8% of AI-influenced visits. That figure represents the direct referral link from an AI platform - the kind of traffic that shows up in standard analytics as an AI referral. It is, by far, the smallest majority-attributable portion of the downstream effect. The largest share, by a factor of more than six, arrives through search.

According to Similarweb, this pattern demonstrates that users remember the recommended brand and actively seek it out rather than being directed there by an AI referral. It is a behavioral signal of intent, not a passive click-through.

The brand-pair data

The study publishes visit rates for each brand pair, showing what percentage of users visited the AI-recommended brand versus the competitor brand within seven days of the ChatGPT conversation.

In Finance, when ChatGPT recommended American Express, 7.2% of users visited American Express within seven days while 3.1% visited Capital One. When the recommendation switched to Capital One, 14.2% visited Capital One and 3.8% visited American Express. The asymmetry in absolute visit rates - Capital One receiving 14.2% when recommended versus 7.2% for American Express - may reflect differences in the starting awareness levels of each brand, in the specificity of the queries triggering each recommendation, or in competitive dynamics within the Finance category. The study does not attribute causation to any single factor.

In Travel, when ChatGPT recommended Skyscanner, 9.5% of users visited Skyscanner and 7.6% visited Kayak. When Kayak was recommended, 12.0% visited Kayak and 3.4% visited Skyscanner. The gap widens considerably in the Kayak-recommended scenario: a 3.5x differential compared to approximately 1.25x in the Skyscanner-recommended scenario.

In Beauty, when Sephora was recommended, 7.9% of users visited Sephora and 3.3% visited Ulta. When Ulta was recommended, 7.6% visited Ulta and 4.6% visited Sephora. The Beauty figures show the narrowest competitive differentials of the three verticals, with competitor visit rates remaining relatively high regardless of which brand received the recommendation.

Across all six brand scenarios, the recommended brand consistently outperforms its competitor in downstream visits. The 2.5x average multiplier holds as an aggregate across the full dataset.

Where AI sits in the consumer journey

The January 2026 survey adds a self-reported dimension to the behavioral panel data, asking users at which stage of the purchase journey AI tools and search engines are most useful.

According to Similarweb, for the "discovering/getting initial ideas" stage, 35.0% of users said AI tools are most useful at that point, compared to 13.6% for search engines. For the "researching/comparing options" stage, the split narrows: 30.0% for AI tools versus 20.0% for search engines. For "finding where to buy/the best price," the gap narrows further still: 24.3% for AI tools versus 22.1% for search engines.

The pattern is directionally clear. AI dominates the earliest stage of the consumer journey. Search catches up as the journey progresses toward transaction. According to Similarweb, this explains the attribution gap mechanically: AI influences users at the start of their journey, and the traffic comes later - when they turn to search to find the brand they have already decided to visit.

The report describes this as confirming why relying on AI referrals alone to measure the impact of AI visibility misses the bigger picture. The referral, when it exists, occurs at the wrong stage of the journey to capture the full effect.

The engagement data examined

The 2x session duration finding is supported in the report by two parallel metrics: pageviews and time on site. AI-influenced visitors view 12.0 pages on average compared to 6.5 for non-AI-influenced visitors - an 85% increase. Time on site runs 11.8 minutes for AI-influenced visitors against 5.6 minutes for non-AI-influenced visitors - a 111% increase.

According to Similarweb, these users have already done their research in the AI conversation, narrowed their options, and chosen a brand before they reach the site. That intent shows in the numbers. Standard visitors, by contrast, are still in research mode: they browse, compare, and leave without the same depth of engagement.

The report notes a direct commercial implication from this engagement gap. AI-influenced visitors arrive with higher intent, which means invisible brands - those absent from AI recommendations - hand that higher-intent engagement to competitors. The framing is explicit: invisible brands do not just miss the mention. They lose the visit.

Rand Fishkin on the measurement shift

Rand Fishkin, co-author of Zero Click Marketing and founder of SparkToro, Alertmouse, and Snackbar Studio, contributed to the study and is quoted at length in the report. According to Fishkin: "This study is a near-replica of how advertisers in the 20th Century measured the impact of billboards, TV, and radio advertising using lift in store visits or sales. Whether we're talking about a brand's visibility on the side of a highway in 1926 or its presence in an AI tool's response in 2026, it's clear that influence is happening. What marketers must do now is change the way they run measurement and attribution."

Fishkin adds a separate observation on methodology: "I have but one addition to this excellent advice - marketers must uncover the platforms their audiences are using to discover brands and the prompts (or, at least, the prompt intents) they're typing in. Luckily, clickstream panel data is an excellent way to do just that."

The billboard analogy is technically precise. Offline brand awareness measurement has, for decades, relied on lift studies: exposing a panel to brand messaging, then tracking behavioral change in a subsequent window. The Similarweb study applies the same logic to AI recommendations, using a seven-day attribution window to measure lift in site visits. The methodology is not novel in principle - it is novel in context.

Five steps for closing the attribution gap

The report includes a five-step framework for marketers seeking to connect AI visibility to measurable outcomes. According to Similarweb, the steps are:

First, stop measuring AI by referral traffic alone. Standard analytics only tell part of the story. AI-driven visits often arrive through search, direct, and other channels that appear disconnected from the original AI conversation.

Second, treat visibility as a leading indicator. AI visibility creates intent that surfaces later as traffic and revenue. The brands gaining an advantage are the ones measuring that connection before it appears in traditional attribution models.

Third, protect the brand on the SERP. According to Similarweb, 56% of AI-influenced visits arrive via search. Branded terms should be optimized and defended with paid search to prevent competitors from capturing demand generated by AI recommendations.

Fourth, connect recommendations to outcomes. Understanding AI's impact requires tracking real user journeys, not just prompts and rankings. Following users from recommendation to site visit is what establishes commercial value.

Fifth, benchmark against competitors. AI visibility is zero-sum. Measuring share alone is not enough. The key question is where a brand is being recommended instead of competitors, and where competitors are being recommended instead.

Three implications for CMOs

The report identifies three direct implications for chief marketing officers, framing AI visibility explicitly as a performance issue rather than a brand awareness metric.

The first is that visibility is a performance channel. According to Similarweb, AI recommendations drive downstream site visits with measurable impact on traffic. This is not brand awareness in the abstract - it is a channel with commercial outcomes.

The second is that attribution models have a blind spot. AI-driven traffic arrives via search and direct, not via AI referrals. Without a new measurement framework, the ROI of AI visibility is systematically underreported.

The third is that invisibility hands traffic to competitors. The dynamic is zero-sum. The visit happens regardless of whether the recommended brand measures it. If a brand is not recommended, a competitor is - and the competitor receives the traffic, the engagement, and the conversion opportunity.

Adelle Kehoe on what comes next

The report's "Looking forward" section is attributed to Adelle Kehoe, Director of Product Marketing at Similarweb. According to Kehoe, the last 12 months have been a scramble to gain visibility in AI search, and businesses are now starting to ask a more important question: what value does that visibility actually create?

Kehoe's commentary identifies three forward-looking priorities: visibility itself, referral traffic, and paid advertising. On referral traffic, Kehoe notes that there is not much of it, but the rules are changing. According to Kehoe, ChatGPT has recently experimented with outbound links for recommended brands, generating roughly three times more outbound traffic. The volumes remain small, but the trend matters.

On paid advertising, Kehoe notes that both ChatGPT and Google now offer brands ways to pay for visibility in AI experiences. According to Kehoe, much like the SEO and SEM relationship, paid placements can help brands close visibility gaps and secure presence in critical conversations. Success will increasingly depend on closer collaboration across marketing teams to balance visibility, traffic, and spend.

The report's closing takeaway section states that AI visibility is becoming a competitive advantage, and that the brands measuring it, benchmarking it, and acting on it will be best positioned to capture demand as consumer discovery continues to shift.

The attribution gap and its technical context

The 55.9% branded search figure is the most technically consequential finding in the study. When a user asks ChatGPT a question in a category - travel insurance, moisturizers, credit card rewards - and receives a recommendation that includes a specific brand name, that user does not necessarily click a link inside ChatGPT. According to Similarweb's panel data, the majority do not. Instead, they open a new browser tab and search for the brand by name.

That branded search query arrives in Google Analytics, Search Console, or any third-party attribution tool as organic branded traffic. There is no referrer from ChatGPT. No UTM parameter. No session chain. The AI interaction, which directly caused the visit, is invisible to the attribution system.

This mechanism has been documented at a technical level in related contexts. When Google fixed an AI Mode tracking bug in May 2025, the fix addressed stripped referrer headers in AI-powered search links that were causing clicks to register as direct traffic in analytics platforms. A parallel problem exists across any AI interface that does not produce a clickable, tracked outbound link - including conversational chat sessions where the user manually copies a brand name and searches for it independently.

Google Analytics in August 2025 began formally suggesting that marketers set up custom channel groups specifically for AI chatbot traffic - separating visits from known AI referrer URLs into dedicated buckets. That capability captures the 8.8% of AI-influenced visits that produce a direct referral. It does nothing to capture the 55.9% that arrive through branded search.

The branded search finding also reframes what "protecting brand in search" means when AI is part of the customer journey. It is not only a defense against competitor bidding on branded keywords, which has been a standard paid search concern for years. It is an infrastructure requirement for capturing the downstream value of AI visibility. The attribution anomalies created by Google's AI Max search campaigns already complicate the relationship between branded keyword coverage and traffic attribution. As AI-driven branded search queries grow in volume, the commercial stakes of brand protection in search increase alongside them.

Why this study differs from prior measurement approaches

The marketing community has spent much of 2025 and early 2026 building tools to track what is observable: citation counts, visibility scores, referral URLs from AI platforms. Semrush documented how the company tripled its own AI share of voice from 13% to 32% in one month through systematic optimization, while acknowledging that tying visibility improvements to revenue remained "extremely complicated." Amplitude launched AI Visibility tracking in October 2025to connect AI mentions to conversions within its analytics platform. Similarweb published an AI citation analysis framework in November 2025, documenting that citation sets change 50% monthly and that only 11% of citations overlap across major AI platforms.

All of these tools operate on the AI side of the interaction. They measure what the AI platforms do: which brands appear, how often, in what contexts. The downstream study is the first to follow users from the AI conversation through to the eventual website visit, across channels that have no technical connection to the original AI session.

Similarweb launched its GenAI Intelligence Toolkit in July 2025, a dual-tracking platform measuring both brand visibility in AI answers and the referral traffic that results from AI platform clicks. That product connected two layers - visibility and traffic. The downstream study adds a third: the branded search journeys that occur between the AI recommendation and the eventual site visit, which neither AI-specific analytics nor standard traffic attribution currently captures.

ChatGPT referrals to news publishers increased 25 times year-over-year between January 2024 and May 2025, according to Similarweb's own earlier data, while zero-click searches on Google climbed from 56% to nearly 69% over the same period. The two trends compound: AI platforms send more referrals, but more of those referrals are never recorded as such.

Research published by NP Digital in May 2026 found that original research is cited in AI search at a rate of 82%, leading all content formats by a wide margin. Video content scored 2%. The downstream study now quantifies what the content visibility gap costs in terms of subsequent website traffic and engagement.

For small publishers, the consequences of reduced AI visibility compound existing pressures. Chartbeat data published in April 2026 shows small publishers lost 60% of search traffic over a two-year period, while ChatGPT referrals still account for under 1% of total publisher page views. The gap between visible referral traffic and invisible downstream behavior is precisely what the Similarweb panel methodology is now able to measure. AI share across the major platforms has itself shifted dramatically - ChatGPT fell from 76.4% of generative AI web traffic in June 2025 to 52.7% by May 2026 as Gemini climbed to 27.3%, expanding the range of platforms across which downstream effects must be measured.

The study team

The study was produced by Similarweb's Market Insights division. According to Similarweb, the research team includes Laurie Naspe as Director of Market Insights, Daniel Reid as Principal Analyst, Sam Sheridan as Data Analyst, Rice Tong as Product Marketing Manager for AI Search Intelligence, and Darrell Mordecai as Content Marketing Manager for Web Intelligence. Rand Fishkin contributed in an advisory capacity and is quoted directly in the report.

Adelle Kehoe, Director of Product Marketing at Similarweb, represents the report externally. Sam Sheridan, described by Similarweb as having a background in data science with previous experience in the retail and CPG sectors, designed the data analysis.

Timeline

Summary

Who: Similarweb, the competitive intelligence platform headquartered in Tel Aviv and New York, produced the study. The core research team is led by Laurie Naspe (Director, Market Insights), with data analysis by Sam Sheridan and contributions from Daniel Reid, Rice Tong, and Darrell Mordecai. Rand Fishkin, co-author of Zero Click Marketing and founder of SparkToro, contributed analysis and is quoted directly. Adelle Kehoe, Director of Product Marketing at Similarweb, wrote the report's forward-looking commentary.

What: A panel-based clickstream study tracking real user journeys on desktop web in the United States from July 2025 through December 2025, supplemented by a January 2026 survey. The study follows users who received a ChatGPT brand recommendation across Finance, Travel, and Beauty verticals - specifically American Express vs Capital One, Skyscanner vs Kayak, and Sephora vs Ulta - and tracks their website visits in the following seven days. Key findings: AI-recommended brands are 2.5x more likely to receive a visit; 55.9% of that downstream traffic arrives via branded search; AI-influenced visitors average 12.0 pages and 11.8 minutes on site compared to 6.5 pages and 5.6 minutes for non-AI-influenced visitors; AI tools are rated most useful for discovery by 35% of users versus 13.6% for search engines.

When: User journey data was collected between July 2025 and December 2025. The discovery-stage survey was conducted in January 2026. The full report was published on June 21, 2026.

Where: The study covers desktop web browsing in the United States. The three verticals are Finance, Travel, and Beauty. Brand pairs were selected within each vertical based on competitive overlap.

Why: Standard analytics systems cannot connect AI recommendations to downstream website visits because the majority of AI-influenced traffic does not arrive through a trackable referral link. Instead, users remember a recommended brand from an AI conversation and search for it by name days later - appearing in analytics as organic branded search, not as AI-influenced traffic. The 55.9% branded search figure quantifies the scale of that structural attribution gap and establishes why measuring AI visibility only through direct referral traffic systematically underreports its commercial impact.