3.4 billion sessions point the same way: AI may be moving websites downstream

Earlier today, AEO Updates examined a provocative pattern in Brainlabs data from 54 advertisers: organic sessions fell while direct AI referrals grew, and the people arriving from AI generated advertiser-defined key events at a higher rate than organic-search visitors. The article proposed a cautious interpretation: <a href="/articles/ai-search-website-downstream-brainlabs">AI search may be moving the website later in the decision journey</a>, after answer engines have already handled more education, comparison and consideration.

That was a hypothesis derived from one agency dataset. A separate First Page Sage report now points in the same direction using a much larger client panel. Its analysis covers more than 3.4 billion sessions from 218 websites across 12 industries between January 2023 and July 2026. It reports lower organic traffic, rapidly growing AI referrals, fewer clicks when Google AI Overviews appear, stronger post-click behaviour among the people who still visit, and dramatically different effects by search purpose.

The two studies do not use the same sample, definitions, time period or methodology. They should not be pooled into a single benchmark. Nor do they prove that AI caused the observed traffic changes. What they provide is something more defensible: two independent agency datasets in which traffic volume falls while the remaining or AI-referred visitors appear more advanced in their journeys.

<h2>A larger dataset strengthens the pattern, not the causal claim</h2>

First Page Sage says it normalised its panel for site growth, seasonality and advertising spend, excluded sites that entered or left during a year, and defined AI-platform traffic as sessions whose referrer resolved to an AI assistant. Organic search represented 51.3% of panel sessions in January 2023 and 42.8% in July 2026. Normalised organic session volume fell 23.6%, while AI-platform traffic rose from 0.1% to 6.2% of sessions and increased 5,580% from a tiny starting point.

The quarterly series shows how quickly the observable handoff channel developed. AI platforms accounted for 0.11% of sessions in the first quarter of 2023, crossed 1% in the fourth quarter of 2024, reached 5.13% in the first quarter of 2026 and represented 6.24% in the second quarter of 2026.

Those figures describe First Page Sage’s client panel. They are not a market-wide census. The company also calculates traffic loss against a counterfactual baseline derived from the panel’s pre-2023 growth rate. A site can therefore be classified as losing traffic even if its absolute visits increased, if it grew more slowly than the counterfactual predicted. The report documents an important change within a large dataset, but the size of that change depends partly on the baseline model.

<h2>AI Overviews were associated with fewer clicks and different visitors</h2>

First Page Sage compared the same keyword set and date range when a Google AI Overview appeared and when it did not. Average organic click-through rate was 8.7% without an AI Overview and 3.6% with one. That is 87 versus 36 organic clicks per 1,000 impressions, a 58.6% reduction. The zero-click rate increased from 62.7% to 82.4%.

The people who still clicked behaved differently. Average session duration increased from 2 minutes 41 seconds to 3 minutes 18 seconds. Pages per session rose from 2.3 to 2.9, while visitor-to-lead rate increased from 1.7% to 2.6%.

It would be easy to turn those numbers into the comforting claim that better visitors compensate for lost traffic. They did not. Applying the reported click and lead rates implies that lead volume still fell by approximately 37%. The higher visitor-to-lead rate recovered part of the loss, but nowhere near all of it.

This is the first important connection with the Brainlabs article. Both datasets report fewer conventional organic visits alongside stronger downstream behaviour among the smaller group that arrives from AI-influenced or AI-referred journeys. Neither establishes that the answer engine made those people better prospects. Selection is an obvious alternative explanation: people who receive everything they need inside the AI experience do not click, while those who still need pricing, availability, a consultation, an account or a transaction have a reason to visit.

<h2>The purpose data provides the strongest clue</h2>

The most consequential part of the First Page Sage report is its segmentation by search purpose. Informational queries represented 46.2% of the panel. Organic sessions associated with them fell 43.9%, and 91.7% ended without an external click. Transactional queries represented 11.7% of the panel, with sessions down only 5.7% and a 37.2% zero-click rate.

Navigational and commercial-investigation searches sat between those extremes. Navigational sessions fell 19.4%, with a 76.3% zero-click rate. Commercial-investigation sessions declined 14.2%, with a 58.1% zero-click rate.

This does not reconstruct individual journeys. First Page Sage uses its own intent-classification framework, and the study is observational. But it provides a plausible explanation for why a website could receive fewer visits while the visitors who remain appear more commercially advanced. The largest disruption is associated with learning. Far more of the traffic associated with acting remains.

That distinction matters for AEO measurement. Two brands can each record 50% Visibility while occupying completely different parts of the decision process. One may dominate broad category education. The other may remain present in comparison, shortlisting and recommendation prompts. The percentages are mathematically identical, but their commercial proximity is not.

Purpose should therefore be treated as a core Prompt Taxonomy dimension, not merely a reporting filter. A serious measurement programme needs to show where Visibility occurs and how performance changes as prompts move from informational questions towards consideration, recommendation and action.

<h2>Direct AI referrals look like handoffs, not replacement traffic</h2>

First Page Sage also reports stronger downstream behaviour for visitors arriving directly from named AI systems. ChatGPT visitors had a 4.7% visitor-to-lead rate, compared with 1.9% for Google organic, approximately 2.47 times higher. Perplexity recorded 4.1%, Claude 3.9%, Gemini 2.8% and Copilot 2.4%. Claude produced the highest lead-to-customer rate at 22.6% and the longest average session duration at 6 minutes 41 seconds.

These are panel-specific rates based on First Page Sage’s standard definition of a marketing-qualified lead. They do not establish that one AI system causes stronger prospects than another. The traffic mix, industries, client selection, purchase cycles and reasons for clicking can all affect the results.

The more useful interpretation is that an AI referral records a particular transition. A person may have already asked questions, refined requirements, compared alternatives and received a recommendation before opening the website. The click marks the moment the AI environment hands the person to a brand-controlled experience for something it cannot complete itself.

That makes AI referral traffic a potential <strong>handoff metric</strong>. It does not measure the total AI audience or every decision influenced by an answer engine. It measures the subset of journeys in which the AI system sends an observable referral to the website.

The limits of referral data become clear when a user discovers a brand in ChatGPT, closes the application, searches the brand on Google and later converts. Conventional analytics may credit organic search. If the user types the URL, the session may appear as direct. The AI influence disappears even though it helped shape the decision.

<h2>Industries are not moving at the same speed</h2>

First Page Sage reports substantial variation across 12 industries. Organic sessions fell 47.3% for media and publishing, 41.8% for education and 38.2% for healthcare. Local services declined 6.8%, real estate 9.2%, and ecommerce and retail 11.6%.

The company reports a 0.89 correlation between AI Overview prevalence and organic traffic loss across those 12 industry observations. That is a strong association. It does not mean AI Overviews caused 89% of the loss, nor does it isolate AI Overviews from every other difference between sectors, websites and queries.

The referral side also differs. AI systems represented 11.8% of sessions for B2B SaaS, 9.4% for professional services and 8.1% for financial services, but only 2.3% for ecommerce and retail, 1.9% for real estate and 1.4% for local services. Media and publishing had the largest organic decline yet received only 4.8% of sessions from AI referrals.

AI referrals are therefore not simply replacing organic search. Some sectors appear to be losing far more conventional traffic than they recover through observable AI handoffs.

<h2>The AI referral mix is diversifying</h2>

ChatGPT still dominated First Page Sage’s AI referral traffic in 2026, with 68.4% of sessions. Gemini accounted for 14.7%, Perplexity 6.9%, Claude 5.3%, Copilot 3.1% and other systems 1.6%.

ChatGPT’s share fell from 84.2% in 2024 while its referral volume continued to grow. Gemini gained 10.6 percentage points over the same period, with smaller gains distributed across Claude, Copilot and other assistants.

These are referral shares within the First Page Sage panel, not global market share. Even so, the trend reinforces an important measurement principle: an AEO programme should not treat one answer engine as a proxy for the entire environment. Systems differ in audience, product design, retrieval, citations and propensity to send a click.

<h2>The website may be becoming a decision destination</h2>

For decades, websites have educated, explained, compared, persuaded and converted. AI can increasingly mediate the first four activities before a consumer arrives. The website may therefore become more concentrated around verification, pricing, availability, booking, signup, service and transaction.

The Brainlabs analysis introduced that hypothesis. First Page Sage provides a second observational dataset pointing in the same direction and adds a particularly useful clue: informational traffic is being affected much more heavily than transactional traffic.

That still does not prove the pathway for any individual person. It does make a traffic-only measurement model increasingly inadequate.

The measurement chain AEO practitioners need is broader:

<strong>Observed demand → Visibility → Brand representation → Consideration → Recommendation → Handoff → Business outcome</strong>

Traffic is one transition within that chain. It is not the whole system.

<h2>AEO Updates Takeaway</h2>

The defensible conclusion is not that AI caused organic traffic to fall 23.6%, that AI referrals inherently create better prospects, or that lost clicks no longer matter. First Page Sage’s panel still implies a substantial net loss in lead volume when AI Overviews appear, even after stronger post-click behaviour is considered.

The more important conclusion is that search purpose changes the meaning of the observed traffic. AI appears most capable of absorbing informational journeys, while visits tied more closely to action remain comparatively resilient. The people who still click may therefore represent a later and more selected stage of demand.

AEO measurement must move beyond asking how much traffic AI sent. It needs to show whether a brand appeared, how it was represented, whether it survived comparison, whether it was recommended, when the user was handed to a first-party experience and what happened next.

The website is not necessarily disappearing. It may be receiving a different part of the journey.

<h3>References</h3>

1. <a href="https://firstpagesage.com/seo-blog/ai-impact-on-website-traffic-report/">First Page Sage, “AI Impact on Website Traffic: 2026 Report,” August 25, 2026</a>

2. <a href="https://aeoupdates.com/articles/ai-search-website-downstream-brainlabs">AEO Updates, “AI search may be moving the website down the funnel,” August 25, 2026</a>

3. <a href="https://firstpagesage.com/seo-blog/the-most-cited-websites-by-ai-models-for-buying-intent-queries/">First Page Sage, “The Most Cited Websites by AI Models for Buying-Intent Queries,” July 2, 2026</a>

4. <a href="https://digiday.com/media/in-graphic-detail-ai-visibility-is-no-longer-about-referral-traffic/">Digiday, “AI visibility is no longer about referral traffic,” July 21, 2026</a>