LLM Traffic in 2025: Early Performance, Real Intent, Uneven Results

Large language models aren’t just a curiosity in the traffic mix anymore. For some companies, they’ve already overtaken branded and direct referrals. ChatGPT leads the way, responsible for 4 out of 5 of all LLM-driven sessions. Perplexity and Gemini follow at a distance, but collectively, these three platforms now drive the majority of AI-attributed discovery.

But traffic is only part of the picture.

Bounce rates remain high — 26.3% on average — with smaller companies seeing the least engagement. At the same time, those who stay often stay for longer: the average LLM session runs over 3 minutes, and mid-market companies in particular show signs of deep research behavior.

Leads from LLM traffic are rare but serious. Across all accounts, 86% of hand-raisers are high intent, often requesting demos or contact. The strongest outcomes appear in companies with 501–1000 employees, which deliver the highest conversion to pipeline and revenue.

This report of 118 accounts breaks down where LLMs are driving value today, where the gaps lie, and how early adopters are turning this discovery engine into a strategic advantage.

‍

LLMs are already a top referral source, led by ChatGPT

ChatGPT drives more referral traffic than many branded or direct sources. In some accounts, it sends the largest share of sessions overall.

Across the LLM landscape, traffic is concentrated in a few platforms:

• ChatGPT: 82% 
• Perplexity: 12.1%
• Gemini: 4.9%
• Others: less than 1%

Visibility inside ChatGPT’s and Perplexity’s outputs is now a meaningful growth lever. Optimizing for those two platforms covers nearly the entire LLM-driven traffic surface today.

This level of consolidation makes optimization more focused, but also more urgent. Visibility depends on whether your content is selected, summarized, or cited in the way these models prefer.

Here’s our take:

• Structured content — especially content with clear answers, citations, and semantic clarity — may be what earns placement in these models.
  
• Some companies are likely winning visibility through accidental alignment with what the model prioritizes. Others are beginning to engineer for it.
  
• LLM referral traffic today behaves more like product-led growth than traditional search. The context is invisible. The content has to do more, faster.
  

This is an early channel with uneven mechanics. But the volume is real, and it’s already outperforming legacy channels for the companies paying attention.

‍

High bounce rates reveal a deeper issue with LLM traffic

LLMs are generating visibility. But that visibility doesn’t always lead to engagement. In many cases, users land on a page and leave without interacting at all.

Across the accounts analyzed, bounce rates remain high:

• Average bounce rate: 26.3%
• Half of all accounts: above 23.6%
• Top-performing 10%: under 14%

Bounce patterns shift by company size:

• <200 employees: 27–28% bounce rate
• 5000+ employees: ~21% bounce rate

Smaller companies tend to attract more casual or early-stage visitors. Larger enterprises see stronger intent and lower drop-off.

This is where the LLM experience breaks down. Users are showing up, but they’re not finding what they expected — or not fast enough. That disconnect could be due to vague summaries, misleading citations, or content that simply doesn’t deliver on the promise of the AI output.

Unlike traditional referral paths, there’s no breadcrumb trail. No meta description. No SERP. The page has to deliver context and clarity instantly, without relying on the user to orient themselves.

Here’s our take:
Pages need to assume zero context. If someone arrives from ChatGPT or Perplexity, they likely skipped everything that normally leads up to the visit. The job of the page is to pick up that conversation midstream.

Pages that convert well tend to mirror the model’s structure — concise headers, answer-first copy, and summaries that frontload value. Traffic from LLMs behaves more like in-product discovery than like search. There’s no time to warm up. The landing page is the pitch.

‍

LLM visitors that stay, stay engaged

LLM-driven sessions often begin without context. Users land mid-conversation, mid-problem, or mid-search — depending on how the model framed your content. Despite that, session-level behavior shows signs of real engagement.

Across accounts, the average session length is 3.5 minutes. That’s a meaningful amount of time, especially for traffic that arrives without a structured path. These visitors aren’t being funneled in through paid campaigns or nurtured over email. They’re clicking on AI-generated summaries and deciding, quickly, whether to stick around.

When they do stay, they read. They scroll. They scan for answers. They’re actively evaluating their options. 

Even in accounts with high bounce rates, those who remain often engage for several minutes. That suggests the problem isn’t with the content itself but whether the page immediately reinforces what the user expected based on the model’s summary.

‍

Most hand-raisers from LLMs are high-intent—when they show up 

Across the board, LLM traffic doesn’t produce a high volume of hand-raisers. But the hand-raisers it does produce are overwhelmingly intentful. They’re demo requests, contact forms, and sales-facing actions — the kinds of conversions that usually signal clear buying interest.

On average, 86% of LLM-sourced hand-raisers are high-intent, based on how each account defines lead quality. 

Half of the accounts in this dataset reported that every hand-raiser attributed to LLMs was high-intent. 

At the same time, 25% of accounts didn’t see a single high-intent hand-raiser, reinforcing the uneven distribution across teams and segments.

That duality — high quality, low consistency — is one of the most important patterns in the data. LLMs are beginning to surface serious buyers, but only in certain environments.

The clearest signal of commercial value appears in companies with 201–1000 employees. Within that band, the 201–500 segment generates the most volume of hand-raisers, likely because it also drives the most LLM sessions overall. 

But when it comes to performance across the funnel, companies with 501–1000 employees stand out. In this group, roughly 17% of LLM sessions convert to hand-raisers, and of those, 5.6% progress into pipeline and 2.8% reach closed-won.

No other size segment comes close to that level of follow-through (falling between 0.48% to 0.8% in closed won opportunities. Larger companies (1001-5000 employees) show intent, but deal progression is slower and less predictable. 

Smaller companies generate some interest but often fall off before serious qualification. The mid-market — particularly in that 500 to 1,000 employee range — shows a complete LLM funnel in motion.

These visitors arrive ready to engage and convert at an unusually high rate. 

That readiness doesn’t seem to be linked to volume, either. Companies in this band aren’t seeing massive session counts. What they’re seeing is sharper alignment between what the model surfaces and what the visitor needs next.

Here’s our take:

• When LLM visibility connects with the right persona and use case, it outperforms traditional top-of-funnel sources on both intent and progression.
• The mid-market segment — specifically companies with 501–1000 employees — is where that alignment is most consistent. 
• Teams seeing only a trickle of hand-raisers shouldn’t write off the channel. In this data set, a single hand-raiser was often enough to justify the effort — especially when that lead moved quickly into active pipeline.
• LLMs are starting to function as lightweight intent generators. They're not flooding the top of the funnel, but they’re carving out a new middle where the right buyer can leap forward without needing to be nurtured.

‍

Conversion remains the weakest link in the LLM funnel

We're seeing stronger engagement, higher-quality hand-raisers, and a clearer sense of intent. But those signals rarely make it to the finish line. The conversion journey, from visit to pipeline to revenue, is where most companies lose momentum.

LLM visits to hand-raisers:

• Median conversion rate: 5.24%
• 25% of accounts have a conversion rate below 0.49%

Hand-raisers to qualified pipeline:

• Median conversion rate: 2.66%

Qualified pipeline to closed-won:

• Median conversion rate: 0%

The most striking breakdown happens at the final stage. Across the majority of accounts, only a small subset of LLM-sourced opportunities are closing. That does not mean zero deals have closed. But it does show how few companies have managed to turn this traffic into repeatable revenue.

This is not a funnel problem in the traditional sense. There is volume at the top. There is intent. There is time-on-page and even form fills. What is missing is sustained progression. The hand-raiser becomes a lead, but the lead does not become a deal.

There are a few likely reasons. Many sales teams still do not know when a lead originates from an LLM interaction. Without that context, it is easy to deprioritize the follow-up or handle the lead as if it came from search or paid. Lead scoring systems may not assign the right weight. Landing pages may offer a clear hook, but fail to guide the visitor into a meaningful next step.

The deeper problem is that LLM sessions operate outside the structure most teams are built to handle. They do not follow campaign flows. They do not match channel tags. They often arrive mid-journey with no metadata trail. And when that happens, the infrastructure around the lead breaks down.

Here’s our take:

• The drop in performance between hand-raiser and revenue is not because these leads lack intent. It is because teams are not yet treating them as a distinct buyer path.
• Routing, scoring, and sales engagement strategies still prioritize volume over source behavior. In this model, that approach misses the nuance.
• Fixing the conversion gap starts with visibility. Sales teams should know when a lead came from ChatGPT or Perplexity. They should know what content surfaced. And they should be trained to respond accordingly.
  
  

‍

LLMs are building a new path to pipeline

LLM traffic today is uneven, low-volume, and often misunderstood — but it’s already delivering real value in specific pockets. Mid-market companies, particularly those with 501–1000 employees, are seeing measurable conversion from sessions that began with a ChatGPT or Perplexity answer. 

While most accounts aren't yet closing LLM-sourced deals, the ones that do show a clear throughline: deeper engagement, stronger alignment between content and intent, and faster progression to pipeline. This is still an early-stage channel, but one that rewards clarity, structure, and speed. 

Teams that optimize for AI-surfaced discovery — not just search or paid — are laying the groundwork for a new kind of top-of-funnel: less visible, more volatile, and in the right conditions, far more efficient.