Open GA4, filter for AI referrals, and you’ll probably see a number that barely registers – half a percent, maybe less. It looks like ChatGPT, Perplexity and the rest are sending you almost nobody.
That number is wrong. Not slightly wrong – fundamentally wrong. The traffic is there. Your analytics just can’t see most of it.
Welcome to dark traffic: one of the biggest blind spots in modern web analytics, and the reason so many businesses are underselling the impact AI search is already having on their pipeline.
What is dark traffic?
Dark traffic is any visit to your website that arrives without a trackable source. Your analytics platform has nowhere to file it, so it defaults to one of two places: the “Direct” bucket, or nowhere at all.
Historically, dark traffic was a rounding error – a handful of app-based clicks, an email client that stripped a tracking parameter, the odd bookmark. Nothing to lose sleep over.
AI search changed that. When someone clicks a citation inside ChatGPT, Perplexity or an AI Overview, many platforms strip the referrer header before the visitor ever lands on your site. Your analytics sees someone appear from nowhere and logs them as Direct traffic – exactly as if they’d typed your URL into their browser from memory.
They didn’t. They came from an AI answer that recommended you. Your reporting just doesn’t know that.
Why your AI numbers can’t be trusted
Recent data puts the scale of this at 70.6% of AI traffic arriving with no trackable referrer at all. For every AI visit your analytics correctly identifies, more than two are hiding in Direct.
This isn’t a data-quality problem you fix with a setting change – it’s structural. Referrer stripping happens before the data ever reaches your platform.
When we asked 171 marketers how they track AI traffic, 54% said, unprompted, that they know AI is sending visitors – they just can’t find them in the data.
It’s also why every industry benchmark quotes a different number.
➡️ Ahrefs put average AI traffic at 0.17% of sessions.
➡️ SE Ranking’s much larger study landed at 0.32%.
➡️ Search Engine Land found 1.08% across more than 3 billion sessions.
None of these are wrong – they’re all measuring the same visible sliver, using different samples. The other 70%+ never shows up in anyone’s dataset, because it can’t.
If someone quotes you a single “typical” AI traffic percentage, treat it as context, not a target. Your own trend line is the only comparison that matters.
How AI creates dark traffic – and why it’s not the only cause
Referrer stripping is the main driver, but it isn’t the only mechanism at play.
AI does the selling – “organic” takes the credit
- AI recommends, then you Google. Someone asks an AI assistant for a recommendation, gets your brand name, and opens a new tab to search for you directly rather than clicking the citation. GA4 logs that as organic or branded search – the AI conversation that started it gets no credit at all.
- Cross-device journeys. Someone researches on ChatGPT on their phone, then visits your site later from a laptop. No session stitches the two together.
- In-app browsers. A meaningful share of AI platform traffic happens inside in-app browsers, which are notoriously inconsistent at passing referrer data through to analytics.
Put these together, and you get a channel that’s quietly influential and almost invisible in a standard reporting setup.
Why this matters more than the numbers suggest
If AI traffic were both small and low-value, ignoring it would be reasonable. It isn’t. One in three marketers we surveyed said AI-referred visitors convert better than other channels – smaller in volume, sharper in intent, because the platform has already done the comparison shopping before the click happens.
That’s the uncomfortable maths: the visitors most likely to become customers are the ones your reporting credits to nothing, or worse, to “Direct” – a channel most teams don’t optimise for because it looks like background noise.
It skews decisions upstream too. A channel that doesn’t show up in the data doesn’t get budget, doesn’t get content investment, and never gets credited when it works.
How to spot AI hiding in your dark traffic
You can’t eliminate dark traffic, but you can build a strong circumstantial case for how much of it is AI-driven. Three signals are worth watching:
- Homepage-excluded direct traffic: Strip out the homepage and look at Direct visits to interior pages – blog posts, service pages, specific product URLs. Nobody types a long blog URL from memory. Landing directly on a deep page with no referrer means something pointed them there.
- Branded search lift: Watch Search Console for spikes in branded impressions and clicks that don’t line up with a campaign, PR push or seasonal pattern.
- Timing correlation: If a spike in dark traffic to a specific page lines up with a spike in branded search in the same window – with nothing else to explain it – that’s a far stronger signal than either number alone.
None of these prove a session came from AI on their own. Tracked together as a trend over 60-90 days, they build a case that’s hard to dismiss.
AI traffic is likely more valuable than you think
Dark traffic isn’t a glitch – it’s the default state of most AI-referred visits under a standard analytics setup. Somewhere between half and three-quarters of your AI traffic is likely sitting in your Direct channel right now, uncredited and unoptimised.
Standard GA4 won’t fix this on its own. You need to combine AI referral data, branded search trends and homepage-excluded direct traffic to see the real picture – and that’s exactly what we’ve broken down, step by step, in our full guide to tracking AI traffic, including the free dashboard we built to do it automatically.
Download our AI traffic dashboard
Get more reliable insights into your AI data.
Download here
