Our AI Traffic Looked Nothing Like the Benchmarks

If you have read anything about AI search this year, you have met the same handful of statistics. AI referrals are growing fast. ChatGPT drives the overwhelming majority of them. AI traffic sits at roughly 1% of the average site’s visitors.

That second one is shakier than it looks. Depending on who is measuring, over what window, and in which market, published figures for ChatGPT’s share of AI referrals run anywhere from about 60% to over 90%. The benchmarks do not fully agree with each other before you even get to your own site.

We kept repeating these numbers to clients. Then we pulled our own.

What we actually found

Across roughly three months of our own site data, from late April to late July 2026, we recorded 5,529 sessions. Of those, 21 came from AI assistants. That is 0.38% of traffic, a fair bit under the ~1% figure that gets quoted as typical.

The split is the part that surprised us:

• Gemini: 18 sessions
• ChatGPT: 1 session
• Perplexity: 1 session
• Copilot: 1 session

Gemini accounted for about 86% of our AI referrals. ChatGPT, the platform that dominates every published estimate of AI referral traffic, sent us exactly one person.

Now the caveat, because it matters more than the finding: 21 sessions is a tiny sample. We are not presenting this as a new benchmark, and you should be suspicious of anyone who would. One busy week could reorder that list entirely.

But that is rather the point. If a number this small can invert the headline statistic for our site, the headline statistic was never going to tell you much about yours.

Why yours will look different

A few reasons the averages fail at the individual-site level:

Your category decides your platform. People research different things in different tools. A B2B software question, a local service search, and a recipe hunt do not happen in the same place. Aggregate data blends all of it into one number that describes nobody.

Google is embedding Gemini everywhere. It sits in Search, Workspace, Android, and Chrome. For a business whose audience already lives inside Google’s products, Gemini exposure is close to automatic in a way ChatGPT is not. That is our best read on our own result.

Measurement is genuinely leaky. Some AI tools pass a clean referrer and some do not. Traffic that arrives from an assistant can land in your reports as direct, unattributed. Whatever your AI number says, the real one is probably higher. Treat it as a floor, not a count.

How to find it in your GA4

This takes about two minutes and most businesses have never done it.

1. Open Reports, then Acquisition, then Traffic acquisition.
2. Look at the Session default channel group column for the AI Assistant channel. Google added this grouping so AI referrals stop hiding inside general referral traffic.
3. To see which tools specifically, switch the dimension to Session source. You will get the actual hostnames: chatgpt.com, gemini.google.com, perplexity.ai, copilot.com.
4. Widen the date range to at least 90 days. At these volumes, a 28-day window is mostly noise.
5. Add Engaged sessions as a column. Volume alone will mislead you.

That last one is worth dwelling on. Of our 18 Gemini sessions, 9 were engaged, about half. Our single Perplexity and Copilot visitors both engaged. Small numbers, but the pattern is consistent with what others report: people arriving from an AI assistant have usually had their basic question answered already, so the ones who click through tend to be further along than a cold search visitor.

What to actually do about it

Here is where we will disagree with a lot of the advice being sold right now.

At 0.38% of traffic, AI referrals are not where our next client comes from, and pretending otherwise would be dishonest. If someone tells you to rebuild your content strategy around a channel sending you 21 visits a quarter, check whether they are selling an AI visibility product.

What is worth doing:

Start measuring it now. You cannot spot a trend you never tracked. The value of today’s number is not the number, it is having a baseline when you look again in six months.

Watch the direction, not the level. Ours moved from a single session in one 28-day window to twenty in the next. Still trivial in absolute terms. Not trivial as a slope.

Do not build a separate strategy for it. The work that earns AI citations is the work that earns good rankings: answer real questions directly, be specific, keep facts current, and structure pages so a machine can extract a clean answer. We went into this in more detail in our piece on why citations now beat rankings. There is no second content strategy to buy.

Check your own data before believing a statistic. Including the ones in this post.

The honest summary

AI traffic is real, it is growing, and for most small and mid-sized businesses it is still a rounding error. Both halves of that sentence are true, and most coverage picks one and drops the other.

The useful move is not to panic or to dismiss it. It is to open GA4, find your actual number, write it down, and look again next quarter. Ours was 0.38% and mostly Gemini, which is not what we expected and not what the benchmarks predicted. Yours will be its own thing.

If you would rather someone read your analytics properly and tell you what is actually happening rather than what is trending, that is a fair bit of what our SEO team does, and it is the starting point for any AI search work worth paying for.