Is Anyone Arriving From AI Chatbots? How to Measure It

More people now ask an assistant a question and click a link in the answer instead of scanning a page of search results. If your content gets cited, some of those clicks land on your site. The obvious question is how many.

Answering it is harder than it sounds. Some of those visits arrive with clear markers, some arrive looking like direct traffic, and a large amount of "AI traffic" in server logs is not human at all. This post explains what you can and can't measure, and how to set up a sensible view.

Two different things get called AI traffic

Keep these apart, because mixing them wrecks the analysis.

Referral visits from people. A person reads an answer in a chat product, clicks a citation, and arrives at your page. Their browser may send a referrer showing the assistant's domain, or the link may carry a tracking parameter that says where it came from. This is the traffic you care about for audience and conversions.

Crawler and agent requests. AI companies run bots that fetch pages to train models or to retrieve content live when a user asks something. These requests show up in server logs and, if your analytics script ran, sometimes in reports. They are not visitors. They don't convert and they distort pageview counts.

OnyxMetric deals with both. Visits get classified into channels, and there is an AI channel for sessions that come from known chat assistants. Separately, the bots area includes an AI overview, which exists to compare crawler activity with actual referral visits.

How a visit gets labeled

OnyxMetric assigns each session a channel such as Organic Search, Direct, Referral, Social, Email, and AI. For AI, it looks at where the visit came from. That includes the referring domain of known chat assistants and, when a link carries them, source and campaign tags that name an assistant. The list covers the major chat products, and it's maintained so that new ones can be added as they become common.

You don't need to write regular expressions or hand-build a referrer group. You open the channel breakdown and AI appears as its own row when there is traffic from it.

Why the number is always an undercount

Be skeptical of any report that claims exact AI-referral totals, including ours. Several things hide visits:

• Missing referrers. Some chat apps, especially native mobile apps, open links without sending a referrer. Those visits fall into Direct.
• Copy and paste. A user who copies your URL out of the answer and pastes it into a new tab is a direct visit by any measure.
• No click at all. People read your content in the answer, trust it, and later search for your brand by name. That shows up as branded organic search, not as AI traffic.
• Link stripping. Some interfaces rewrite links through intermediate redirects, which can drop the original source.

So treat the AI channel as a floor. The real influence is bigger than the measured visits, but you can't see by how much.

A practical way to watch it

1. Look at the channel

Open the channel breakdown over a longer window, 30 or 90 days. AI traffic is typically small and uneven, so short windows are noisy. Look at the trend, not a single week.

2. Look at the landing pages

Filter to the AI channel and look at which pages people land on. This is the most useful view. Assistants cite pages that directly answer questions: how-to guides, comparisons, definitions, documentation. If your top AI landing pages are all one kind of page, you've learned what kind of content gets cited, and what to write more of.

3. Look at what they do

Compare AI-referred sessions with others on a few simple measures: pages per session, time on site, and goal completions. Visitors who clicked through from an answer have often already had their basic question addressed, so expect different behavior: sometimes shorter visits, sometimes higher intent. Don't assume. Set up a goal for your main conversion and compare directly.

4. Compare with crawler activity

In the bots overview, look at which AI crawlers are fetching your pages. If a crawler visits often and you see few referral visits, your content is being read but not cited. If crawler activity is near zero, check your robots rules and make sure you aren't blocking access by accident.

Tag links you control

You can't tag a citation the assistant generates. You can tag links you place yourself, in places where AI products may later read them, such as profiles, public docs, or press pages. Use clear source parameters so those visits are attributable. And when your own team shares a link inside a chat tool, tag that too, or you'll file internal clicks under the wrong label.

Keep crawlers out of the visitor numbers

If automated requests are inflating your counts, turn on bot filtering and review what is being blocked. The goal is a visitor number that reflects people. Counting crawls as visits makes every ratio, bounce rate, conversion rate, pages per session, worse in ways that are hard to diagnose.

What to do with the result

Measurement is only useful if it changes something. Three decisions it can inform:

1. Content. Write more of what gets cited. Make answers explicit near the top of the page, use clear headings, and keep facts current.
2. Landing experience. Visitors from answers arrive mid-thought. Make sure the page they land on explains what it is and offers a next step without requiring them to dig.
3. Access. Decide deliberately what you let AI crawlers fetch. That's a business choice. Block or allow with intent, and check afterward that the logs match your policy.

Expectations

For most sites today, AI referrals are a small share of traffic. Some sites see a few percent, a few see more, particularly developer tools and reference content. The share is growing but unevenly. Check monthly, keep the channel visible next to organic search, and avoid rebuilding your strategy around a number you know is incomplete.