How to check whether ChatGPT is sending visitors to your website

01The two signals that identify an AI visit

When someone clicks a link inside an AI assistant, your analytics can recognise the visit in two ways.

The referrer. The browser often tells your site which page the visitor came from. For ChatGPT that is usually chatgpt.com, and older links may still show chat.openai.com.

The utm_source tag. Links that ChatGPT search shows as sources are generally tagged with utm_source=chatgpt.com, so the visit announces itself in the address:

The other assistants rely mostly on the referrer. These are the domains you are likely to see:

These behaviours are set by each company and they change, so treat the table as a starting point and check your own reports for what actually arrives.

02Why these visits are easy to miss

In a default analytics setup, AI visits rarely get a row of their own. They tend to end up in one of three places:

• Buried in the referral list, one line among dozens of small domains.
• Counted as direct traffic, when the browser or app sends no referrer, or when someone copies the link out of the chat and pastes it into a new tab.
• Left unclassified, in tools that cannot read a source tag when no medium comes with it.

This is why the figure you find is a floor. If an assistant names your business and the reader then types your address by hand, that visit looks identical to any other direct visit.

03How to check, step by step

The steps are the same in almost any analytics tool.

1. Open your sources or referrers report for the last 30 days.
2. Search it for the names above: chatgpt, perplexity, gemini, claude, copilot. Also check the campaign or UTM view for utm_source=chatgpt.com.
3. Add the totals together. That is your AI visit count for the period.
4. Open the landing pages for those visits. The pages assistants link to are often your guides and comparison pages, not your home page.
5. Note the date and repeat in a month. One reading tells you little. A trend tells you whether anything is working.

In Seonavo, you can skip the search. It reads both the referrer and the utm_source value and groups matching visits into an AI channel, split by assistant, for example AI: ChatGPT and AI: Gemini. The summary on the overview page also states the total in a sentence. Moving between your own pages is not counted as a source.

04Crawlers are not visitors

Assistants also send automated requests to your pages to read them, using names such as GPTBot, ChatGPT-User, ClaudeBot and PerplexityBot. You will see them in server logs. They are software fetching text, not people reading it, so they should not be counted as visits.

Seonavo filters these out before they reach your reports. That keeps the AI channel limited to people who clicked a link and arrived. If you ever compare it with raw server logs, expect the log to be much larger for this reason.

05How to read what you find

A small AI count is normal for most sites. What matters is what those visitors do once they arrive.

• Read depth and engaged time. Did they scroll through the page or leave after a few seconds? Compare them with your search visitors rather than assuming they are better or worse.
• The pages they land on. If one article keeps being cited, keep it accurate and up to date, and link from it to the next step you want people to take.
• Goals. Tag the button or form that matters and see whether AI visits complete it. Judge on your own numbers, not on general claims about AI traffic.

If you changed something recently, such as a title, a description, structured data or a batch of edits from an optimisation tool, compare the weeks before and after. Change one thing at a time and wait at least a few weeks, otherwise you cannot tell what caused a shift.

06What the numbers cannot tell you

• Mentions without links. An assistant can recommend you and send nobody. Only a visit leaves a trace in your analytics.
• Visits that lose their source. Copied links, apps and privacy settings can strip the referrer, which pushes real AI visits into the direct total.
• Visibility inside answers. Tools that test prompts and report whether you are cited measure something different from visits. Use each for its own question.

A simple way to fill part of the gap is to ask new customers or subscribers how they found you.

07Common questions