Is ChatGPT actually sending traffic to hotel websites?

Yes, in small volumes, and the volume differs so much between properties that the only figure worth anything is your own. Google Analytics has grouped these visits under an AI Assistant channel since May 13, 2026, so a ninety day read takes about two minutes. If your site is newer than that setting, or you want a longer history, a custom channel group can be applied retroactively over the data you already hold.

Why do assistant visits show up as direct traffic?

They land there because nothing told your website where the visitor had come from. Google's Direct channel is defined as a session with no source and no medium, which is where a pasted link, a typed address or an app that strips the referrer all end up. None of that is a fault in your analytics or something you can repair afterwards, so assume the real figure is higher than the one on screen.

How do I add Copilot and Perplexity to that channel?

Build a custom channel group in the admin area, make a channel that matches those hostnames, and move it above Referral in the ordering. Google's help page for custom channel groups gives a worked AI assistant example that names Microsoft Copilot and Perplexity alongside ChatGPT, Gemini and Claude, and custom groups apply to your existing reports rather than only to new data.

How do I tell a crawler from a guest in my server log?

Two things on each line settle it. One is the address the visitor came from, which shows somebody was sent to you. The other is the name the software gives itself, which shows a program made the request instead of a person. OpenAI, Perplexity, Anthropic and Google all publish the names their crawlers use and the addresses they run from, so a line claiming to be a vendor's crawler from an address outside that list is somebody else borrowing the name. Google Analytics already filters known bots for you and gives you no way to see what it removed.

Does a citation count as a referral?

No, and keeping them apart is most of the discipline here. Bing Webmaster Tools counts how often Copilot and the AI answers in Bing used your pages as sources. Your analytics counts people who arrived. Expect the citation count to run well ahead of the visit count, which is what this channel normally looks like.

Can I track a booking that happens without a browser?

Not in your website analytics, because it never touched your website. When an assistant completes a reservation it does so on a partner's system, so that booking reaches you as a platform reservation in your channel manager with the platform's commission on it. What your analytics can show is the other route, where an assistant sends a person to your own site to finish, which is the one worth protecting.

Do I need a visibility tool for this?

Not for measuring referrals. Google Analytics, your server log and Bing Webmaster Tools cover the whole check and none of them charges you. A visibility tool runs a fixed list of questions past a model each month and reports how often you were named. The score counts the quoting. It says nothing about how many people arrived.

Getting found9 minute read

Tracking AI assistant referrals to your own website

Christian, Co-founder, Fullstreak; co-owner, MINT @Naschmarkt, Vienna

Published 20 Sept 2026

At a glance

Who this is for: Operators who want to be named when a guest asks an assistant, or a search engine, where to stay.
What you get: Google Analytics has put assistant visits in a channel of their own since May 2026, and your server log has kept the same record for years. Here is the ninety day check, the referrers and crawler names to look for, and the four ways the number gets misread.
In this piece: 7 sections, 9 minute read
Published: 20 September 2026
Filed under: Getting found

On this page7 sections

The ninety day read in Google Analytics
Add the assistants Google has not included
The same check in your server log
Telling a guest from a crawler
Four ways this number gets read wrongly
What our own record shows
What to do with the number once you have it

Google Analytics has sorted these visits into a channel of their own since May 13, 2026. Open Reports, then Acquisition, then Traffic acquisition, set the range to the last ninety days, and look for the row marked AI Assistant. If you do not run analytics at all, the web server keeps the same record: every visit arrives carrying the name of the site that sent it. The totals will be small, and much of what the assistants do for you never appears as a visit at all.

Somebody asks ChatGPT for a quiet place near the lake for a long weekend in October, gets four suggestions with links under them, taps yours and lands on your rooms page, and that tap is the referral. The whole check takes a morning, and none of the tools it uses costs anything.

The ninety day read in Google Analytics

Google stamps a visit as coming from an assistant when the site that sent it appears on a list Google keeps. The announcement on May 13, 2026 named ChatGPT, Gemini and Claude. Google has never published the whole list, so read that row as a floor rather than the total.

Two changes to the report are worth making before you read it. Use the plus sign beside the channel column to add Landing page, and the report names the page that answered the question. At most properties that turns out to be a deep page somewhere rather than the home page. Then change the drop down above the table from channel to Session source, which is Google's name for the site the visit came from. The assistants appear there under their own web addresses, along with anything Google's list has missed.

Add the assistants Google has not included

Copilot and Perplexity were not in the May announcement. If either is landing in your Referral row, build a custom channel group and put an AI Assistants channel above Referral in the order. Google's instructions for custom channel groups include a worked AI assistant example naming ChatGPT, Gemini, Microsoft Copilot, Claude and Perplexity, and the useful part is that a custom channel group can be applied to your reports retroactively. Apply one and Google re-reads the year of data you already hold, instead of starting a fresh count today.

The same check in your server log

Your server has been writing this down since the day the site went up, with no tag, no consent banner and no account to open. Every page it serves gets a line, and one field on that line holds the address the visitor came from. The documentation for the web server software most hosting runs describes it as the site the visitor reports having been referred from.

Ask whoever looks after your website for two counts from the last ninety days of the access log. First, how many visits arrived from each of the addresses in the table below, which is the assistants sending you people. Second, how many requests came from each of the crawler names in the same table, which is the assistants coming to read you. One of those counts answers whether the channel brings guests. The other is more basic. If no crawler has been near you in ninety days, your own hosting is probably turning the assistants away and nobody has noticed. If you have no web person, most hosting control panels keep the log under a heading like Raw Access Logs or Visitors, and some hosted website builders do not hand it over at all, in which case the analytics half of this check is all you have.

Assistant	The address it appears as in a log	Its crawlers, from the vendor's own documentation
ChatGPT	chatgpt.com	OAI-SearchBot for search, ChatGPT-User for pages a person asked it to open, GPTBot for training
Gemini and Google	gemini.google.com. Ordinary Google answers arrive as normal Google search traffic, not as a referral	Googlebot crawls; Google-Extended is a separate control for Gemini training
Copilot	copilot.microsoft.com, bing.com	Crawled through Bing, with citations reported separately in Bing Webmaster Tools
Perplexity	perplexity.ai	PerplexityBot for search, Perplexity-User when somebody asks it to open a page
Claude	claude.ai	ClaudeBot, Claude-SearchBot, Claude-User

Checked on September 2026 against the documentation each company publishes.

Telling a guest from a crawler

If you count crawler requests as guests, the channel will look far busier than it is, and you will make decisions about a number with no people in it. Analytics mostly protects you here. Google says traffic from known bots and spiders is automatically excluded using its own research plus the International Spiders and Bots List, and that you can neither switch it off nor see how much was removed. Your server log has no such filter, which is exactly why the two counts above are asked for separately.

When a line looks suspicious, the address it came from settles it. Four companies publish the ranges their crawlers use, and your web person can check a log entry against them in a minute.

OpenAI: openai.com/searchbot.json, openai.com/chatgpt-user.json, openai.com/gptbot.json
Perplexity: perplexity.com/perplexitybot.json and perplexity.com/perplexity-user.json
Anthropic publishes a list for its three crawlers, which its own help page links
Google publishes ranges for its common crawlers, described in its crawler documentation

Anything claiming to be one of those crawlers from an address outside the list is somebody else wearing the name.

Four ways this number gets read wrongly

A citation and a visit are counted in different places. Microsoft's AI Performance report in Bing Webmaster Tools counts how often your pages were used as sources across Copilot and the AI answers in Bing, and those counts are citations. Most people read the answer and never open the link underneath it. Record the two separately, since only one of them counts people.

Some assistant visits get filed as direct traffic. Direct is where Google files a visit that turned up with no source attached to it. Anything that arrives without telling your site where it came from goes there: a link copied out of an answer and pasted into a browser, a name the guest typed in after reading it, an app that passes nothing along. Some of your assistant traffic is sitting in that row and there is no way to pull it back out after the fact.

Your booking engine can eat the credit. If the booking step lives on a different web address than the rest of the site, Google Analytics treats the two as separate users and separate sessions unless cross-domain measurement is switched on. A guest an assistant sent to your rooms page turns into a second, anonymous visitor at the moment they start booking, and the reservation gets filed under the booking engine. To find out whether this applies to you, click Book Now on your own site and read the web address that comes up. If the part before the first slash changes, the booking step is on a different domain, and cross-domain measurement needs switching on. It lives in Admin, under Data streams: open your web stream, choose Configure tag settings, then Configure your domains, and list both addresses there. Until that is done, no attribution report on the site means much.

One month is too small a sample to read. The report counts sessions rather than people, and at a property of eight to twenty-five units the monthly figure swings on a handful of visits. Compare one ninety day block against the next.

What our own record shows

I run MINT @Naschmarkt, a group of serviced apartments in Vienna, with my co-founder. Since 21 July the site has carried a piece of tracking code we built ourselves. It records the source of a visitor's very first arrival and carries that source with them on every visit afterwards.

In the first four weeks of that record, 51 people gave us an email address through a form. Of those, four had reached us first through an assistant, with ChatGPT accounting for three and Claude for one. Because the code stays with a contact right through to a completed reservation, we can see how each of those four ended up.

None of the four got in touch on that first visit. They came back days later, in sessions that any analytics report files as direct traffic, and only then did they fill in the form. In a report that credits the last click, all four of those enquiries sit under Direct.

Google Analytics has a free version of that idea, and most owners never open it. The Traffic acquisition report you started with is built on where each session came from. The User acquisition report is built on where each person came from the very first time, under a dimension Google calls First user source. Read both. When the two disagree about the assistants, the gap you are looking at is people who came back later.

What to do with the number once you have it

If an assistant is sending people to one page and ignoring twenty others, that page is doing something the others are missing, and what to put on a page so an assistant can quote it is the recipe for repeating it. If nothing is landing anywhere, check whether the assistants can reach you at all, starting with where each one gets its answer.

Then set the ninety day number beside your other channels and rank it honestly before you decide how much of next quarter it deserves. At most small properties it sits well behind the Google Business Profile and the guests who already stayed with you, which is roughly the order those jobs are worth doing in.

At under roughly $350,000 a year of bookings, keep this in house. Every part of the check is free and the whole thing fits in a morning. We do not take on paid work below that line, because the commission a direct channel saves at that size will not cover a first year of building it, and who this is built for sets out the arithmetic. Above it, if you want somebody to read your ninety days with you, book a call.

Questions operators ask

Is ChatGPT actually sending traffic to hotel websites?: Yes, in small volumes, and the volume differs so much between properties that the only figure worth anything is your own. Google Analytics has grouped these visits under an AI Assistant channel since May 13, 2026, so a ninety day read takes about two minutes. If your site is newer than that setting, or you want a longer history, a custom channel group can be applied retroactively over the data you already hold.
Why do assistant visits show up as direct traffic?: They land there because nothing told your website where the visitor had come from. Google's Direct channel is defined as a session with no source and no medium, which is where a pasted link, a typed address or an app that strips the referrer all end up. None of that is a fault in your analytics or something you can repair afterwards, so assume the real figure is higher than the one on screen.
How do I add Copilot and Perplexity to that channel?: Build a custom channel group in the admin area, make a channel that matches those hostnames, and move it above Referral in the ordering. Google's help page for custom channel groups gives a worked AI assistant example that names Microsoft Copilot and Perplexity alongside ChatGPT, Gemini and Claude, and custom groups apply to your existing reports rather than only to new data.
How do I tell a crawler from a guest in my server log?: Two things on each line settle it. One is the address the visitor came from, which shows somebody was sent to you. The other is the name the software gives itself, which shows a program made the request instead of a person. OpenAI, Perplexity, Anthropic and Google all publish the names their crawlers use and the addresses they run from, so a line claiming to be a vendor's crawler from an address outside that list is somebody else borrowing the name. Google Analytics already filters known bots for you and gives you no way to see what it removed.
Does a citation count as a referral?: No, and keeping them apart is most of the discipline here. Bing Webmaster Tools counts how often Copilot and the AI answers in Bing used your pages as sources. Your analytics counts people who arrived. Expect the citation count to run well ahead of the visit count, which is what this channel normally looks like.
Can I track a booking that happens without a browser?: Not in your website analytics, because it never touched your website. When an assistant completes a reservation it does so on a partner's system, so that booking reaches you as a platform reservation in your channel manager with the platform's commission on it. What your analytics can show is the other route, where an assistant sends a person to your own site to finish, which is the one worth protecting.
Do I need a visibility tool for this?: Not for measuring referrals. Google Analytics, your server log and Bing Webmaster Tools cover the whole check and none of them charges you. A visibility tool runs a fixed list of questions past a model each month and reports how often you were named. The score counts the quoting. It says nothing about how many people arrived.

Filed under

Who wrote this

Christian

Co-founder, Fullstreak; co-owner, MINT @Naschmarkt, Vienna

I own MINT @Naschmarkt in Vienna with Maxine, and I run Ekamoira, the agency Fullstreak came out of. Almost every booking we took arrived through the platforms, so we built the site that takes the booking, the plumbing that keeps one calendar, and the pages Google and the AI assistants quote. MINT ran it first. Fullstreak is that work, made repeatable for operators like you.

MINT @Naschmarkt, Vienna

Every vendor claim traces to the vendor's own page, checked on the month shown under each table. MINT numbers come from our own books.

See where your own bookings leak.

Twenty minutes on a call. You get a read on where your bookings start today, what the platforms take, and what a direct channel would take to build. It is yours whether or not you hire us.

Count what the platforms take