---
title: "AI Referral Traffic in WordPress: Count Visits From Assistants"
url: https://hostmy.blog/ai-referral-traffic-wordpress/
date: 2026-11-06
modified: 2026-11-06
lang: en
author: "Aditya Sharma"
description: "Three ways to count visits from AI assistants on WordPress, a seven day test of what each assistant really sends, and how to read a number that will be low."
categories:
  - "RankReady"
image: https://hostmy.blog/wp-content/uploads/2026/10/hmb-card-2796-v3-1024x538.jpg
word_count: 1547
---

# AI Referral Traffic in WordPress: Count Visits From Assistants

AI referral traffic is the set of human visits that arrive on your site after someone clicks a link in an AI assistant's answer. It is small, easy to miss and worth counting properly. Most analytics setups have no bucket for it, some of it arrives with no referrer at all, and some of it hides inside other numbers.

This post is a measurement guide, not a vendor summary. It shows three ways to count assistant visits, a seven day test you can run on your own site to find out what each assistant actually sends, and how to read a number that will almost always be low. Where a vendor documents something, the post says so. Where we could not find it documented, the post says that too and gives you a way to test it yourself.

## What counts as AI referral traffic

A referral is a visit that arrives because someone clicked a link on another site. For AI, a reader clicks a link in an assistant's answer and lands on your page, so the assistant is the referrer. The hosts to look for are chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, claude.ai and copilot.microsoft.com. Hosts change, so treat that list as a starting point and add to it when your own data shows a new one.

That is a different event from a crawler fetching your page. A crawler fetch is a machine reading you. A referral is a person arriving. Conflating the two is the commonest mistake in this area, and [OAI-SearchBot: the crawler behind ChatGPT search](/oai-searchbot-wordpress/) explains why a visit from a crawler is not a citation and not a reader.

## What OpenAI and Google document about referral traffic

[OpenAI's Publishers and Developers FAQ, as archived in June 2026](https://web.archive.org/web/20260609031847/https://help.openai.com/en/articles/12627856-publishers-and-developers-faq), said that ChatGPT automatically included the UTM parameter utm_source=chatgpt.com in referral URLs from ChatGPT search results, and that publishers who allowed OAI-SearchBot could track that referral traffic in analytics platforms such as Google Analytics. That statement covers ChatGPT search results only. We could not find the same thing documented for other assistants, so treat their links as untested and use the seven day test below to find out what each one sends. OpenAI's [crawler documentation](https://developers.openai.com/api/docs/bots) separately describes OAI-SearchBot and how to allow or block it.

Google is a different case. Its guide to [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features) says that sites appearing in AI features such as AI Overviews and AI Mode are included in overall search traffic, reported in the Performance report in Search Console under the Web search type. In practice that means Google's AI traffic is reported as Google search, so look for it in Search Console and not in a referrer list. [How to rank in AI Overviews: checks for a WordPress blog](/rank-in-ai-overviews-wordpress/) covers that side.

## Method one: your analytics

Most analytics tools record the source of each visit. In Google Analytics, the acquisition reports show sessions by source and medium, and a filter that matches several names at once saves time. The pattern chatgpt|perplexity|gemini|claude|copilot is a reasonable start for a regular expression filter on the source. Exact menu names change, so follow Google's current help for the steps.

The strengths are that it is already there and that it shows behaviour after the click. The weaknesses are that it depends on a script that a consent banner or an ad blocker can stop, and that it only sees what the browser reports.

## Method two: your access log

Your server records the referrer of every request, and nothing a visitor installs can hide it. In the common combined log format, the referrer is the fourth field when the line is split on double quotes. This lists the referrers that look like assistants, with counts.

`awk -F'"' '{print $4}' /var/log/nginx/access.log | grep -iE 'chatgpt.com|perplexity.ai|gemini.google.com|claude.ai|copilot.microsoft.com' | sort | uniq -c | sort -rn`

The log path varies, as [reading your server log by hand](/read-server-log-ai-crawlers/) explains. This second command counts requests whose address carries the utm_source=chatgpt.com parameter that OpenAI documented for ChatGPT search results, as archived in June 2026. A result of zero is also a finding: it tells you that no visit on your site carried it, and the parameter is not something to rely on for every link.

`grep -c 'utm_source=chatgpt.com' /var/log/nginx/access.log`

The log counts requests, not people, and it includes bots that send a made-up referrer. Treat the result as an upper bound for what a person did, and filter by status code and user agent if you want to tighten it.

## Run a seven day test on your own site

Documentation tells you what a vendor intends. A test tells you what reaches your server. This one takes a few minutes a day for a week and needs no tools beyond the log command above.

`1. Pick one page that answers a clear question.
2. Ask each assistant you use that question, in your own words.
3. If the answer links to your page, click the link once.
4. Write down the time, the assistant and the device.
5. Next day, find that time in your access log.
6. Record the referrer host and whether the address carried a parameter.
7. Check whether your analytics showed the visit.
8. Repeat from a phone app and from a desktop browser.`

Fill in a small table as you go. The columns that matter are the assistant, the device, the referrer host the log recorded, whether a parameter was present, and whether analytics counted the visit. Where the log shows a host and analytics shows nothing, a script was blocked. Where the log shows an empty referrer, the referrer was stripped on the way, and that visit will land in your direct traffic.

Referrer handling can differ between an app, an in-app browser and a desktop browser, which is the reason to repeat the test from each device your readers use. Delete your own test visits from the totals afterwards, or you will count yourself as a reader.

## Method three: a count made on the server

A plugin or a small piece of server code can read the referrer on the server before any script runs, and keep a count. Because it runs before the page reaches the browser, a blocked script or a declined consent banner does not stop it. What matters is the trade-off against the other two methods.

| Method | Sees | Misses |
| ------ | ---- | ------ |
| Analytics | Behaviour after the click | Visitors who block scripts or decline consent |
| Access log | Every request with its referrer | It counts bots and cannot tell people apart |
| Server-side count | Referrals before any script | Anything with no referrer, and history before install |

No method is complete. Use two and compare them against the results of your own seven day test.

## Why the number will be low, and how to read it

Several things push the count down. Many assistants show an answer and a link that few readers follow. Some apps and browsers strip the referrer, so a visit arrives as direct traffic. A reader may copy the address and paste it. And for Google, the traffic is reported as search. The count you see is a floor.

That is why [what zero click search actually costs a blog](/zero-click-search-cost/) matters here. A reader who got the answer in the assistant never visited, and no counting method will show them.

Do not read the absolute number. Read the direction and the ratio. If crawlers fetch your pages and no referrals ever arrive, you are being read and not clicked, which is not the same message as being ignored. If both rise together, your pages are being used and followed. If referrals appear for a few pages only, look at what those pages have in common: the question they answer, how early they answer it and how they are titled.

[The Cited: where your readers really come from](/where-your-readers-really-come-from/) and [AEO tools: what to measure, fix and skip](/aeo-tools-wordpress/) set out how this fits with the other measurements.

## Mistakes to avoid, and a monthly routine

`1. Counting a crawler fetch as a visit.
2. Reading a month of single digits as failure.
3. Comparing a week with a month.
4. Forgetting that referrers are stripped.
5. Counting your own testing as traffic.
6. Expecting the number to explain why you were cited.`

Once a month, take the log count and your analytics count for the same dates, note both, and put them beside the number of AI crawler fetches. Keep the three in one place. After a few months, the pattern is the useful result. [The 15 second test: ask an AI about your own blog](/fifteen-second-brand-test/) gives a qualitative check to sit beside the numbers.

## Where RankReady fits

[RankReady's referral tracking](/plugins/rankready/ai-referral-tracking-wordpress/) is the server-side count from method three. Its page documents what it recognises, what it skips and what it stores, so this post does not repeat that. RankReady's documentation says plainly that single digits in the first month are normal, because AI answers are mostly read without a click.

It is free and runs alongside your analytics, as a companion and not a replacement. It does not tell you why a visit happened, and it does not promise any volume. The free [AI Readiness Checker](/plugins/rankready/ai-readiness-checker/) shows whether assistants can reach your pages in the first place.

If you ran the first command on your own log tonight, would you know what the answer should be?