The AI channel often looks dead because of how it is recorded

A builder tells us the AI work produced nothing. We look at the analytics and the Direct bucket has grown steadily for two months. These are frequently the same finding, read two different ways.

Run the free check

Traffic from AI assistants often arrives with no referrer information, which sends it to the Direct bucket in analytics alongside bookmarks and typed URLs. Before concluding the channel is not working, build a channel group that captures the assistant domains as referrers, then watch Direct on the pages that would only be found through an answer, since a rise there on deep pages is the signature of untracked AI traffic.

Why the traffic disappears

When somebody clicks a link, the browser normally tells the destination where the click came from. Analytics uses that to file the visit under a source.

Traffic originating inside an AI assistant does not reliably carry that information. Depending on the assistant, the platform, whether it was a desktop app or a browser, and how the link was opened, the referrer may be absent.

Analytics has to file the visit somewhere, and a visit with no referrer goes to Direct, which is the bucket for people who typed the address or used a bookmark. So a genuine AI referral is recorded as though the person already knew your address, which is precisely the opposite of what happened.

The signature to look for

You can often detect the missing traffic without any new setup, by looking at which pages the Direct visits land on.

Direct traffic, properly understood, should be concentrated on your homepage and on pages people would plausibly bookmark. Nobody types the address of a deep page answering a specific question.

So if Direct visits are landing on interior pages, especially ones that answer a specific question, that is not people typing the URL. Something sent them, and the something did not identify itself.

Look at the trend as well as the shape. Direct rising on deep pages, over weeks, while nothing else changed, is the signature worth investigating.

Capturing what can be captured

A portion of the traffic does carry a referrer, and that portion is worth separating out so you have at least a floor measurement rather than nothing.

In your analytics, create a channel group or segment that matches the assistant domains as referral sources. The ones worth including are chatgpt.com, perplexity.ai, claude.ai, copilot.microsoft.com and gemini.google.com, and it is worth adding new ones as they appear rather than setting this up once and forgetting it.

This will undercount, and it is important to hold that in mind when reading the result. What it gives you is a real number that moves in the right direction when the channel grows, which is enough to tell whether the work is doing anything.

The method that removes the ambiguity

If you want certainty rather than inference, there is a straightforward approach available to any builder.

Publish something that exists only in one place, then watch for arrivals at it. A page answering a specific question, linked from nowhere, not in your navigation, not shared anywhere. If people arrive at that page, they were sent by something that found it, and you can be confident about what.

The other reliable signal is asking people. A single question on signup, kept to one line, asking where they heard about you. The response rate is imperfect and the answers are honest, and for a small product a handful of people saying they asked an assistant is worth more than a great deal of inference.

Server logs are the ground truth

Analytics runs in the visitor's browser, which is why it misses so much. Your server log records every request that reached you, whether or not any script ran, and it is the closest thing to an honest record you have.

The log is also the only place you can see the crawlers themselves rather than the humans they eventually send. Searching it for the agent names, OAI-SearchBot, PerplexityBot, Claude-SearchBot and the rest, tells you whether the systems are actually fetching your pages, which is a question analytics cannot answer at all.

This is a genuinely useful early signal, because crawler visits precede citations. Seeing OAI-SearchBot fetch a page you published last week means the mechanism is running, months before any visitor arrives from it. Seeing no AI crawler activity at all, on a site that has been live for a while, points straight back at robots.txt or at pages that cannot be read.

Whether you can reach the logs depends on your host. Some platforms expose them and some do not, so this is worth checking once rather than assuming either way.

Reading the numbers without fooling yourself

Two cautions, because this is an area where it is easy to see what you want to see.

First, do not attribute all Direct traffic to AI. Some of it is genuinely direct: you, your friends, people you told, and anybody who bookmarked the site. Look at the change and the landing pages rather than the total.

Second, the numbers at this stage will be small, and small numbers move around for reasons that have nothing to do with your work. A week where Direct doubled from four visits to eight is not a trend. Look at months, not weeks, and resist building a narrative on a handful of sessions.

The honest position for most new products is that the AI channel is measurable in direction but not yet in precise magnitude. That is a perfectly reasonable thing to say, and it is much better than a confident number that is wrong.

What to set up today

Half an hour, once, and then the measurement works from here on.

  • Create the AI channel group with the assistant domains listed above.
  • Add a report or saved view showing Direct traffic broken down by landing page, so the interior-page signature is visible without digging.
  • Exclude your own visits, otherwise every number on this page is contaminated by you checking the site.
  • Add the one-line question at signup asking how people found you.
  • Write down today's baseline figures with the date, because in two months you will want to compare and will not remember.

Part of a larger guide

This page is one part of The AI channel. The other parts:

Questions people ask

Why does my AI traffic not show up in analytics?
Because traffic originating inside an assistant often arrives with no referrer information, and a visit with no referrer is filed as Direct alongside bookmarks and typed addresses. The visit is recorded as though the person already knew your address, which is the opposite of what happened.
How can I tell if my Direct traffic is really AI traffic?
Look at which pages the Direct visits land on. Genuine direct traffic concentrates on the homepage and pages people would bookmark, because nobody types the address of a deep page answering a specific question. Direct visits landing on interior pages, rising over weeks, is the signature worth investigating.
Which domains should I add to an AI channel group?
chatgpt.com, perplexity.ai, claude.ai, copilot.microsoft.com and gemini.google.com, adding new ones as they appear rather than configuring it once and forgetting. This will undercount, because much of the traffic carries no referrer, but it gives you a real floor that moves when the channel grows.
Is there a way to be certain about the source?
Publish a page that exists in only one place, linked from nowhere and not in your navigation, then watch for arrivals. Anyone who reaches it was sent by something that found it. Asking people directly at signup, in one line, is the other reliable signal.
Should I attribute all Direct traffic to AI?
No. Some of it is genuinely direct: you, people you told, and anyone who bookmarked the site. Read the change and the landing pages rather than the total, and exclude your own visits first or every figure is contaminated by you checking the site.
How much should I read into small movements?
Very little. At this stage numbers move around for reasons unconnected to your work, so a week where Direct went from four visits to eight is not a trend. Look at months rather than weeks. Measurable in direction but not yet in precise magnitude is an honest and useful position.

Get a baseline before you measure change

The free Get Found Check gives you a dated starting point for how reachable and citable your site is, so the analytics you set up today has something to be compared against.

Run the free check Have Licheo do it for you