# Reading the AI Traffic page

> How to read Kymo's AI Traffic page — the people AI sent you, which assistant sent them, their journeys, the pages that earn them, and the weekly report.

Canonical: https://kymo.in/docs/guides/reading-ai-report
Last updated: 2026-10-19

The AI Traffic page answers three questions in order: is AI reading your pages, is it sending you anyone, and what do those people do next. Read it top to bottom.

## The three numbers at the top

- **AI read your pages** — how often an AI assistant fetched one of your pages to answer a question someone just asked. Only answer crawls count here. Indexing and training crawls are left out, for the reason in [What the AI Traffic page counts](https://kymo.in/docs/concepts/ai-visibility).
- **AI sent you** — the people who arrived on your site from an AI answer, found by the referrer or the `utm_source` of their visit. The line under it says how many of them Kymo also matched to a read it logged.
- **Then converted** — the people whose first visit ever came from an AI answer and who completed one of your goals. The line under it gives the number of goal completions and the goals behind them.

All three count people once per day, because the visitor ID rotates at midnight UTC. Someone who comes back on three days counts three times in a 7-day view.

> **Note:** The reads need the server-side snippet. The people and conversions need only the script tag. If you install only the script tag, the page still shows who AI sent you, and the first number stays empty until the snippet is in place. See [Install](https://kymo.in/docs/start/install).

## Same page, two periods

Under the three numbers, one line compares two groups of pages: the pages AI read in the period before, and the pages it did not read. For each group it gives the share that got at least one AI visitor in this period, and how many times larger one share is than the other.

The reads come before the visits, so the visits did not cause them. The line still shows a link between reads and visits, not proof that one read sent one person: popular pages get more of every kind of visit. It shows the result whichever way it goes, and it stays hidden when either group has fewer than 10 pages.

The line compares two periods because AI visitors often arrive days after the read. An assistant reads a page, stores what it learned, and quotes it in answers for days or weeks.

## Where they came from

One row per assistant: ChatGPT, Gemini, Perplexity, Claude, Copilot and others. Each row shows the people that assistant sent, the conversions from people it brought in first, and a small line of its traffic over the period. This is the fastest way to see which assistant actually matters to you, and the answer is often not the one you would guess.

## Journeys

One card per person, from the AI answer to what they did next. The first step is the page the assistant sent them to. The steps after it are the pages they viewed that same day, in order. A card that ends in a goal shows which goal, and how many minutes after the AI answer it happened.

When someone converts on a later day, the card starts at that day's first page and says which assistant first sent them, and how many days earlier.

> **Note:** A journey never joins one person across days. Kymo is cookieless and its visitor ID changes every day, so "first sent by Gemini, 4 days earlier" comes from a record the browser keeps of where it first came from, not from following the person. See [Privacy](https://kymo.in/docs/trust/privacy).

## Pages AI sends people to

One row per page, with five columns: how often AI read it, how many people AI sent to it, how many of them converted, and a verdict. The table has three views:

- **Sending you people** — the pages that brought someone in, ordered by conversions and then by people. This is what is working.
- **Read, no visitors** — the pages AI reads again and again that send you nobody. This is your list of pages to fix.

A page gets the "No one clicked through" verdict only when AI read it at least 3 times in the period before and no one arrived from an AI answer in this period. A page read only this period has not had time to send anyone yet, so it gets no such verdict.
- **All pages** — everything, biggest gap first.

Click a row to see the inferred prompt family, which AI engines read it, the trend against the previous period, the confidence of the inference, and the intent of the page.

"Converted" is keyed to the page AI first sent someone to, not the page they converted on. A page can show conversions in a week when AI did not read it at all: it brought the person in earlier, and the conversion is its result.

> **Note:** Inferred prompts are informed guesses, always phrased as "likely" and prefixed "e.g." Kymo does not receive the question anyone actually asked. It infers intent from the page type and the pattern of reads. Read them as signal, not transcript.

## The weekly report

A short written read of the period, built from your own numbers with no model call:

- A **lead** finding at the top.
- **Actions** — findings that imply something to do, such as a page AI reads a lot that sends nobody, or a page an assistant asked for that returned a 404.
- **Notes** — findings worth knowing that need no action.

The "Copy for AI" button renders the same report as markdown, with the table and the people per assistant, so you can paste it into an assistant.

## Came back another way, and reads over time

At the bottom, two pieces of context.

**Came back another way** counts visits from people AI found for you who returned on a later day through another route, such as typing your address or searching your name. It counts visits, not people. Kymo stores no cookie, fingerprint or ID that follows a person. The browser keeps only a note of where it first came from, so Kymo can tell a visit came from someone AI brought in, but not whether two visits were the same person.

**AI reads over time** plots answer, indexing and training reads as three lines. It is the one place on the page that shows all three, as context for the numbers above.
