Two places on the dashboard show the split. They answer different questions.
The traffic chart
The main chart carries a second series: AI crawler views, drawn on its own right-hand scale.
The main traffic chart over 30 days with both series visible — the human area on the left scale and the AI crawler series on the right scale — with the legend showing all three entries. Pick a range where the crawler series has visible shape rather than being flat. Light theme, desktop.
Two scales, on purpose. Crawler traffic and human traffic are usually orders of magnitude apart. On one scale, whichever is smaller flattens to the axis and you learn nothing about its shape. Two scales let you see both.
The consequence: the two series are not comparable by height. A crawler line sitting above the human area does not mean more crawlers than people — it means the crawler series is at the top of its scale. Read each series against its own axis.
They do not add up. Crawlers and humans are two populations, not slices of one total.
The crawler series disappears when you turn the bot filter off, because in that state the human series already includes crawlers and drawing a second series would be the same data twice.
The AI-vs-human split bar
At the bottom of the dashboard: one stacked bar of page views classified by user agent, over your whole selected range.
The AI-vs-human split stacked bar with a human segment and several named crawler families, each labelled. Light theme, desktop, cropped to the card.
That is a proportion rather than a time series. It answers "of everything that fetched my pages this month, what share was each operator" — the number people screenshot.
Like everything on the AI panel, it ignores the bot filter.
What to look for
A crawler spike with no human movement is your content being collected. Common after you publish something, and again whenever a model operator does a crawl pass.
A crawler spike followed days or weeks later by human traffic on the AI channel is the sequence you actually want: your content was read, and then people arrived through an assistant citing it. Check Channels for the second half.
Crawler traffic that is a large multiple of your human traffic is normal on small sites and not a problem in itself — except that crawler views count against your monthly event limit. See What counts as an event.
Crawler traffic falling to zero after you changed your robots.txt means the
change took effect. Note the direction of proof though: crawler traffic
continuing after a robots.txt change means either the crawler has not
re-read it, or it is ignoring it. This panel cannot tell you which.
The one comparison that does not work
Do not read the ratio between the two series as "how much of my audience is AI". The crawler side is undercounted by an unknown amount because most crawlers do not execute JavaScript, so any ratio you compute is wrong in a direction you cannot measure.
Trends over time are fine — the undercount is roughly constant, so a rise is a real rise. It is the level that is not trustworthy. See How reliable is it.