How to read a bot traffic spike

Do not block anything yet

The instinct at this moment is to find the biggest source and cut it. That instinct is wrong often enough to be expensive. A spike has three common explanations and only one of them is worth acting on today.

• A crawler found you. Harmless, sometimes flattering, and it should never have been billed to you anyway.
• A campaign changed. New placement, new zone, new supply partner, and the quality came with it.
• Someone is sending you traffic that is not real. This is the one that needs action.

Step 1. Split declared bots from the rest

Open the bots screen and look at the GIVT and SIVT split first. GIVT is declared automation: search crawlers, monitoring, uptime checks. A GIVT spike is almost never your problem. SIVT is the category that had to be caught rather than declared, and a SIVT spike is a real event.

The vocabulary here is not ours. It comes from the invalid traffic standards the ad industry already uses, which is why an ad network will recognize it when you quote it back to them.

Step 2. Check whether the spike has a name

If the traffic is a named crawler, the row will say so and show what verified it. A crawler that claims to be Googlebot and fails verification is a different animal from one that passes, and Google publishes the check itself so you can confirm our verdict independently.

Named AI crawlers are the common case for publishers in 2026. Both OpenAI and Anthropic publish which user agents they operate and how to control them, so a spike from one of those is usually just your content being read.

Step 3. Give the spike an address

A percentage tells you nothing you can act on. Open the attribution screen and find which zone, placement, or sub ID carries the new traffic. If the spike has no source attached, it is not something an ad network can block for you.

Step 4. Wait for the volume floor

A source with 12 visits and an 80 percent bot share is not an 80 percent bot source. It is 12 visits. TrueVisit leaves sources below the floor unscored on purpose, and you should apply the same patience: a day of data is usually enough, an hour rarely is.

The two expensive mistakes

• Blocking a zone on the strength of its worst hour. Blocking a zone blocks every campaign running on it, so judge it on everything it sent.
• Treating a suspected verdict as proof. Suspected means the behavior scored above a threshold, not that anything was verified. It deserves a second look, not a block.