Every decision has a bot score from 0 to 100: the more evidence of a bot, the higher. Five thresholds turn the score into a class (human_high, human, unknown, bot or bot_high), and your action per class, on the Set-up page, says what each class gets in block mode. You change the thresholds on the Bot score page; every change applies from the next request.
The thresholds
Lower thresholds call more visitors bots, and more people with them; higher ones let more bots through. Keep the defaults until you have looked at your decisions: in monitor mode, the Decisions page shows what block mode would have done with each one, as measuring before you block explains.
| Threshold | Default | What it sets | Lower it | Raise it |
|---|---|---|---|---|
| Bot | 60 | The bot score from which a visitor can be a bot. | More visitors are bot, and more people with them. | Fewer visitors are bot: more bots stay unknown. |
| Bot high | 80 | The bot score from which a visitor can be a bot with strong evidence. | More bots are bot_high and get its action, block by default, in place of bot's. | Fewer bots reach bot_high; the rest stay bot. |
| Suspect | 30 | The bot score, without where the browser runs and what it did before, from which a visitor is no longer clean: the human classes stay below it, and a bot needs it. | Fewer visitors are clean, so fewer reach the human classes, and bot comes sooner. | More visitors are clean, and bot needs more evidence about the visitor itself. |
| Behaviour | 30 | How much of the bot score the visitor's behaviour must add for bot and bot high: circumstance never reaches them on its own. | Less of what the visitor did is enough: strong circumstance and a little behaviour can make a bot. | A bot class needs more of what the visitor did; a script that behaves quietly stays unknown. |
| Human high | 70 | The human score that, with two kinds of proof of a person, holds bot off: the decision is unknown instead. | More visitors with proof of a person are held off bot. | Fewer are held off: more of them stay bot. |
The thresholds keep their order: suspect below bot, bot above 45 and below bot high, bot high at most 100, behaviour from 1 to bot, human high from 1 to 100. The Bot score page refuses a set that breaks it, and Back to the defaults there puts all five back.
From the score to a class
bot_high: a bot score at Bot high, and the visitor's behaviour adds at least the Bot threshold.bot: a bot score at Bot, at Suspect without where the browser runs and what it did before, and the visitor's behaviour adds at least Behaviour.humanandhuman_high: a bot score below Suspect, without where the browser runs and what it did before, and what your feedback told Bot Protection about that browser.unknown: everything else. A person on a first visit, with no feedback yet, isunknown: what one session shows, however much, never makes a human class.
Your feedback that a browser is a good customer holds off both bot and bot_high for that browser: the decision is unknown instead.
The kinds of evidence
The Decisions page names, for each decision, the kinds of evidence behind its score. Only behaviour can reach bot on its own; each of the other kinds stays below the Bot threshold by itself, which is why Bot stays above 45.
| Kind | What it looks at |
|---|---|
Behaviour behaviour | What the visitor did: how it moved, typed and went through your pages, the way a script does and a person does not. |
Inconsistency inconsistency | What the browser says it is and what it shows disagree, or what it claims cannot be true. |
Environment environment | Where the browser runs: a data centre, a virtual machine, a network many bots share. It raises the score, never makes a bot on its own. |
Missing signals missing_signals | Something a browser sends did not come. An ad blocker or a slow network can cause it, so it never makes a block on its own. |
History history | What this browser did on your site before, such as signing up again in another session. |
A visitor of yours caught by mistake: mark the decision Wrong on the Decisions page, and we fix it in the engine. A report is read by people, and never changes your thresholds by itself; for a whole period, use Report a problem.