Skip to content
Glossary

Bot detection is a probability estimate, not a test

No site can prove a visitor is a program. It scores signals, sets a threshold, and accepts that both kinds of mistake will happen. Understanding that is most of understanding why you were blocked.

The four families of signal

Network reputation


The address you arrive from carries history. Datacentre ranges score differently from consumer connections, and a range that other people abused yesterday is a range you are judged by today. This is usually the largest single term and the one you influence least.

Client characteristics


Headers, TLS handshake shape, and the properties a page can read from the browser. These describe what kind of client you are. Consistency matters more than any single value: a combination that could not occur on a real device is a stronger signal than any one attribute.

Behaviour over time


Request rate, navigation order, dwell time, whether resources are fetched. This is where most automation is actually identified, and it is the family a well-behaved client naturally scores well on because it is largely a description of restraint.

Declared identity


The newest family, and the one moving fastest. Web Bot Auth and comparable schemes let a client cryptographically sign that it is a specific known agent, which turns the question from detection into verification.

What a block usually means

Likely cause The useful response
Immediate block, first request Network reputation Change egress, or ask for an allowlist
Works, then blocks after minutes Request rate Slow down; this is the common case
Interstitial challenge on every visit Client characteristics Use a real browser, not an HTTP client
Blocked only on certain pages Page-level policy Check robots directives and terms
Blocked after login Account-level terms Read what the account agreed to
Intermittent, no pattern Threshold near your score Reduce rate and stabilise the client

As of 2026-08-31 · Competitor details come from each vendor's published pricing page on that date.

Diagnosing a block without guessing

This is a reliability investigation. Treat it as one.

  1. Read the response, not just the status

    Many blocks carry an explanation, a reference id or a contact route in the body. Teams routinely retry for days against a page that says why in the first paragraph.

  2. Establish whether it is you or your address

    Try the same request from a different network by hand. If a browser on a phone succeeds, the signal is network reputation and no amount of client work will help.

  3. Halve the rate and re-measure

    Rate is the cheapest variable to change and the most common cause. If halving fixes it, you have found the answer and it was never about detection sophistication.

  4. Check what you agreed to

    Robots directives, terms, and for authenticated access the account terms. If the answer is that this access is not permitted, the engineering question is settled.

  5. Ask

    An identifiable user agent with a contact address gets asked to slow down far more often than it gets blocked. Many operators will allowlist a stated purpose that they would otherwise refuse.

Common questions

Is bot detection the same as CAPTCHA?

No. A CAPTCHA is one possible response to a score, not the detection itself. Most detection is invisible and most decisions are made before any challenge is shown, which is why a system that only reacts to visible challenges is reacting to the tail of the process.

Why does my automation get blocked when a browser does not?

Usually rate and consistency rather than any single tell. A person generates a handful of requests with irregular gaps and fetches every resource; automation often generates hundreds with even gaps and fetches only the document. The second pattern is visible without any fingerprinting at all.

What is Web Bot Auth?

A scheme for a client to prove which agent it is by signing requests with a key tied to a published identity, so a site can allow a known agent deliberately rather than infer intent from behaviour. Several large networks now support declaring agents this way, and it is the direction the problem is being solved in.

Do you help bypass bot detection?

No. We do not sell proxy rotation, fingerprint spoofing or CAPTCHA solving, and if your target is an adversarial public site then several other vendors fit your problem better than we do. Our position is that automation should be able to say what it is.

Automation that identifies itself

If your workload only works while hidden, we are the wrong tool, and we would rather say so before you integrate.