How Button works

Button examines every visit through four independent layers of evidence. It classifies the visit only on strong signals and keeps the rest as evidence you can inspect. Then you decide what that traffic may influence.

From page load to decision

Understanding comes first. Nothing on your site changes until you choose an action.

  1. Record every visit

    One script, added directly or through your tag manager, records each page load and the ad click that brought it. You start listen-only: nothing is blocked and your ad account isn't connected.

    Read the documentation →
  2. Gather the evidence

    Each visit is examined at four layers: the network it comes from, how it communicates, the browser environment it runs in, and how it behaves.

    Traffic Detective →
  3. Classify on strong signals

    Visits are classified as trusted or questionable, each with the reasons behind it. You can follow every click to the pages it reached. A page reached is not a verified purchase.

    Ad Health →
  4. Decide what counts

    If you choose, load or withhold your ad tags by classification on the paths you pick, or verify form submissions server-side. Or simply keep observing.

    Smart Responder →ButtonCaptcha →

Four layers of evidence

Each layer answers a different question about the same visit. No layer is read on its own.

Where it comes from

Network

  • IP reputation
  • Infrastructure characteristics
  • Proxy and datacenter indicators
  • Network consistency

Context, not identity: shared, mobile and residential networks carry real people too.

How it communicates

Protocol

  • HTTP characteristics
  • TLS characteristics
  • Request behavior
  • Consistency between them

A client claiming to be one browser while communicating like another stands out.

What it runs in

Browser

  • JavaScript execution
  • Browser and device characteristics
  • Runtime behavior
  • Rendering properties

The question is whether the claimed environment is coherent: no impossible attributes, contradictory hardware or automation-generated state.

How it acts

Behavior

  • Click patterns and clustered clicks
  • Session consistency
  • Conversion behavior
  • Repeated activity

Essential because unwanted activity can come from real browsers on real devices.

Why no single signal decides

A browser fingerprint is useful but incomplete. Attributes can be spoofed. Many signals last only a browser version or two. A single quirk also has innocent explanations. Button doesn't search for one bot signature. It correlates independent observations across the four layers.

Strong signals classify. Everything else is forensic evidence. Weaker observations never decide a classification on their own. They're kept to corroborate it, explain it and show edge cases. The result is a classification with named reasons, not a probability score or a single fraud rating.

Replay. A static challenge can be captured once and replayed many times. Button's answer is to keep changing what runs in the browser. That raises the cost of replaying it, instead of relying on a challenge that stays the same.

Not all automation is unwanted

Classifying a visit as automated doesn't mean it should be blocked. The question is what generated it, what evidence supports that, and what it should be allowed to influence.

Often useful

  • Search engine crawlers
  • Ad verification tools
  • Helpful AI crawlers
  • Authorized testing systems

Often unwanted

  • Click automation
  • Invalid interactions
  • Scraping systems
  • Advanced automation built to look human

Every classification should answer

  • What traffic was observed?
  • Why was it classified this way?
  • Which strong signals contributed?
  • What supporting evidence exists?
  • What business impact does it have?

What Button doesn't do

  • It doesn't change your bids, budgets or campaigns, or update ad-platform IP exclusion lists.
  • It doesn't refund clicks already billed. Withholding a tag affects what your site reports back from then on.
  • A questionable classification isn't proof that a click was fraud, and a trusted one isn't proof of a buyer.
  • A goal page reached isn't a verified sale. Purchases need your own order or CRM records.

A funnel with bots in it

Bots arrive in bursts, scan your pages and go nowhere. Real visitors move toward strong intent. Turn Button on to see the difference in this model.

0Orders
$0Revenue
$0Ad spend
$0Profit

Illustrative simulation, not customer data. Every arriving click costs $10, bots included; every real visitor who reaches the end is a $60 order. Bypass is the share of traffic let through unchecked. Totals are since the page loaded.

Get started

See your own funnel. With the evidence.

Start by observing visits and reviewing the evidence.