Anti-adblock detection accuracy: how to read a 99% claim

When a vendor advertises "99% accurate", the first question is: 99% of what? Detection quality is not a single number. It splits into two that can move independently, and a tool can look excellent on one while failing on the other.

Precision answers: when the tool says a visitor is blocking, how often is it right? Recall answers: of all the visitors who really are blocking, how many does the tool catch? A tool can have near-perfect precision and poor recall, or the reverse. A bare "accuracy" figure hides which one you are getting.

When you see a 99% claim, ask the vendor to name it. AntiAdBlock Core reports 99.7% precision specifically, meaning a positive verdict is almost always correct. That is the number that protects your audience from false accusations, and it is the one a serious vendor states without hedging.

A false positive is when the tool tells a visitor with no ad blocker that they are blocking. To that reader the site looks broken or accusatory, they are shown an allowlist message they cannot act on, because there is nothing to allow. It is a bad experience inflicted on an innocent visitor.

This error is expensive because it hits your most valuable people: real readers who already accept ads. Annoy enough of them and you lose trust, engagement and sometimes the visit entirely. A tool with sloppy precision can cost more in goodwill than it recovers in impressions.

This is why precision is the headline number that matters. High recall with low precision is a tool that catches blockers by being trigger-happy at everyone. A 99.7% precision figure means the false-positive rate is low enough that the recovery messaging only reaches people who genuinely block.

A false negative is the opposite error: a visitor is running a blocker and the tool fails to notice. There is no broken page, no complaint, no symptom at all, which is exactly what makes it dangerous. The blocked impression is lost and nothing tells the publisher it happened.

False negatives are where Manifest V3 hurts most. uBlock Origin Lite blocks at the network layer and does little cosmetic hiding, so a tool relying on DOM bait quietly misses it. The publisher sees a clean dashboard and assumes coverage is fine while a growing share of blockers slip through.

Recall is what governs false negatives, and recall is the figure vendors are quietest about. Ask specifically: what is the detection rate against uBlock Origin Lite, across its modes? A tool that cannot answer that is almost certainly leaking blocked impressions it never reports.

An accuracy number is a measurement taken on a date, and ad-blocking does not stand still. The same tool measured three months later can score very differently, because filter lists have moved against its probes and blockers have shifted in popularity.

A tool with no retraining loop only ever decays from its launch-day figure. Its quoted accuracy becomes a historical artefact, not a current property. The honest way to read a static tool's 99% claim is as the best it ever did, on the day it was measured, under conditions that no longer hold.

A tool that retrains keeps its figure live. AntiAdBlock Core retrains nightly against EasyList, uBlock and AdGuard deltas, so its precision is maintained rather than remembered. When you compare accuracy claims, weight a maintained number far above a one-time number, even if the one-time number looks higher.

Five questions separate a measured claim from a marketing one. First: is your headline number precision or recall? A vendor who cannot answer instantly does not measure carefully. Second: how was it measured, on synthetic tests or live traffic, and how large a sample?

Third: what is your detection rate specifically against uBlock Origin Lite? Fourth: how often is accuracy re-measured, and does the engine retrain between measurements? Fifth, and most telling: can I verify the number on my own traffic before paying?

That last question is the real test. A vendor confident in its accuracy makes verification easy. AntiAdBlock Core's free tier exists for exactly this, install it, watch precision and recovery on your own audience, and trust the number you measured rather than the one on the landing page. Accuracy you can reproduce is the only accuracy worth buying.

Once you have a vendor with transparent accuracy data, the next step is verifying that the adblock killer script they provide covers the full blocker landscape, including MV3 extensions that older accuracy benchmarks did not measure. The anti-adblock that actually works guide shows how to validate ongoing performance once the script is live.