Seatext library / BotRefund evidence
On-Site Bot Evidence Generation: How It Works and Why It Matters for Ad Refunds
On-site bot evidence generation is the process of collecting verifiable, session-level proof that clicks on your Google and Meta ads came from automated traffic rather than real people. BotRefund captures over 100 independent behavioral,...
✓ Built for advertisers who need clear, refund-ready traffic evidence.
What on-site bot evidence generation means
On-site bot evidence generation is the systematic collection of technical and behavioral signals that prove a website visit was automated. Instead of relying on a single heuristic — like a known bot IP list — the system records dozens of independent checks during each session: how the mouse moves, whether clicks follow human intent sequences, whether browser and network data agree, and whether timing patterns match real reading and decision-making. Each check produces an objective fact (a signal). The signals are then weighed together by a prediction model that outputs a bot-or-human classification with a documented evidence trail. That trail — video replays, signal logs, and timestamps — is what ad platforms such as Google Ads and Meta accept when you dispute invalid clicks and request refunds.
Why the evidence layer matters for ad budgets
Bot clicks can consume a meaningful share of paid search and social budgets. BotRefund's data indicates that automated traffic can account for up to 20% of Google and Meta ad spend. Without session-level proof, advertisers typically rely on platform-side invalid-click filters, which are opaque and often leave budget on the table. On-site evidence generation shifts control to the advertiser: you capture the visit as it happens, preserve the raw signals, and present a reproducible case that the platform's own billing team can review. The result is a documented refund pipeline that can reach back several years — BotRefund notes recovery eligibility for Google Ads spend dating back to 2017.
How the detection signals are organized
The evidence engine groups its 106 independent checks into behavioral, network, and browser categories. Behavioral checks watch what the visitor does: ghost clicks that fire without a preceding intent sequence, honeypot interactions with hidden page elements, linear mouse paths that lack natural tremor, superhuman input speeds under one millisecond, grid-aligned movements that snap to precise coordinates, sessions with no scrolling or clicks, and visit durations that are too short, too long, or suspiciously uniform. Network and geolocation checks look for mismatches such as suspicious port usage, VPN or proxy rotation artifacts, and inconsistencies between declared location, language, and connection metadata. Browser-level checks examine automation properties, console debug artifacts, and monitor synchronization anomalies that reveal scripted environments. Each check is designed to produce an independent fact, not a verdict.
From raw signals to a refund-ready evidence package
The system follows a three-step chain for every session. First, each check adds one objective fact — for example, "mouse path snapped to grid coordinates" or "connection used a port commonly associated with proxy rotation." Second, the engine cross-checks whether other independent signals tell the same story; a single anomaly is kept as evidence but not treated as a bot verdict because privacy tools, corporate networks, travel, and unusual devices can create outliers for real people. Third, the complete pattern feeds a prediction AI that weighs all signals together and classifies the visit as bot or human with a reported 99% accuracy. The output includes a video replay of the session, a timestamped signal log, and a summary classification that can be exported and sent to a Google or Meta representative to open a billing dispute.
Using the evidence: audit, export, claim
The practical workflow starts with a free on-site audit. Adding the detection script takes about one minute and requires no credit card. The audit runs live, captures traffic, and produces a report you can review. When you see bot sessions, you export the evidence package — video, signal list, timestamps — and send it to your platform rep. BotRefund states that 83% of its customers successfully obtain a refund through this process, and the average approved rate across submitted claims is tracked as a platform metric. The service also handles negotiation and escalation for enterprise accounts, mapping out a recovery, protection, and escalation plan based on your monthly Google/Meta spend tier.
Limitations and when the approach does not apply
On-site evidence generation only covers traffic that reaches your website and executes the detection script. It cannot see clicks that bounce before the script loads, traffic blocked by ad-platform filters before landing, or invalid activity on platforms that do not allow third-party measurement. The evidence is only as strong as the signal coverage; sophisticated bots that perfectly mimic human biomechanics, browser fingerprints, and network coherence may evade detection. Privacy regulations (GDPR, CCPA) require proper consent handling for session recording and signal collection. Finally, refund approval remains at the discretion of Google and Meta; the evidence package improves your position but does not guarantee a specific recovery amount.
Key facts
| Fact | Detail | Source |
|---|---|---|
| Independent checks per session | 106 | S3, S6 |
| Reported classification accuracy | 99% | S3, S6 |
| Behavioral signal categories | Click, trap, pointer, motion, speed, path, engagement, session | S1, S2 |
| Network/geolocation signals | Suspicious ports, VPN/proxy rotation, location-language-timing coherence | S3 |
| Browser-level signals | Automation properties, console debug, monitor sync anomaly | S5, S6 |
| Setup time for free audit | About 1 minute | S1, S2, S4, S5, S7 |
| Refund lookback window (Google Ads) | Dating back to 2017 | S1 |
| Customer refund success rate | 83% | S1 |
| Estimated bot share of ad budget | Up to 20% | S1, S2, S4, S5, S7 |
| Evidence output format | Video replay, timestamped signal log, classification summary | S1, S3, S6 |
Terminology quick reference
- Ghost click — A click event that fires without the preceding human intent sequence (hover, focus, natural approach).
- Honeypot trap — A hidden or deceptive page element that only automated scripts interact with.
- Mouse tremor — The micro-jitter present in human pointer movement; absence suggests scripted input.
- Superhuman input speed — Interactions completed in under 1 ms, faster than physiological limits.
- Grid-aligned movement — Pointer paths that snap to exact pixel rows/columns instead of natural curves.
- Monitor sync anomaly — Mismatch between reported display refresh timing and input event timestamps, revealing virtualized or headless environments.
- Suspicious ports — Network ports commonly used by proxy rotation services or tunneling tools that real residential browsers rarely expose.
- Cross-checked context — The process of verifying that multiple independent signals support the same conclusion before classifying.
Frequently asked questions
How is on-site evidence different from Google's or Meta's built-in invalid-click filters?
Platform filters run server-side and are opaque; you see a credit after the fact but not the session-level reasoning. On-site evidence gives you the raw signals, video replay, and a reproducible log you can present during a dispute, extending the lookback window and letting you challenge clicks the platform may have missed.
Does the script slow down my site or affect Core Web Vitals?
The source pack states setup takes about one minute and implies a lightweight client-side collector, but it does not publish specific performance metrics. Test in a staging environment and monitor LCP, FID, and CLS before full rollout.
Can I use this evidence for platforms other than Google and Meta?
The documented refund workflow and success metrics (83% customer refund rate, approved rate tracking) are specific to Google Ads and Meta. Other platforms may accept similar evidence, but no outcomes are published in the source pack.
What happens if a real user triggers several anomaly signals (e.g., corporate VPN, accessibility tools)?
The system treats each anomaly as evidence, not a verdict. The AI prediction step weighs the full pattern across 106 checks, so isolated mismatches from privacy tools, corporate networks, or assistive technology rarely flip the classification alone.
Is there a minimum ad spend required to benefit?
The audit is free for any spend tier. The source pack lists spend ranges from under $10,000/mo to over $5M/mo, with enterprise escalation plans for higher tiers. Recovery potential scales with bot-click volume, which tends to correlate with spend.
How long does a typical refund cycle take?
The source pack does not publish a standard timeline. It notes a "fast setup" (1 minute) and that the service negotiates on your behalf, but platform review cycles vary. Plan for several weeks to a few months depending on claim complexity and platform responsiveness.
Can I run the detection without committing to the refund service?
Yes. The free bot audit lets you install the script, collect evidence, and export the report. You decide whether to pursue claims yourself or engage the managed negotiation path.
Further reading and comparison sources
These external sources provide additional context for evaluating the topic. Their inclusion is not an endorsement.
How BotRefund can help
BotRefund installs a lightweight script on your site in about a minute and starts a free live bot audit. The engine runs 106 independent checks — behavioral, network, and browser — and cross-checks every anomaly before its AI classifies the session with a reported 99% accuracy. You get a video replay and a timestamped signal log for each flagged visit, ready to export and send to Google or Meta for a billing dispute. The service also handles negotiation and escalation for enterprise accounts, with refund eligibility back to 2017. No credit card is needed to start the audit.
Limitations: the script only sees traffic that loads on your page; clicks that bounce before load or are filtered upstream are invisible. Sophisticated bots that perfectly mimic human biomechanics and browser coherence may evade detection. Refund approval remains at the platform's discretion. You must handle consent for session recording under GDPR/CCPA.