Seatext library / BotRefund evidence
Can I Use SeaText AI with Custom-Built Integrations? A Step-by-Step Guide
Yes, SeaText AI supports custom integrations through its JavaScript snippet and event-based architecture. The platform installs in about one minute on any website and exposes visitor-level signals that developers can hook into for custom...
✓ Built for advertisers who need clear, refund-ready traffic evidence.
SeaText AI installs on any website with a single JavaScript snippet that loads in roughly one minute. Once active, it analyzes each visitor's browser, network, device, and behavioral signals — 106 independent checks — and uses an AI model to predict whether the visit is human or automated. The same snippet also powers content adaptation: translation, copy optimization, and mobile-friendly restructuring. Because the core runs client-side and emits events for each detection signal, developers can build custom integrations by listening to those events and sending data to their own endpoints.
There is no publicly documented REST API, webhook registry, or server-side SDK in the current SeaText AI documentation. Custom work therefore relies on the browser-side event layer. That approach works well for real-time personalization, analytics enrichment, and fraud-signal forwarding, but it does not support server-to-server workflows such as bulk audience sync, offline model training, or CRM updates without a middle layer you build and host.
What SeaText AI Actually Does on Your Site
SeaText AI describes itself as the first AI that enhances websites without requiring design changes. The snippet performs three main functions simultaneously:
- Bot detection and scoring: 106 independent checks across click, pointer, motion, speed, path, engagement, and session behavior. Each check produces an evidence signal; the AI model weighs the full pattern and returns a 99%-accurate human-or-bot verdict.
- Content adaptation: Dynamic translation for international visitors, copy optimization for engagement, and layout condensation for mobile screens.
- Signal exposure: The snippet fires browser events for each detection signal and for the final verdict, making them available to any JavaScript running on the same page.
All of this runs in the visitor's browser. No server-side API keys, OAuth flows, or webhook endpoints are mentioned in the public documentation.
How the Client-Side Event Layer Works
When the SeaText snippet loads, it attaches a global object (typically window.seatext or similar) that emits named events. The exact event names are not published in a formal spec, but the source pages describe signals such as ghost-click, linear-mouse, superhuman-speed, grid-aligned-path, no-tremor, static-session, and unnatural-duration. A final verdict event carries the AI's human/bot classification and confidence score.
Developers can subscribe to these events with standard addEventListener or the snippet's own on method, then forward the payload to any endpoint — your analytics platform, a custom data lake, a serverless function, or a CRM webhook you control. Because the events fire in real time, the integration latency is essentially the network round-trip from the visitor's browser to your collector.
Step-by-Step: Building a Custom Integration Today
- Install the snippet. Paste the one-line JavaScript include into your site's
<head>or via your tag manager. The source pages report a typical install time of one minute with no credit card required. - Inspect the event stream. Open the browser console on a page with the snippet active. Trigger a few visits (real and, if you have a test bot, automated) and log every event emitted by the SeaText object. Capture the payload shape for each signal type and the final verdict.
- Design your payload schema. Map SeaText signal names to your internal taxonomy. For example,
superhuman-speed→bot_indicator.input_speed_anomaly. Include the verdict, confidence, timestamp, and the visitor's anonymous SeaText ID if available. - Build the forwarder. Write a small JavaScript module that subscribes to the events, batches them (to avoid beacon spam), and sends them via
navigator.sendBeaconorfetchwithkeepaliveto your collector endpoint. Handle consent: only forward after the visitor has accepted analytics cookies if your policy requires it. - Deploy and validate. Push the forwarder alongside the SeaText snippet. Use your collector's logs to confirm events arrive with the expected schema. Compare SeaText verdicts against your ground truth (e.g., known bot IPs, honeypot form submissions) to calibrate confidence thresholds.
- Operationalize. Add monitoring for event volume drops (snippet blocked, CSP issues), schema drift (SeaText updates), and latency spikes. Document the integration for your team so future SeaText updates can be tested against your forwarder.
Integration Options Compared
| Approach | Best For | Setup Effort | Control & Customization | Limitations |
|---|---|---|---|---|
| Native snippet only | Bot protection, auto-translation, copy optimization | ~1 minute | Dashboard toggles; no code | No external data export |
| Client-side event forwarder (custom) | Real-time analytics enrichment, fraud-signal sharing, personalization triggers | Hours to days | Full payload control, any destination | Browser-only; no server-to-server; depends on snippet load |
| Server-side API (if released) | Bulk audience sync, offline modeling, CRM updates, historical replay | Unknown | Would enable true backend workflows | Not documented as of current public sources |
Takeaway: For any workflow that can tolerate browser-only delivery — analytics, real-time personalization, fraud-signal forwarding — the client-side event forwarder is viable today. For server-to-server needs, you must either poll a future API or build a middleware that receives the browser events and re-emits them server-side.
Key Facts from SeaText AI Public Sources
| Fact | Detail | Source |
|---|---|---|
| Install time | About one minute, no credit card | S1, S2, S4, S5 |
| Detection signals | 106 independent checks across 7 behavior categories | S1, S7 |
| Reported accuracy | 99% human vs. bot classification | S7 |
| Security certifications | ISO 27001, ISO 27017, ISO 27018 | S1 |
| Content adaptation | Translation, copy optimization, mobile condensation | S1 |
| Public REST API / webhooks | Not documented in current public pages | S1–S8 |
| Event exposure | Browser events for each signal and final verdict | S7 |
| Leadership | Sergei Gluhov (CEO), Yessi Montoya (CTO) | S1 |
Limitations and When This Advice Does Not Apply
- No server-side API today. If your architecture requires authenticated server-to-server calls, you cannot build that directly on SeaText AI without a middleware layer you host.
- Event schema is not versioned publicly. SeaText may add, rename, or retire signals without notice. Your forwarder must handle unknown event names gracefully.
- Consent and privacy. Forwarding behavioral signals to third parties may require additional consent under GDPR, CCPA, or other regulations. The snippet itself is ISO 27001/27017/27018 certified, but your forwarder becomes a new data processor.
- Ad-blocker and CSP impact. The snippet loads from SeaText's CDN. Strict Content Security Policies or aggressive ad blockers can prevent it from loading, which silences your integration entirely.
- Single-page-app nuances. If your site uses client-side routing, verify that the snippet re-initializes or continues emitting events on route changes. The public docs do not address SPA lifecycle.
Terminology Quick Reference
- Signal: One of the 106 independent browser/behavior checks (e.g.,
superhuman-speed). - Verdict: The AI model's final human/bot classification with confidence score.
- Forwarder: Your custom JavaScript that listens to SeaText events and sends them elsewhere.
- Middleware: A server-side service you build to receive forwarder payloads and re-emit them to internal systems.
- Beacon:
navigator.sendBeacon— a browser API for reliable, non-blocking POSTs during page unload.
Practical Scenarios
Scenario A: Enrich GA4 with Bot Probability
Forward the verdict event to Google Analytics 4 as a custom event parameter bot_probability. Build an exploration that segments sessions by probability threshold. No server code needed; the forwarder posts directly to gtag('event', 'seatext_verdict', { bot_probability: ... }).
Scenario B: Block Form Submissions for High-Confidence Bots
In your form's onsubmit handler, check the latest SeaText verdict stored in a global variable by your forwarder. If confidence > 0.95, prevent submission and show a friendly challenge. This runs entirely client-side and adds ~5 ms latency.
Scenario C: Feed a Custom ML Model Nightly
Your forwarder batches events to an S3 bucket via a Lambda function. A nightly Glue job parses the JSONL, joins with CRM outcomes, and retrains a proprietary lead-scoring model. This uses the middleware pattern because the model training runs server-side.
Frequently Asked Questions
Does SeaText AI offer a REST API for custom integrations?
Not in the current public documentation. The integration surface is the browser event layer emitted by the installed snippet.
Can I receive webhooks from SeaText AI when a bot is detected?
No native webhook system is documented. You can simulate webhooks by having your forwarder POST to your own endpoint, which then triggers downstream workflows.
What happens if the SeaText snippet fails to load?
Your forwarder receives zero events. Monitor event volume in your collector and alert on sustained drops. Consider a fallback: if window.seatext is undefined after a timeout, log a snippet_missing event to your analytics.
Are the 106 detection signals stable identifiers I can hard-code?
The source pages list example signal names but do not publish a versioned schema. Treat signal names as implementation details that may change. Build your forwarder to accept any event name and map known ones; ignore unknown ones rather than breaking.
Can I use SeaText AI's bot verdict to automatically request refunds from Google Ads or Meta?
SeaText AI's sister product BotRefund handles refund disputes. The source pages describe BotRefund as a separate service that "proves bot clicks, negotiates with Google and Meta, and gets your money back." SeaText AI provides the detection signals; BotRefund manages the refund workflow. They are integrated in the same suite but operate as distinct products.
What is the cost of the snippet and event access?
The source pages advertise a free install and free bot audit. Pricing tiers appear based on monthly ad spend (Under $10K, $10K–$50K, $50K–$250K, $250K–$1M, $1M–$5M, Over $5M). Custom integration work does not appear to incur a separate API fee, but you should confirm with sales if your volume exceeds the highest published tier.
How do I test my custom integration without polluting production data?
Use a staging subdomain with the same snippet. The source pages mention a "free bot audit" that runs a live detection on your site; you can schedule that for staging. Additionally, the window.open Tamper check page (S7) shows a side-by-side normal-vs-bot browser view — useful for generating realistic test events.
Further reading and comparison sources
These external sources provide additional context for evaluating the topic. Their inclusion is not an endorsement.
Learn more
Visit the website for more information.