A device intelligence API is easy to buy and easy to misread. Teams wire it in, store the raw response, and months later nobody can say which field should have stopped a signup. The payload had forty keys; the decision needed three.
This post is about the return value: which fields drive a real decision, what each one means, and how to turn them into a policy you can explain to support, to finance, and to a customer who asks why they were challenged.
It assumes you already call device fingerprinting from signup, login, trial or checkout. If you are still choosing between building and buying, our device fingerprinting overview covers the tradeoffs; here the question is what comes back.
Start with the decision, not the field list
Every field you cannot act on is storage cost and privacy liability. Before you read a payload, write down the outcomes your product supports: allow, step-up (captcha, email verification, passkey), block, and occasionally manual review. Then keep only the fields that move a request between them.
A response you can reason about
{
"visitorId": "…",
"confidence": 0.92,
"bot": { "detected": false, "kind": null },
"network": { "vpn": false, "proxy": false, "tor": false, "datacenter": true },
"tampering": { "detected": false },
"reputation": { "events": 0 },
"requestId": "req_…"
}
Log the decision, the fields that produced it and the requestId. That trio lets you reconstruct a case six weeks later without keeping every payload forever.
visitorId and confidence: the fields you build on
visitorId is a stable handle for a device and browser combination. It should survive a cleared cookie, stay comparable across requests, and never be treated as a person: it is a key for joins, not a verdict.

confidence says how well that identifier was supported by real signals on this request. When it drops, your policy should too. Let a low-confidence identifier inform a step-up, never a block on its own — a weak identifier treated as proof creates support tickets, not safety.
Bot and AI-agent signals: separate automation from intent
A single “is a bot” boolean is not enough. You need to know whether automation was detected and what kind: scripted traffic hitting a signup form is a different problem from an AI agent reading your pricing page.
Policy follows from that difference. Block automation at account creation, checkout and OTP requests; count it but do not block it on marketing pages; keep bot sessions out of product analytics so your funnels describe humans. Separating scripted bots from AI agents is what keeps a policy honest in both directions.
Network signals: VPN, proxy, Tor and datacenter
VPN use is not fraud. Privacy-conscious customers, remote teams and mobile carriers all produce network signals that look unusual. What matters is the combination: a datacenter address, a fresh device and a domain registered days ago is a different request from a VPN login on a device you have seen for a year.

So return those facts separately and let your geo rules read them. Collapse them into one “bad network” flag and you lose the ability to treat legitimate VPN users well.
Tampering and reputation: what the client cannot hide
Tampering means the environment is misrepresenting itself: patched browser APIs, spoofed canvas or audio output, automation hooks left in place. A tampering flag is stronger evidence than any single anomaly, because it describes intent rather than geography.
Reputation is the same idea across accounts. Return how often a device has been seen and in which kinds of events, so policy can separate a first-time visitor from a device attached to a dozen recent signups. Our network page describes how those signals are pooled and what is never stored in the clear.
From fields to policy
- Allow: no bot, no tampering, healthy confidence, clean reputation.
- Step-up: VPN or datacenter network, low confidence, one weak anomaly.
- Block: tampering combined with automation, or a reputation pattern across accounts.
- Review: high-value checkout with mixed signals.

Then test it before you enforce it. Run the flow in a mirror period, compare your would-be decisions with what actually happened, and turn on blocks for the cheapest path first. Prynt Identify returns the signals and your code makes the call; the docs show the webhook shape and how sealed results are verified.
FAQ
Should the API return a verdict or just signals?
Signals with clear reason codes. A vendor that returns “block” hides its reasoning, and you cannot tune, appeal or explain it to a customer. Keep the verdict in your own policy layer.
How long should we keep the raw response?
Keep the decision, the reasons and the requestId for as long as you keep the account. Keep full payloads for a short window only, and treat identifiers as personal data under your own retention rules.
Can the same payload identify the company behind a visit?
No. A device identifier describes a browser, not an organization. Company identification needs a different source of truth — Prynt Reveal names companies from their own network and links people only from first-party data.
Try it on your own traffic
Prynt Identify returns visitorId, confidence, bot and AI-agent detection, VPN, proxy, Tor and datacenter signals, tampering and reputation from one call, with 5,000 identifications a month on the free plan and no card. Create an account at pryntid.com and read the payload in the docs.
Try it free
Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.