Privacy
Last updated: 4 November 2026
Role
When you install SentientUI on your site, you are the data controller for your visitors. SentientUI acts as a processor: it handles the data below only to run optimisation, measurement and reporting for you. We do not pool your visitors' data with other customers' or train a model across customers. See the DPA for processing terms and subprocessors.
What the SDK sends when tracking is allowed
Depending on your integration and consent settings, the browser SDK and snippet send:
- A visitor id. A random identifier in a first-party cookie (
_snt_uid_<key>), kept for 365 days from the visitor's last visit and shared across your site's own subdomains (setcookieDomain: falseto keep it to one host), plus local and session storage keys for assignment continuity and retrying unsent events - Visit context. Device class (mobile / tablet / desktop, worked out in the browser — the user agent itself is not stored), traffic source and the referring site's domain (never the full referring URL), the landing path without its query string, UTM values, local time of day and weekday, the SDK version, and whether the browser reports that it is automated
- Ad click identifiers when present on the landing URL (
gclid,fbclid,ttclidand similar). Their presence classifies how a visit arrived (paid search, paid social, email) and, where you enable it, can shape what that visitor is shown; the identifier values are never used to decide what anyone sees and are never shared with an ad platform - Events. Page views (path only — query strings and fragments are removed on our side too), which version of each component was shown, and the goals and conversions you define, with any value, currency and metadata your site attaches. Metadata is stored as sent, so do not put personal data in it
- Engagement signals (on by default): time spent and scroll depth per page section, rage clicks, text copying (the length of the selection only, never the text), scroll pauses and tab switches
- Interaction statistics, computed in the browser and sent as totals, used to tell people from automated agents: counts of pointer moves, clicks, touches, scrolls, key presses, characters typed, backspaces and navigation keys; how long clicks are held; one number for how regular the time between key presses is; how long the gaps between actions are; viewport size, screen pixel density, the reduced-motion preference, and whether graphics are hardware- or software-rendered (the graphics card's name stays in the browser). Which keys were pressed and what was typed are never sent
- Page structure. For each page section: its tag, id and class names (up to 200 characters), its heading (up to 160 characters), text length, number of buttons and links, ARIA role, and yes/no flags for content patterns such as pricing or testimonials. Adaptive regions (
<AdaptiveSlot>) also send their rendered text (up to 400 characters) so your original version can be recorded — turn that off withreportBaselineText={false}. Fuller page text is sent only if you setcaptureDomText - A user id if you call
identify(), stored as you provide it (up to 256 characters) so one person's visits join up across devices. With verified identify, the signature is checked and not stored. A verified sign-in can carry the traits we inferred for that person (for example how price-conscious they seem) to their new device - A hashed email and its domain, only if you pass an email to
identify()so your own team's visits are kept out of your numbers: the SDK sends a SHA-256 hash of the lowercased address and the part after the @, never the address itself. Email rules you add in Settings are stored the same way, with a masked form (for example c***@example.com) so you can tell them apart. We also record which of your sites a visit was on - A visitor type if your site declares one (for example a plan tier); values that look like an email address or a name are dropped
- Purchase-stage history, only where you enable it: on your site alone, how many purchases a visitor has made and when they last bought or added to cart — derived from the goal events you already send, never order contents
What we derive from the IP address
During each request that starts a visit, we look the address up in an offline database to get the country, the network operator (ASN) and whether the network is a data centre or residential, and we store those three facts with the visit. We also store a keyed, one-way hash of the network the address belongs to (the first three parts of an IPv4 address, or the first 48 bits of IPv6), scoped to your project, so several visitors from one network count as one voice when recording your original page. The address itself is not stored in our database.
Shortened addresses (the network part only) appear in our request logs and in audit records of account actions. Full addresses appear briefly in rate-limit counters at our caching provider — about two minutes for visitor traffic, up to 48 hours for the public site scan and demo — and then expire.
The lookup uses IP Geolocation by DB-IP, licensed under CC BY 4.0. It runs inside our infrastructure — no visitor address is sent to DB-IP.
What your server sends, if you integrate it
- AI crawler requests (Next.js
AdaptiveRoot,captureAgentRequestor the agent middleware): the path, the user agent and the recognised agent's name, only for requests whose user agent matches a known AI agent or crawler. No IP address, no cookie. This runs regardless of consent settings because it concerns automated clients, not people - Server-side goals, conversions and refunds: an event or order id, value, currency, time and a visitor or cart reference. Never the customer's name, email, address or items bought
What we do not collect
- Names, email addresses or contact details — unless you pass them in a user id or metadata
- Form field values or what visitors type
- Full URLs with query strings, or the full referring URL
- Cross-site tracking: the visitor id is first-party and never shared with another site
- A device fingerprint: visitors are recognised only by the random cookie. The coarse environment details above are used to tell people from automated agents, not to identify anyone
- Third-party advertising pixels on your site
Automated visitors
Visits whose browser reports automation, or whose user agent belongs to a known bot or AI agent, are shown your site's original markup rather than any adapted version. Those visits, and visits our bot score rates as non-human (from signals such as how the visit behaved and whether it came from a data-centre network), are kept out of learning, conversion rates and your billed visitor count. A newer behavioural score built from the interaction statistics above is calculated but does not yet change any of these outcomes.
AI processing
Some features call a language model (Anthropic's Claude, routed through Vercel's AI Gateway). What we send: aggregated statistics about your project and its audience groups, the public content and section headings of your pages, messages you write to the in-dashboard assistant, and — only if you connect a repository for code changes — the relevant source code. These prompts are built from aggregated statistics, not individual visitors' event histories. Model output is never generated at the moment a visitor is served: written versions are created ahead of time and, with review mode on, wait for your approval.
Shopify stores
When a store is connected through the SentientUI Shopify app we also process, per paid, cancelled or refunded order, its Shopify order id, total, currency and time, attributed to the visitor's pseudonymous session — never the customer's name, email, address or what they bought. The checkout pixel binds Shopify's cart/checkout token to that session (deleted after 90 days). A customer-tag → visitor-type mapping, if the merchant sets one, is applied in the storefront and only the resulting visitor type is sent to us, not the tags. Merchants who buy a plan through the app are billed by Shopify, which sends us the plan and subscription status.
Shopify's customers/redact removes the named orders' ids from our records (the anonymous amount stays in revenue reports until normal retention removes it), customers/data_request is answered to the store owner, and shop/redact deletes the app's stored keys and session.
Personal data and identifiers
Online identifiers (including pseudonymous visitor ids) and linked behavioural history can be personal data under GDPR / UK GDPR, especially when activity is linked over time. Treat the visitor id and any user id accordingly: obtain consent where ePrivacy / GDPR require it before tracking runs.
Cookies and consent on your site
Tracking is on unless you gate it. Gate it on your consent platform with consentFrom, or with the consent prop; on Shopify storefronts the snippet follows Shopify's own customer-privacy setting. Do Not Track and Global Privacy Control are honoured by default and override a granted consent. Before consent the SDK sets no cookie or storage and sends no visitor id; it may still fetch your site's configuration or its best-performing version, so those requests reach our servers. When consent is withdrawn the SDK deletes its cookie and storage. See the consent documentation.
Cookies on this website
On sentient-ui.com and the dashboard you choose which categories we may use — optimisation & analytics (the SDK above) and marketing (the Meta Pixel) — and can change or withdraw that choice at any time; a withdrawal takes effect immediately: . For visitors outside the EU, EEA, UK and Switzerland, optimisation & analytics starts switched on; everywhere, marketing starts switched off.
Separately, this website and the dashboard use Vercel Web Analytics and Speed Insights on every page: cookie-less page-view and performance measurement that stores nothing on your device. Shared-report and connect links are redacted before anything is sent.
Retention
- Raw events: kept for your project's retention window — 90 days on Free and Starter, 180 on Growth, 365 on Scale and Agency, as agreed on Enterprise. A project admin can choose a shorter window (at least 30 days) in Settings
- A visitor's record and everything tied to it (versions shown, goals, attributions): deleted once that visitor has been inactive for the project's retention window
- AI crawler request logs: 30 days on Free and Starter, 90 days on Growth and above
- Kept until you delete the project: learned results per version and audience, page structure (including section headings), dashboard assistant conversations (older messages are condensed into a summary), and monthly unique-visitor counts used for billing, which keep the visitor id — except for an erased visitor, whose entry keeps counting but has its id replaced with a random placeholder
- Database backups: taken daily and kept for 90 days
Export and deletion
Project owners can export via the management export endpoints. To erase one visitor, call DELETE /v1/visitor with their visitor id: collection and serving for that visitor stop at once, and their stored data is deleted within about 25 hours. Erasing by user id from the dashboard (or the management API) deletes at once (very large requests finish within the hour). Either way, the visitor's entry in the monthly billing count is kept so your bill does not change, but its id is replaced with a random placeholder. Erased data is not removed from backups, which expire within 90 days.
Deleting a project in the dashboard deletes its data immediately. Cancelling a paid plan does not delete anything — the account moves to the Free plan. You can delete your account yourself in Settings → Account. If you own the account, this deletes your team and all its projects and data at once, and cancels any card subscription immediately (the unused part of the current period is not refunded); if you were invited to someone else's team, only your login and membership are deleted. Plans billed through Shopify must first be ended by uninstalling the app. Deleted data expires from backups within 90 days, and our security audit log keeps a record that the deletion happened (your account id, not your email). We respond to privacy requests within 30 days.
Where data is processed
The API, background worker and database run on Fly.io in London (UK); the Shopify app runs on Fly.io in Paris. Language-model calls, backups, and some supporting services are processed in the US — the DPA lists each subprocessor, what it receives and where. Engage counsel for your jurisdiction; this page is a product inventory, not legal advice.
Contact
Questions about this policy: hello@sentient-ui.com

