Privacy
What RefinedAI stores, and who else sees it.
Draft — not yet reviewed by a lawyer.
This document describes how the product actually behaves today, written from the application source. It has not been through legal review and is not a final policy. Sections that need input Aemorph has not yet supplied are marked inline in the same dashed style as this notice.
[Aemorph to supply: last-updated / effective date]
The most important thing on this page
RefinedAI works by sending your tracked phrases to third-party AI assistants and recording what they say back. Every phrase you track is transmitted to OpenAI, Anthropic, Google and Perplexity (and, for search-surface checks, to DataForSEO), together with the country of the market it belongs to. That transmission is not incidental — it is the product.
If a phrase is confidential, do not track it. Those providers handle the phrase under their own terms, not ours.
What we store
The list below is taken from the application's database schema. It is what the product actually writes, not a generic description.
- Account identity — the user id, email address, name and profile image supplied by Clerk when you sign in, plus your plan tier and notification preferences.
- Brands — brand name, website domain, aliases, and the description, category and target-audience details you enter.
- Markets — the country (and language) each brand is tracked in.
- Tracked phrases — the phrases themselves, where they came from, and whether they are active.
- Monitor runs — each scheduled or manual run, its status, and when it ran.
- Model answers — the exact prompt sent, the exact model used, and the literal answer text returned, stored verbatim so you can audit any figure we report.
- Derived analysis — brand and competitor mentions extracted from those answers, with position, sentiment and framing.
- Citations — the URLs the models cited and the entity each citation supported.
- Cost records — one row per model call with token counts, latency and cost, used for budgets and abuse limits.
- Free GEO audit submissions — if you run the public free audit without an account, we store the email address, brand name, domain and country you submit, plus a hashed form of your IP address used only to rate-limit abuse. The raw IP address is not stored.
- Waitlist entries — an email address and the module it was submitted for.
Sub-processors
The following third parties process data on our behalf. This list was compiled from the application source and the production environment configuration.
- Clerk — authentication and user identity.
- Supabase — hosts the primary Postgres application database, where everything in the section above is stored.
- Vercel — application hosting and serving.
- Upstash Redis — provisioned through Vercel KV; used for rate limiting and short-lived caches of in-progress work.
- Inngest — executes background jobs, including the scheduled monitoring runs.
- OpenAI, Anthropic, Google (Gemini) and Perplexity — the AI assistants your tracked phrases are sent to, and the source of the answers we store and analyse.
- Vercel AI Gateway — an optional routing layer in front of the OpenAI, Anthropic and Google calls. It is disabled by default and enabled per environment; Perplexity calls never route through it.
- DataForSEO — retrieves Google AI Overview and search results for the phrases and markets you track.
- Firecrawl — fetches web pages, both cited URLs during verification and your own site during a site crawl.
- Stripe — payment processing. Subscriptions are not currently sold, so in normal use no payment data is created; the integration exists in the application and is listed here for completeness.
[Aemorph to supply: the hosting region(s) each sub-processor runs in, and how international transfers of personal data are handled]
Cookies and tracking
Clerk sets session cookies so that you stay signed in. Those are required for the application to function.
Your light/dark theme choice is kept in your browser's local storage, not in a cookie, and is never sent to us.
There is no analytics, advertising or third-party tracking script in this application. If that changes, this section must change with it.
Retention, deletion and backups
We are not going to state a retention period we cannot honour. Deleting a brand cascades to its markets, phrases, runs, stored answers, mentions and citations in the primary database.
[Aemorph to supply: retention periods for each data category, the process and timeline for account deletion and data export, and the backup retention policy of the hosting providers]
Security
Reaching your data requires authentication. The application signs users in through Clerk; the programmatic API requires a per-user API key; and inbound webhooks are signature-verified. The public free audit, which needs no account, is rate-limited by hashed IP.
[Aemorph to supply: verified security statements — encryption at rest and in transit, internal access controls, penetration testing, incident response and breach-notification commitments. Nothing beyond the paragraph above has been verified, so nothing further is claimed here.]
Your rights and how to reach us
We have deliberately not asserted compliance with any specific privacy regime, or any certification, because no such assessment or certification has been completed.
[Aemorph to supply: the privacy regimes this service is being held to (for example Singapore's PDPA and/or the GDPR), the rights available to individuals under them, and the request process and response times]
[Aemorph to supply: data controller legal entity name, registered address, and the contact point for privacy requests (and a Data Protection Officer, if one is appointed)]
Changes to this document
This page is generated from the application source. When the sub-processor list or the stored-data list changes in the product, this page is expected to change in the same release.
[Aemorph to supply: how material changes will be communicated to existing users]