Privacy policy
Last updated 31 October 2026
BeigeCheck is run by Ghostart Ltd in the United Kingdom. We keep what the service needs to work, for as long as your company has chosen, and we don’t sell any of it. The AI provider that scores drafts isn’t allowed to train its models on them. The free BeigeCheck on our homepage keeps nothing you paste.
Who we are
BeigeCheck is a trading brand of Ghostart Ltd, a company registered in England and Wales under company number 16720663, with its registered office at 64 Cambridge Avenue, New Malden, KT3 4LF. In this policy “we” and “us” mean Ghostart Ltd. For questions about your data, write to hello@beigecheck.com.
Whose data it is
A company that uses BeigeCheck decides what goes into its workspace: its facts, its voice, its approved writing and the drafts its people check. For that material the company is the data controller and we process it on its behalf, as its processor. We’ll sign a data processing agreement with any customer who asks for one.
For the details we need to run accounts, such as your sign-in, your email address and what you use, we’re the controller.
What we collect
From you and your company:
- Your account: your name, your email address and how you sign in.
- Your company’s workspace: its name, website, the facts and house rules it approves, its voice brief, and every change to them with who made it and when.
- Documents your company adds. We read each one for facts to propose. Writing the company marks as approved is kept as a sample of its voice until an owner removes it; the text of other documents isn’t kept.
- Drafts sent for checking, and the scores and suggested changes for each, kept as described under “How long we keep it”.
- Your company’s website, when an owner or approver asks BeigeCheck to read it: up to ten public pages, read like a document. The facts and words it finds are proposed for approval, and the homepage and About page are kept as samples of how your company writes.
- Invitations: the email address an owner invites, and whether the invitation was accepted.
- A weekly email to each person, sent only in a week when something was approved at their company, listing what changed. Every one has a link to stop it, and we record when it was sent and whether you’ve stopped it.
- Your own voice, if you set one up: the pieces of your own writing you add, your answers to the questions on the Your voice page, and the voice brief drafted from them. You can remove any piece whenever you like.
Created as you use the service:
- Which AI tool you connected (Claude, ChatGPT or Microsoft Copilot) and when you last used it.
- How much of your company’s allowance each check and document used, and what the AI processing cost us to run.
- Technical logs from our hosting provider, such as the time of a request and whether it succeeded. We don’t log the text of drafts.
The free BeigeCheck on our homepage checks what you paste in memory and sends the result straight back. It doesn’t need an account, and it stores nothing.
What an AI tool sends us
When you connect BeigeCheck to Claude, ChatGPT or Microsoft Copilot, the AI tool calls BeigeCheck while it works for you. Each call carries only what that step needs: the kind of writing, a draft to check, or a question about your company’s facts, together with your sign-in. We don’t receive the rest of your conversation, and we can’t read your other chats.
Your conversation itself is held by the company that makes the AI tool, under the terms you or your employer have with them. Anthropic, OpenAI and Microsoft don’t process data for us when you use their apps.
Microsoft 365 Copilot and your administrator
A company adds BeigeCheck to Microsoft 365 Copilot through its administrator. When the administrator connects BeigeCheck, Microsoft’s notice asks them to confirm that the company, as data controller, lets Copilot retrieve data from BeigeCheck. That matches how we work: BeigeCheck answers Copilot with the company’s own facts, voice and check results, and processes the drafts Copilot sends as the company’s processor.
What Copilot sends us, where we process it and how long we keep it are the same as for every AI tool, as set out in this policy. Copilot’s connection to BeigeCheck only reads: it can’t change your company’s facts, voice or rules.
Folders you connect: SharePoint, Google Drive and Notion
An owner or approver can connect a SharePoint or Google Drive folder, or Notion pages, the company already keeps. They sign in with Microsoft, Google or Notion and choose the folder, or in Notion the pages. BeigeCheck’s access is read-only and limited to what that person’s account can open. It reads the Word, PDF and text files and Google Docs in the chosen folder and the folders inside it, or the chosen Notion pages and the pages inside them, and nothing else in the account.
We read each document the way we read one you upload: to propose facts and house rules for an approver, using our AI model provider as set out under AI processing. We keep each file’s name and version, so we read it again only when it changes, and the facts and rules it gave, with the file’s name as their source. We keep a document’s text only when the company has said the folder’s writing should also teach its voice.
Microsoft, Google or Notion gives us a key that keeps the folder readable. We store it encrypted, and only our servers use it. Disconnecting the folder in BeigeCheck deletes the key, and you can also remove BeigeCheck’s access in your Microsoft, Google or Notion account at any time.
BeigeCheck’s use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. We use Google Drive data only to provide these features to the company that connected it. We don’t use it for advertising, we don’t sell it, we don’t use it to train AI models, and people see it only with your permission, to keep BeigeCheck secure, or when the law requires it.
How we use it
- To provide BeigeCheck: answering your AI tool with your company’s facts and voice, checking drafts, reading documents, and preparing changes for an approver.
- To run your account: signing you in, sending invitations and service emails, keeping to your company’s allowance, and answering support requests.
- To keep BeigeCheck working and secure, including spotting misuse.
- To improve our checks, using drafts only from companies whose setting allows it (see below), and only in de-identified, aggregated form.
We don’t sell, rent or trade personal data, and we don’t use it for advertising.
Your company chooses what happens to drafts
Each workspace has one of three settings for the drafts its people check:
- None. The draft text is never stored. We keep the scores and the kind of change suggested, without any of the draft’s words. This is the setting for enterprise customers unless they choose otherwise.
- Aggregate. Draft text and the sentences we flagged are kept for 90 days, and may be used, de-identified and combined with other companies’ drafts, to improve BeigeCheck’s checks. This is the setting for companies that sign up themselves.
- Full. Draft text is kept for 365 days and used to learn your own company’s voice.
A company’s owner can ask us to change its setting or its retention period at any time. A daily job clears draft text, and anything quoting it, once its period ends.
Where draft text is kept, and you ask your AI tool to check your own edited version of a draft it wrote for you, we compare the two to spot changes people in your company keep making. A change made often enough is proposed to an approver as a word to use or a house rule. With the None setting, edits aren’t compared.
AI processing
Some checks, and reading documents, use an AI model from Anthropic. We send it the draft or document and the company knowledge it needs, through Anthropic’s business service, whose terms don’t allow Anthropic to train its models on what we send. The rule-based checks run on our own servers and send nothing to anyone.
Who processes data for us
We use these providers to run BeigeCheck, each under a contract that protects your data:
- Supabase: our database, in Ireland.
- Vercel: hosting, with our servers in London.
- WorkOS: sign-in, in the United States.
- Anthropic: the AI model behind some checks and document reading, in the United States.
- Mailgun: email, in the European Union.
Beyond these providers, we share personal data only when the law requires it, if the business is sold (and we’ll tell you first), or with your consent.
Data outside the UK
WorkOS and Anthropic process data in the United States. Where data leaves the UK or the European Economic Area, we rely on recognised safeguards: the UK International Data Transfer Agreement or Addendum, or the EU Standard Contractual Clauses.
Who can see what
- People in your company’s workspace see its approved facts, voice and rules. Owners and approvers also see proposed changes and decide on them.
- Owners see who’s in the workspace, which AI tools they’ve connected, and how much of the allowance has been used.
- Your own voice is yours: only you see your pieces and your voice brief, and only you approve it. Owners and approvers can’t see them. Your AI tool uses your voice brief for writing that goes out in your name.
- Each person’s connection to BeigeCheck is their own sign-in, and one company’s workspace is never visible to another.
- A small number of our own staff can see accounts, companies and usage to support customers and run the service. They see draft text only when a customer asks for help with a specific check.
How long we keep it
- Draft text: as your company’s setting says, then it’s cleared (never stored, 90 days, or 365 days).
- A draft you send for sign-off: kept while it waits for an approver, whatever your company’s setting, because they need to read it. Once it’s decided, it follows the setting above.
- Scores and the kind of each suggested change: while your company’s account exists.
- Approved facts, voice and rules, with their history: while your company’s account exists.
- Your account: while you have one.
If you or your company close an account, we delete or anonymise the personal data within 30 days, except what the law requires us to keep, such as invoices.
Security
Data is encrypted in transit and at rest. Our database separates every company’s data from every other company’s by row-level security, and the connection your AI tool uses can only read. If a breach affects your data, we’ll tell you, and the regulator where the law requires it.
Cookies
When you sign in we set one cookie that keeps you signed in. We don’t use analytics, advertising or tracking cookies, so there’s nothing to accept or decline. The free BeigeCheck sets no cookies at all.
Your rights
Under UK and EU data protection law you can ask to see the personal data we hold about you, correct it, delete it, take a copy, or object to or restrict how we use it. Write to hello@beigecheck.com and we’ll reply within one month. If your request is about a company’s workspace, we’ll work with that company, as it’s the controller.
You can also complain to the Information Commissioner’s Office (ico.org.uk) or your local data protection authority.
Our legal basis
- Contract: providing BeigeCheck to you and your company.
- Legitimate interests: service emails, security, preventing misuse, and improving our checks with de-identified data where a company’s setting allows it.
- Legal obligation: tax, accounting and lawful requests.
Children
BeigeCheck is a tool for businesses and isn’t meant for anyone under 18. We don’t knowingly collect children’s data.
Changes to this policy
We’ll update this policy when BeigeCheck changes, and email account holders about any change that matters. The date at the top shows the current version.
Contact
Ghostart Ltd, 64 Cambridge Avenue, New Malden, KT3 4LF, United Kingdom. Privacy questions and requests: hello@beigecheck.com.