A padlock on a bright blue door
Privacy

What we hold, who can see it, and what never touches a model.

This is written from the system rather than from a template. Everything below was checked against the code before it was written. Where the honest answer is awkward, it is here anyway — a privacy statement you cannot rely on is worse than no statement at all.

Last changed 28 October 2026
At a glance

Three kinds of information, and what reads each one.

Nothing you merely give us — as opposed to an account you deliberately connect — ever goes near a model.

We never sell your data. Not to anyone, not ever, at any price.

Fig. 1 — Where your information goesThe statement governs
What you type or uploadOur database What is public about youAn AI assistantA person reads itYour report An account you connectClaude, via HumanDataA person owns itYou approve Stays there. No model reads it. Only if you choose to
A summary of the statement below
WhatDoes a model read it?Who answers for itYour control
What you type or upload hereYour questionnaire, your files, your CRM NoYour answers are written to a Postgres database and they stay there. Inside HumanData, the people working your file can see it. The database refuses to return a record to anyone who is not on that list. Stop, delete, or take it with you.
What we can read about youYour website, your public filings, published papers, public program databases, and the notes we have written ourselves YesWe research and draft your funding report with the help of an AI assistant. A person reads and is answerable for every report before it reaches you. And you can switch even that off.
An account you connectYour Google, or your LinkedIn Yes, and you choose itYour connected data is read by a language model — Claude, operated by HumanData. A person owns every decision and reads every message before it goes anywhere. It is read-only and revocable in one click.
Each sentence here is taken from the statement that follows. Read the full answer on AI and what happens when you connect an account.
The question people actually ask

Is my information fed to an AI, or used to train one?

No, and nothing you give us can be.

Nothing in this product calls a language model. Not the questionnaire, not the file store, not the report, not the CRM. Your answers are written to a Postgres database and they stay there.

For everything you type or upload here — your questionnaire, your files, your CRM — the only outside services involved are Resend, which delivers our email, Stripe, if you pay us, and Supabase, which hosts the database and the files. None of that is sent to a language model, and none of it is or could be used to train anything.

There is one exception, and you choose it. If you connect an account — your Google, or your LinkedIn — so our agents can work from it, that connected data is read by a model (Claude, via HumanData) to do the work you asked for. You opt into it explicitly, it is read-only, and it is covered in full in its own section below: When you connect an account. Nothing you merely give us — as opposed to an account you deliberately connect — ever goes near a model.

Even the questions are not generated. If your questionnaire has questions on it that somebody else's does not, those were selected by pattern-matching rules held in the database, and each one records which rule put it there and why. Ask us why a question is on your page and there is a literal answer rather than a guess.

Where a model is involved, and it is not nothing.

We research and draft your funding report with the help of an AI assistant, the way a firm might use any other tool. What goes into that work is what we can read about you: your website, your public filings, published papers, public program databases, and the notes we have written ourselves.

That is a different activity from running your data through a model, and we would rather name both than give you a clean answer that is only half true. A person reads and is answerable for every report before it reaches you.

And you can switch even that off.

Tell us in writing and we will record it against your file, not in somebody's sent folder: nothing you submit goes near a model at all, and we work only from what you have sent us. There is a field on your record for exactly this, and it is visible to everyone who works your file.

Email james@humandata.id and say so. You do not have to give a reason.

When you connect an account

What we read from your Google or LinkedIn, and what we never do with it.

Read-only, and only because you asked.

Connecting Google gives us read-only access to your Gmail, Drive & Docs, Contacts and Calendar. Connecting LinkedIn confirms your identity and — if you import your connections — lets us read that list. We use all of it for one purpose: to find the past performance, documents and relationships that strengthen your funding, surface the warm paths to funders and prime contractors, and draft the outreach for you.

This is the one place a model reads your private data — and it is Claude, via HumanData.

To do that work, your connected data is read by a language model — Claude, operated by HumanData — and a HumanData operator is accountable for what it produces. This is the single exception to the promise above: it happens only for an account you deliberately connect, only for the funding work you asked for, and a person owns every decision and reads every message before it goes anywhere.

What we never do.

We never sell your data. Not to anyone, not ever, at any price. We never use it to train a model. We never share it with anyone we have not named to you. We never use it for advertising. And we never send a message on your behalf without your approval — the agent drafts, you decide.

We work for you alone.

Your data is never shared with our other clients. Every engagement is siloed to its own account — your past performance, your contacts and your pipeline stay with you, and are never pooled, cross-referenced, or reused to help anyone else we work with. We hold each client under a fiduciary duty to act in that client’s interest alone, and we keep what we learn about you confidential to you — consultant–client privilege, one client at a time.

You are in control.

It is read-only and revocable in one click — disconnect from your confirmation page, or reply to any email and say so. When you disconnect we stop reading, and we delete what we extracted on request. Not connecting an account changes nothing else about your account or your report.

What we hold

Four things, and we would rather you knew all four.

Two numbered letterboxes set into a blue wall

What you told us. Your organisation or your name, an email address, a website, whatever you typed when you first came to us, and your answers to the questionnaire — typed or recorded. Recordings are audio of your own voice, answering our questions, and nothing else.

What you sent us. Documents you upload: financials, prior applications, determination letters, whatever evidence a funder would want.

What we found. Research we did about you from public sources, and the funding programs we matched you to. This is our work, and it is on your file.

What happened. A log of what we did and when — report sent, link opened, interview started, email delivered. We do not run analytics, advertising trackers or third-party scripts on these pages, and we do not set cookies on the public site. Signing in to your account sets a session, which is what keeps you signed in.

Who can see it

Us, you, and nobody we have not named.

A brass handle and a keyhole on a blue wooden door

We do not sell your information, and we do not share it with funders, lenders or anybody else without asking you first. When the moment comes to put your name in front of a funder, that is a conversation, not a setting.

Inside HumanData, the people working your file can see it. The database refuses to return a record to anyone who is not on that list — the rule is enforced by the database itself rather than by the page you are looking at, which means a mistake in the interface cannot leak someone else's file.

If you talk instead of typing. The questionnaire lets you dictate your answers. That uses your browser's own speech recognition, which in Chrome and Safari sends the audio to Google or Apple to be turned into text — the same thing that happens when you dictate a message anywhere else on your device. We never receive the audio and we never store it. What reaches us is the text in the box, which you can edit or delete before it saves. Typing involves no third party at all, and it is the whole of the difference.

One thing worth knowing about file links. Documents you upload sit in private storage and cannot be read without a signed, expiring link. Interview recordings made before we switched to dictation are held the same way; we no longer make them. Finished artifacts we produce for you — a capability statement, a research profile — are published to a web address that is long and unguessable but not password-protected, so that you can forward it to a program officer without them needing an account. If you would rather yours were not, say so and we will move it.

Your funding report works the same way: a long private link, which you can share and which we can revoke. We can see how many times it has been opened. We cannot see who opened it.

Your say

Stop, delete, or take it with you.

Stop emailing me. Say so and we stop. There is a switch on your record for it, and when it is on the system cannot email you even by accident — the block sits under the thing that sends, not beside it.

Delete it. Ask and we delete your record, your answers, your recordings, your documents and your report. Deleting requires someone to type your organisation's name back, because a confirmation you can click through is not a confirmation. One exception, and it is a legal one: if you have paid us, the payment record survives, because a receipt with no customer on it is a worse problem than a record you wanted gone. Everything else goes.

Take it with you. Ask and we will send you what we hold, including the research we did. It is yours; you can take it to somebody else.

We keep what we hold for as long as we are working with you, and after that for as long as it is useful to you — funding cycles run two to three years and a file we deleted in month nine is one we cannot pick back up in month thirty. Ask us to delete it sooner and we will.

One address, and a person reads it.

james@humandata.id. Tell us what you want and we will do it, or tell you honestly why we cannot. If something on this page turns out not to match what the system actually does, that is a defect and we want to know.