HOW WE USE AI · UPDATED 23 SEPTEMBER 2026
How we use AI
Written to be read rather than to be survivable in court, on the reasoning that a document nobody understood is the one that fails in court.
On Record uses AI models to interview you, learn how you write, and draft in your voice. This page says which models, what they see, what they never do, and who decides what goes out under your name.
Which models, and for what
- Anthropic’s Claude models do the thinking work: asking interview questions, turning your answers into stories and a voice profile, planning, drafting, and checking a draft against your own words before you see it.
- OpenAI transcribes audio when you speak rather than type, gives the interviewer a speaking voice in a spoken interview, and draws an illustration from a description you typed. If Anthropic is unavailable, a request can fall back to an OpenAI model so a job does not fail halfway.
No other AI provider receives your content. Every model call goes through one part of our code, which records each one and what it cost, so there is one place to check and nowhere for a stray call to hide.
If you speak an answer in Chrome or Edge, your browser may transcribe it itself, using Google’s or Microsoft’s speech service under their terms. That is the browser rather than us, and typing avoids it. The privacy policy says more.
What they are sent
The material the task needs, and no more: your interview answers, the stories and voice profile built from them, your pillars and plan, the draft being worked on, and the rules that apply to your voice. Card details never reach us, so they never reach a model. Credentials for connected accounts are held encrypted and are not part of any request.
Training
We do not have a model of our own, and we do not train or fine-tune any model on your content.
The providers say, of the data sent through their APIs:
- Anthropic: “By default, we will not use your inputs or outputs from our commercial products (e.g. Claude for Work, Anthropic API …) to train our models.” Inputs and outputs are deleted from their systems within 30 days, longer only if a request is flagged under their usage policy or the law requires it. Anthropic’s page.
- OpenAI: “data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us).” They keep abuse monitoring logs for up to 30 days, and none for audio transcription. OpenAI’s page.
Those are the providers’ words, checked on 23 September 2026, and they can change them. If either does in a way that touches your content, this page will say so.
A person approves everything that goes out
Nothing is published to a platform and no email is sent until a person has approved it: you, or somebody you have given that permission to. A scheduled post goes out at the time you picked, and only because it was approved first. Every approval is recorded against the person who gave it.
The model suggests; it does not decide. Where your rules, or your company’s, flag something in a draft, the fix is proposed to you and never applied silently.
Your voice is yours
Your voice profile, stories and interviews belong to you, permanently. A company or an agency can pay for your account, work on your drafts with your permission and stop paying, and none of that makes the voice theirs. You can export it — interviews, profile and all — whenever you like.
The profile learns from your own edits. Edits made by other people only teach it if you have said you trust them to, and a material change to how the profile describes you waits for you to confirm it.
Guidance, never copying
A draft is built from your material: your stories, your own sentences, your pillars. If there is not enough to write from, you get a better interview question, not a post with invented specifics.
If you name writers you admire, drafting borrows the shape of how they write — short lines, a story before the point — and never their words, and never names them in your post. What we learn from studying how people in roles like yours post shapes the questions we ask and the formats we suggest. It never puts anybody else’s words into your drafts, and your own writing always wins where the two disagree.
Deleting it
Close your account and your content is deleted at the first daily sweep (02:00 UTC) after 30 days; the app names the exact moment. A post you already published stays on the platform you published it to, and only you can take it down there. What the model providers hold goes on their schedule above. The privacy policy has the detail, including what we keep because the law requires it.
What this page is not
It is a description of how the product works, not a certification. On Record has not been audited against SOC 2, ISO 27001 or any other standard. If your organisation needs answers for a security review, write to legal@onrecord.tech and we will answer what we can, plainly, including what we do not have yet.