Parsed, not stored as a blob
Employers, titles, dates, skills and the achievements under each role. A model handed six pages of unstructured text finds the wrong job as often as the right one.
CV context
Any model can explain how to handle a difficult handover. Only one that has read your CV can say you did it at a logistics firm in 2024, in three weeks, across four depots. This is how it comes to know that.
Reading a CV is free. Keep it current.
What happens
Employers, titles, dates, skills and the achievements under each role. A model handed six pages of unstructured text finds the wrong job as often as the right one.
Updating a CV adds a version. A session that ran against the old one still makes sense when you read it back six weeks later, which is the whole point of keeping history.
The CV attached to a session is the one it answers from, so two parallel job searches do not contaminate each other.
What is actually sent
This is the honest answer to “where does my CV go”, and it is also the reason answers are fast.
Your CV is trimmed to a set size before it joins the question in a prompt. The size does not depend on how long your CV is, so neither does the speed or the price of an answer.
That trimmed CV goes, with the question, to the AI provider writing the answer. There is no way to answer from your experience without sending it. The privacy page names the providers and what each may do with it.
A screenshot of a coding problem does not need your employment history. Context is assembled per question, so what is irrelevant is not sent.
Your CV is in your account to answer your questions. We do not train on it, and deleting it removes it.
Getting more out of it
All of them are about your CV rather than about the product.
A CV written for a recruiter says “improved reliability”. An answer built from that says the same, and it is weak. Add the figure — 40% to 9%, six weeks, four engineers — and every answer that touches that project gets better.
The pipeline nobody thinks about and the migration that finished quietly are exactly the material behavioural questions want. Cutting them for brevity removes the stories.
A CV has no room for the architecture, the trade-offs or the post-mortem. Those belong in documents, which are searched per question. Document context →
Questions
Extracts the text, then parses it into structure: employers, job titles, dates, skills and the achievements under each role. The structure is what matters — it is the difference between an answer that says “in a previous role” and one that names the company and the year.
No. It is trimmed to a fixed budget — a few thousand characters — before it reaches a prompt. That is what keeps the cost of an answer predictable and the first word arriving in seconds, and it is why one credit can mean one credit regardless of whether your CV is one page or six.
Yes. Keep as many as you need and attach one per session — a platform version and a backend version invite different questions and should produce different answers. Updating a CV adds a version rather than overwriting the old one, so a session's history still makes sense later.
PDF and Word, because that is what people actually have. Text is extracted from the file and then parsed; the original file is not what gets used.
No. Adding or updating a CV is free, which is deliberate: anything you are charged for is something you will hesitate over, and hesitating over keeping your CV current makes every answer worse.
It will phrase what is there confidently, which is usually what you want and occasionally more than you would have claimed yourself. It will not invent an employer, a title or a date that is not in the document. Read the answer rather than reciting it — that is the check, and it takes no longer than reciting would.
Any CV, from Settings, at any time. Deleting your account takes every CV, document, session and transcript with it.
Read next
That is the test. Reading it costs nothing, so it is a cheap one.
Create your account