AI disclosure

Hirecraft uses OpenAI models in four places, and nowhere else. Everything else in the product is ordinary code: reading job boards, filling form fields, uploading files, moving cards on the tracker.

• Reading your resume into a profile, once, when you first upload it.
• Scoring new postings against that profile, so the feed can be ordered.
• Rewriting your resume for a particular posting, which you then read as a diff.
• Drafting an answer to a screener question your answer bank does not already cover, and writing a cover letter when a form requires one.

No model presses submit. The decision to send an application is made by a person, every single time, and the engine will not click the button on a form you have not approved.

No model chooses which jobs to apply to. Scoring orders your feed; queueing an application is something you do, from the feed.

No model sees your password, your payment details, or any account you hold elsewhere. Hirecraft does not create accounts on employer sites and does not store credentials for them.

By default every application stops after the form is filled. You see a screenshot of the real page, every field with the value that will be sent, and a label saying where each value came from: your profile, your answer bank, a model, or a default the engine assumed.

Anything a model drafted is highlighted. You can edit any field before approving, and an application with a required field left empty cannot be approved at all, in the browser or from your phone.

A profile can be set to submit on its own. That does not skip any of the checks above: an application is sent without you only when every required question already has an answer from your profile or your answer bank, and the honesty check has not flagged the cover letter. Anything else is held for you to answer, exactly as it is today. The engine does not guess a fact about you in order to avoid stopping, whichever setting you choose.

Tailoring rewrites the bullets your resume already has. It works on the structured profile rather than on a blob of text, which is why every change can be shown to you individually with a switch beside it.

Honest mode is the default. These limits are checked in code after the model answers, not merely requested in the prompt, and a rewrite that breaks one is shown to you struck through and cannot be switched on:

• Honest mode may not introduce a number, percentage, or duration your original line did not contain.
• Neither mode may drop a number your original line did contain.
• A rewrite more than twice the length of the original, or an empty one, is refused.
• Changes that only move punctuation or capitalisation are deleted before you see them.

Aggressive mode reframes more freely toward the posting, and is still bound by the second and third rules, and still shown to you line by line before anything is sent. Off sends the PDF on your profile exactly as you uploaded it.

Job titles, employers, dates, degrees and certifications are never rewritten in any mode. The PDF that reaches the employer is rebuilt from the switches you left on, so the version you approved is the version that is uploaded.

Application forms ask the same questions constantly: notice period, work authorisation, sponsorship, salary expectation, why this company. Your answer bank is consulted first, and a saved answer is used as written.

Only when nothing in the bank fits is a model asked to draft one, and that answer arrives on the review screen marked as drafted, for you to correct or replace. Saving your edit back to the bank means the model is not asked again next time.

Answers to legally sensitive questions come from your profile, never from a model: citizenship, sponsorship, disability, veteran status and the voluntary EEO questions are filled with what you entered, or left blank.

The text of your profile and the text of the job posting, plus the specific questions a form is asking. That is what a rewrite or a drafted answer needs in order to be about you and about that role.

Your content is not used to train models. Hirecraft does not sell it, and does not share it with employers beyond what the application form itself receives.

Treat it like any other cloud tool: do not put something in your profile that you are not allowed to send to a third party.

An application submitted through Hirecraft is submitted by you. It carries your name, your contact details, your work history, and answers you approved. In that sense it is no different from a resume you edited in a word processor and a form you typed yourself.

Some employers ask directly whether AI was used in preparing an application, and some forms have a checkbox for it. Answer honestly. Hirecraft never answers that question for you: a question about AI use is surfaced on the review screen like any other, for you to fill in.

Automating submissions may also conflict with a particular company's terms. Hirecraft applies through public forms at a human pace, one company at a time, and does not attempt to bypass a CAPTCHA or a login. A form that requires an account is reported back to you as a failure rather than worked around.

Pressing approve means you have read what is on the screen and are content for it to be sent in your name. Language models make mistakes, and the guardrails above reduce that risk rather than removing it.

Read the diff. Read the drafted answers. If a line says something your history does not support, switch it off before you approve, and tell us about it so the guardrail can be tightened.

Every submitted application keeps a receipt: the confirmation page, the fields sent, and the documents uploaded. If you ever need to know exactly what an employer received, it is on the application's own page.

You can choose, on your profile, to apply with an address on a Hirecraft mail domain instead of your own. When you do, that address goes in the email box of every form the engine fills, and mail sent to it arrives at Hirecraft. The only thing read by code is a verification code that Greenhouse emails after the submit click: the engine picks it up and types it into the form, and the code is stored against that application. Everything else is forwarded, unread, to your real address with the sender set as Reply-To, and a copy is kept in your in-app inbox so you can see what arrived. No model sees any of this mail. The switch is off by default, and turning it off keeps forwarding what still comes.

How your data is stored, kept and deleted is covered in the privacy policy, and the agreement you are signing up to is in the terms.