Trust

The AI can't put words
in your mouth

Every resume tool promises not to lie for you. Almost none of them can tell you what stops it. This page explains the mechanism, what it covers, and where it stops — because a trust claim you can't inspect isn't worth much.

The mechanism

FactGuard rejects the output, it doesn't just ask nicely

Prompt instructions like "don't invent anything" are requests. Models comply most of the time, which is exactly the failure mode that gets you caught — the one resume in twenty with a company you never worked at.

JobeFlow treats that as a correctness problem instead. After the model produces a resume, a guardrail checks every employer and job title in it against your Career Profile. If something appears that you never entered, the generation is rejected rather than shown to you.

Your skills list gets the same treatment from the other direction. Requirements lifted from the job posting are dropped unless they map onto a skill, certification or project technology you actually recorded — so the posting's wish list can't quietly become your credentials.

What the check catches

Employer you never listedRejected
Job title you never heldRejected
Skill copied from the postingDropped
Your real role, reworded for this jobAllowed
Your real achievement, led with impactAllowed

Tailoring means choosing what to emphasise from what's true. That's the line the guardrail draws.

Grounding, everywhere

The same rule applies wherever AI speaks for you

Candidate match

Every strength cites the requirement it answers and the line from your profile that supports it. You can check each one before repeating it.

Career assistant

Answers come from your retrieved records — which resume went where, what you noted — not from the model's general knowledge.

Interview coaching

Tips coach you on using your real experience. Ones referencing material absent from your profile are rejected before you hear them.

Thank-you notes

Drafted from the interview that actually happened and the questions you prepared. With nothing on record, it stays general rather than inventing a conversation.

Your data

Plain answers about what happens to your material

We don't train on your job search

Your profile, resumes, notes and interview transcripts are sent to the AI provider only to produce the thing you asked for. They aren't sold, and they aren't used to train models.

Sharing is off by default

A shared workspace link is something you create deliberately, and it can be revoked or set to expire. Notes, salary details and recruiter contacts are excluded unless you explicitly include them.

Interview audio isn't archived

Spoken answers are transcribed so you can read and correct what was heard. The audio itself isn't stored — only the transcript you can see.

Ask for your data, or its deletion

Email us and we'll send you a copy of everything on your account, or delete it outright, within 30 days. Self-serve export and delete buttons are on the way; until then a person handles it. Cancelling a plan drops you to the free tier rather than locking you out of your own history.

Straight answers

What JobeFlow won't claim

Marketing pages usually stop before this part. Since the entire pitch is "the AI stays honest", it would be a strange place to start overselling.

We're in private beta

JobeFlow is new. There are no user-count milestones to quote and no testimonials, because inventing either would contradict the whole premise. What we can show is the product and the mechanism.

AI fit verdicts are triage, not scores

When Discover ranks a posting "strong" or "possible", that's a judgment call that can flip between runs near the boundary. Use it to decide what to read, not what you're worth.

Guardrails cover content, not every nuance

FactGuard checks employers and titles against your profile and rebuilds your skills from real ones. It can't judge whether a bullet overstates your part in a team's win — that's still your call to make honestly.

An ATS score is a proxy

Our keyword coverage is deterministic and repeatable, which makes it useful for comparing your own drafts. No tool can tell you how a specific employer's configuration will actually rank you.

Why put this on a sales page? Because you're about to trust a tool with the document that decides whether you get interviewed. A vendor willing to name its own limits is giving you the information you need to judge the rest of its claims.

See it on a real application

Join the private beta and put the guardrail to work on your next job.