You've opened an AI-generated draft and something feels off. The phrasing isn't quite how you'd put it. This is one of the most common reactions we hear, especially during onboarding and your first few reports. Most of the time it's normal, expected, and fixable in a few minutes. This article explains what's happening and gives you a quick way to handle it.
Start here: two different feelings that feel identical
When a draft feels "not like me," it's almost always one of two things. They feel the same in the moment, but they have completely different fixes, so the first job is to tell them apart.
What you're actually reacting to | What it means | What to do |
"This is wrong." A finding, score, fact, or tone that doesn't match the client or the data. | A genuine error. Not a voice issue. | Correct it and, if it's a real defect, tell us (see Something's genuinely wrong below). |
"This is accurate… but it doesn't feel like me." Nothing is incorrect. It just doesn't sound the way you picture your own writing. | The normal recoil. Your brain is comparing the draft to an idealized version of your voice. | Run the Voice Check below, then edit (don't rewrite). It converges. |
If you're not sure which one you're having, the Voice Check sorts it out.
The Voice Check ✅
Before you rewrite anything, ask three questions. These are the tests a psychological report actually has to pass and "sounds like me" was never one of them.
Is it accurate? Do the findings, scores, and formulations match the data and the client?
Is it defensible? Would it hold up in a file review, to a referral source, or in court?
Would you sign it? Not "did it flow from my fingers" but are you willing to put your name behind this document as it stands?
If yes to all three: the draft is doing its job. Edit for polish, sign, and move on. If no to any: you've found a specific, fixable thing. Change that not the whole document.
Why an accurate draft can still feel wrong
Short version: you're not comparing the draft to how you actually write. You're comparing it to how you picture yourself writing and those were never the same.
It's the same reason a recording of your own voice sounds like a stranger, or a candid photo makes you wince. The version in your head is smoother and fresher than the real record. Pull ten of your own finalized reports and read them back to back and you'll find your own patterns. The same openings, the same hedges, the same transitions you'd swear you invented each time. Your real writing is more consistent and more templated than the version you remember. That's not a flaw; it's what writing under a caseload looks like.
So when the AI returns your actual patterns and your gut says that's not me, it usually is. Your internal reference is just set higher than your own history.
Want the full science (voice confrontation, the mirror-vs-camera face, the frozen face effect)? See our blog post: That Doesn't Sound Like Me.
How to make drafts sound more like you
Unlike your recorded voice, this one is trainable and it gets closer every time you correct it. Three habits do most of the work:
Edit, don't rewrite. Change what's genuinely wrong; leave what merely feels unfamiliar. Rewriting a good draft top-to-bottom teaches the model nothing and burns the time you were saving.
Correct in place. Accept or reject the AI's suggested snippets and fix wording directly in the draft. Your edits are the training signal.
Tune your template, not just the text. A lot of "voice" lives in structure aka your intro style, section order, and standard phrasing. Set these once in your report template so every future draft starts closer to you.
Give it a few reports. The match improves noticeably as the model learns from your corrections, and you'll spend less time editing each time.
Something's genuinely wrong
If a draft is inaccurate such as wrong finding, wrong score, wrong facts, or a tone that misrepresents the client that's not a voice issue and you shouldn't work around it. Fix it in the draft, and if it's a real defect, let us know so we can improve your model:
The more specific you are ("it invented a recommendation," "it flattened a borderline score into a definite one"), the faster we can tune it.
One honest note
Even after the model is dialed in, you may still occasionally catch a faint "huh... is that how I write?" in a finished draft. That trace doesn't fully disappear, for the same reason you never quite make peace with your own voicemail. It's not a sign the tool got it wrong. It's just the normal gap between the writer you imagine and the writer on the page and every report you sign is still reviewed and approved by you before it goes anywhere.
Remember: PsychAssist.ai drafts; you approve. Nothing leaves without your signature. Judge each draft by whether it's accurate, defensible, and something you'd sign and let the voice keep tuning itself in the background.
Still stuck, or convinced a draft is genuinely off? Please reach out.