How the Free OET Writing Check works.

The grading pipeline, end to end

1. 01

   You paste your letter

   Referral, discharge, transfer, or advice letter — the same letter types you'd see on the real OET Writing sub-test. The text never leaves your browser session except to be analysed; we don't store it after the response is returned.

2. 02

   We anchor the prompt to the six OET criteria

   The AI is briefed with the same rubric used by official OET examiners: Purpose (Band 0–3), Content, Conciseness & Clarity, Genre & Style, Organisation & Layout, and Language (each Band 0–7). See our explainer for what each criterion means.

3. 03

   The model returns a strict-schema response

   An estimated band, a one-paragraph summary, and exactly six specific issues — one per criterion — that an examiner would likely flag. The response is JSON-validated before we render it, so you never see a half-rendered or malformed report.

4. 04

   The first issue gets a worked correction

   To keep the free tier sustainable, only the first of the six errors comes with a fully-rewritten correction. The other five are diagnostic prompts — they tell you where to look. For the full corrected version of every issue (and a hand-graded band from a qualified assessor), upgrade to the human service.

What the AI is good at

• Surface grammar, vocabulary and register — things a careful proofreader would catch on a first pass. Articles, tense agreement, formality, medical terminology accuracy.
• Structural critique — whether your letter has the standard scaffolding (salutation, paragraphing, sign-off), whether the purpose is announced early, whether the request is buried.
• Concision — flagging padding phrases, nominalisations, and sentences that delay the verb.
• Speed — a usable first read in about 60 seconds, free, with no appointment.

What the AI is not good at

• —Final band prediction with examiner-level confidence. The estimated band is an approximation. Real examiners calibrate against thousands of letters and disagree with each other by ±0.5 bands — AI sits in a similar noise band.
• —Clinical judgement. The AI checks that your letter is well-written; it does not verify that your clinical reasoning matches best practice. That's not the OET's job either, but a human assessor with healthcare background will spot it.
• —Cultural and tonal nuance. Whether your letter sounds like a colleague writing to a peer (as opposed to a student writing an exam) is something experienced assessors detect quickly and AI sometimes misses.
• —Catching every error. The free tier surfaces the six most load-bearing issues — not an exhaustive correction. For an exhaustive correction with every line marked up, that's the human service.

Free AI check vs. human correction

The two tiers are complementary: use the free AI check daily to iterate; use the human service before a real attempt at the OET.

When to upgrade to human review

Common upgrade moments:

• You've used the free check 10+ times and want a calibrated baseline.
• You're consistently scoring Band 5–6 estimated and need to find the missing half-band.
• You've been told your letters “sound right but lose marks” and you want to know why.
• You need a corrected reference letter to model future attempts on.

Privacy & data handling

• Your letter is sent to the AI provider once for analysis and is not stored after the response is returned.
• We retain the first 200 characters of your letter, your email, and your estimated band — to enforce the daily rate limit and contact you if you upgrade. That's it.
• No tracking pixels, no remarketing tags, no third-party scripts other than the Fungies checkout SDK (loaded lazily).