Garbage In, Garbage Out: Why Input Quality Still Matters in Allied Health AI

AI in allied health is getting better very quickly. The models are smarter, the notes are cleaner, and the reports sound more professional. But one old principle still applies: garbage in = garbage out.

In fact, as AI becomes more powerful, this principle may matter even more.

The risk is not just hallucination anymore. The bigger risk is polished inaccuracy. This happens when AI takes incomplete, vague or poorly captured information and turns it into something that sounds professional, structured and clinically convincing.

The note looks good. The formatting looks good. The language sounds clinical. But the underlying information may still be weak.

Even the smartest AI cannot reliably know what was not said, not recorded or not documented. It cannot know the red flag that was not asked about, the objective finding that was not mentioned, the clinical reasoning that stayed in the practitioner’s head, or the key detail missed because of poor audio.

AI can organise information, improve clarity, structure notes and reduce admin burden. But it cannot magically create reliable clinical truth from poor input.

This matters because clinical notes are not just admin. They are part of the clinical record. They support continuity of care, communication with other providers, funding requirements, audit trails, insurer reporting and, at times, medico-legal review.

A beautiful note is not automatically a safe note. A detailed note is not automatically a useful note. A fast note is not automatically a good note.

For allied health practitioners, AI should be treated as a documentation assistant, not a replacement for clinical reasoning. The practitioner still matters. The consultation still matters. The assessment still matters. The review process still matters.

One of the uncomfortable truths about AI documentation is that it can expose the quality of the consultation. If the clinical reasoning is clear, the assessment is structured, and the practitioner communicates well, AI can help produce a much stronger record.

But if the consultation is vague, the questioning is incomplete, or the clinician does not verbalise their reasoning, the AI may have very little to work with. In some cases, this is useful because it shows where clinical documentation can improve. In other cases, it is risky because AI may produce something that sounds impressive while hiding the weakness of the original input behind professional language.

That is why human review remains essential.

Improving AI output is not only about choosing the smartest model. It is about improving the whole workflow: clear audio, structured consultations, verbalised clinical reasoning, relevant objective findings, good templates, review habits and team training.

Whether you are using ChatGPT, Claude, Gemini or a purpose-built allied health AI scribe, the same principle applies. Give the AI clear context. Do not just say, “write a note.” Instead, be specific about the discipline, appointment type, funding body, required structure and what the AI should not invent.

For example: “Write a physiotherapy SOAP note for an initial WorkCover shoulder injury consultation. Use only the transcript. Separate subjective findings, objective findings, assessment, plan, capacity, goals and follow-up. Do not invent findings that were not stated.”

Another useful prompt is: “What clinically important information is missing from this note?”

Or: “Separate documented facts from assumptions. If something was not stated, mark it as not documented.”

These simple instructions can make a major difference.

AI can save time, improve consistency, reduce documentation burden and help practitioners communicate more clearly with GPs, insurers, referrers and patients. But AI is not magic. It is not a mind reader. It does not remove professional responsibility. It does not replace clinical judgement.

The clinics that get the most value from AI will not simply be the ones using the most advanced model. They will be the clinics with stronger clinical standards, clearer workflows and better review habits.

The future of AI in allied health will not just be about better models. It will be about better implementation, better training, better templates, better review habits and better clinical standards.

The goal is not simply to produce more words faster. The goal is to produce safer, clearer and more useful clinical records.

That is why garbage in = garbage out is not outdated. It is more relevant than ever.

Because in healthcare, good input creates good output.

Poor input creates polished clinical risk.