PayslipIQ uses large language models — primarily Anthropic Claude — to read your payslip image and produce the line-by-line explanation. AI systems can and do make mistakes. This page lists the known limits, so you can decide where to trust us and where to escalate to payroll, HMRC or a qualified adviser.
What the AI does well
- Read printed numbers from a clear payslip photo. The uploader takes JPEG, PNG and WebP; it does not take PDFs, so screenshot a PDF payslip first.
- Apply public 2026/27 PAYE, NI, USC, PRSI, pension and student-loan rules.
- Spot common red flags: emergency tax codes, wrong student-loan plans, mis-stated NI categories.
- Translate jargon into plain English with worked examples.
Known limitations
- Image quality. Blurry photos, glare and partial scans reduce accuracy.
- Edge cases. Director-level annual NI, multi-employment deferral, IR35 umbrella, S2P contracted-out legacy and tapered annual allowance are out of scope.
- Hallucination risk. The model may confabulate. Every reply is parsed against a fixed schema and rejected outright if it does not match, so a malformed or half-invented structure never reaches you. We do not independently recompute the model’s arithmetic against the tax tables before showing it to you. That is the single biggest reason every figure here is educational guidance, and why an unexpected number is worth putting to payroll or HMRC rather than acting on.
- Tax-year drift. Default 2026/27.
- No HMRC access. We do not see your coding notice (P2), your record, or any RTI submission.
How to escalate
- Sign in to your HMRC personal tax account.
- Email corrections@payslipiq.co.uk — see correction policy.
- Speak to payroll for per-period questions; an accountant for cross-year strategy.