I asked AI to draft my employment documents. Here’s what it missed.

At SafeHR, we regularly advise employers who are navigating complex workplace issues, from drafting employment contracts to managing dismissals and defending tribunal claims.

Increasingly, we’re seeing businesses turn to AI to draft employment documentation.

AI is fast.
AI is structured.
AI sounds confident.

But when it comes to employment law, confidence and compliance are not the same thing.

To explore this properly, I tested AI by asking it to:

• Draft a full-time electrician’s employment contract.
• Advise on dismissing an underperforming site manager.
• Create a harassment prevention policy.
• Outline the process for dismissing due to gross misconduct.

At first glance, the responses looked polished and professional.

But once you review them against the requirements of the Employment Rights Act 1996 and the principles set out in the ACAS Code of Practice, the gaps become clear.

And some of those gaps create real legal risk.

Here’s what AI got wrong, and why it matters for employers.

1. The “basic” employment contract that wasn’t compliant

The electrician’s contract looked well-structured, clear headings, logical clauses and professional tone.

However, it failed to include several statutory particulars required under section 1 of the Employment Rights Act 1996, including:

• The date continuous employment began.
• Details of pensions and pension schemes.
• Disciplinary and grievance procedures.
• Holiday year specifics.
• Training entitlement.
• Collective agreements (or confirmation that none apply).

These are statutory requirements, not optional enhancements.

If an employer fails to provide compliant written particulars and the employee later succeeds in a substantive tribunal claim, compensation may be increased.

Beyond statutory omissions, there were significant commercial gaps:

• No deductions from wages clause (for overpayments, damage, or unreturned equipment).
• No lay-off or short-time working clause (particularly relevant in construction).
• No right to work conditionality clause.
• No clause addressing loss of professional certification.
• No clear holiday pay calculation method.
• Vague wording around overtime and sick pay.

The document looked professional.

But it did not properly allocate risk or protect the employer.

That distinction is critical.

2. Dismissing an underperforming site manager: What was missing

The AI response on dismissing an underperforming site manager was logically structured and broadly aligned with a capability process.

However, it omitted important legal risk considerations.

It referenced the two-year qualifying period for unfair dismissal, but did not alert employers to automatic unfair dismissal risks, such as:

• Health and safety complaints.
• Whistleblowing (protected disclosures).
• Asserting statutory rights.

These claims do not require two years’ service.

It also did not reference the potential 25% uplift in compensation where an employer fails to follow the ACAS Code of Practice.

Nor did it address discrimination risk overlays, for example, where performance concerns may relate to disability or other protected characteristics.

In practice, these are often the issues that determine whether a dismissal is defensible.

3. The harassment policy that looked good, but lacked protection

The harassment prevention policy was well-written and clearly structured.

But it was generic.

It did not:

• Reference the Equality Act 2010.
• Anchor definitions to protected characteristics.
• Set investigation timeframes.
• Provide a clear appeal route.
• Address reporting concerns involving senior management.
• Refer to training obligations.
• Clarify confidentiality boundaries.

Policies are not simply internal guidance documents.

They become evidence in tribunal proceedings.

A vague or incomplete policy can weaken an employer’s ability to rely on a “reasonable steps” defence in discrimination claims.

4. Gross misconduct: Process is everything

The AI guidance on dismissing for gross misconduct followed a sensible sequence:

• Investigation
• Suspension (if necessary)
• Invitation to disciplinary hearing
• Right to be accompanied
• Written outcome
• Appeal

Structurally, this aligns with best practice.

However, it lacked emphasis on:

• Consistency with previous sanctions.
• The importance of maintaining trust and confidence during suspension.
• The evidential standard under the “reasonable belief” test.
• The risks of procedural unfairness even where conduct is serious.

Tribunals do not simply assess what happened.

They assess whether the employer acted reasonably in all the circumstances.

That level of nuance rarely appears in generic AI drafting.

The broader issue: false confidence

AI does not typically produce obviously flawed HR documents.

It produces documents that look credible.

• Clear formatting
• Professional tone
• Logical structure

For many employers, particularly SMEs, that can create a false sense of security.

However, employment documentation must do more than read well. It must withstand legal scrutiny.

That requires risk assessment, sector awareness, and a detailed understanding of how tribunals interpret both the Employment Rights Act 1996 and the ACAS Code in practice.

SafeHR’s view

At SafeHR, we recognise that AI can be a useful drafting tool.

It can help with structure, initial frameworks, and generating starting points.

But it should not replace informed legal review, particularly where:

• Contracts allocate commercial risk.
• Dismissals are contemplated.
• Policies may be relied upon in litigation.
• Sector-specific protections are required.

The cost of reviewing and strengthening documentation at the outset is modest; the cost of defending a tribunal claim arising from inadequate drafting rarely is.

If you are reviewing your employment contracts, policies, or disciplinary processes, particularly in light of increased use of AI tools, our employment law team would be happy to advise.