AI as a witness: The future of documenting verbal agreements and manager-employee conversations

Every HR professional has been in this situation.

A manager and an employee have a conversation in which the manager promises a salary review in six months. The employee agrees to take on additional responsibilities in the meantime. But this conversation is not documented.

Six months later, the salary review has not happened. The employee raises the concern, but the manager is unable to recall making that commitment. The employee is certain they did. HR is now in the middle of a dispute where both sides are telling the truth as they remember it, and there is no record of what was actually said.

This situation arises in almost all organisations. 

Verbal agreements between managers and employees are one of the most persistent and most costly sources of workplace conflict. They generate grievances, fuel distrust, and consume HR’s time that should be going elsewhere. And the reason they keep happening is that the way organisations have always documented conversations, when they document them at all, is fundamentally insufficient for the modern workplace.

AI changes this by providing the documentation infrastructure that ensures what was said does not disappear the moment the meeting ends. This blog examines the problem that undocumented conversations create, how AI is addressing it, and what HR must understand to get this right.

The cost of undocumented conversations

Most organisations do not track the cost of undocumented verbal agreements. But it accumulates in ways that are both significant and measurable once HR starts looking for them.

• Grievances that could have been prevented: A significant proportion of formal employee grievances originate in a misremembered or disputed conversation. For instance, a commitment that was made and not kept or an expectation that was set in a one-to-one and never confirmed in writing. Both of these situations are preventable. Each of them consumes HR time and damages the employment relationship.
• Performance management disputes: Verbal performance conversations are one of the most common sources of contested dismissal claims. For instance, a manager has a series of undocumented conversations with the underperforming employee, where they discuss the issues, set expectations and agree on certain targets. But the employee's performance does not improve. When the HR team moves towards dismissal, the employee disputes that no such conversation took place. This places HR teams in the difficult position of trying to prove that an undocumented conversation took place.
• Broken trust between managers and employees: When an employee believes their manager made a promise and then denied it, the damage to trust is severe and rarely fully repaired. The employment relationship continues, but it continues on a different foundation. And that foundation shapes every subsequent interaction between the two people, and between the employee and the organisation more broadly.
• Inconsistent application of policy: When managers make verbal commitments that vary from the organisation's written policy, they create informal exceptions that HR did not sanction and cannot track. One manager tells an employee they can work from home four days a week. Another tells their team the policy is two days. Neither conversation is documented. HR ends up managing a workforce operating under individual verbal arrangements.

How does AI documentation solve this issue?

AI-powered conversation documentation tools work by transcribing, summarising, and structuring the content of manager-employee conversations in real time or immediately after they occur.

• Accurate transcription: AI transcription tools produce a verbatim record of what was said in a conversation. This eliminates the reliance on memory and removes the possibility of a dispute about what words were actually used. A manager who committed to a salary review "before the end of Q2" cannot later claim they said "sometime next year" if the transcript records the original words accurately.
• Intelligent summarisation: A full transcript of every manager-employee conversation is not always what HR needs. AI tools can produce structured summaries that extract the key commitments, action items, and agreed outcomes from a conversation, presenting them in a format that is immediately useful for follow-up and record-keeping without requiring HR or the manager to read through an entire transcript.
• Automated follow-up generation: Some AI documentation tools can generate draft follow-up communications automatically, based on the content of the conversation. A manager who agrees to three specific actions in a one-to-one receives a draft email summarising those actions, ready to send to the employee with minimal editing. This closes the gap between what was discussed and what was confirmed, without adding meaningfully to the manager's workload.
• Integration with HR systems: AI documentation tools that integrate with existing HR platforms allow conversation records to be attached directly to the relevant employee file, the relevant performance cycle, or the relevant case record. This means that when HR needs to reconstruct the history of a manager-employee relationship, the documentation is in one place and structured in a way that makes it immediately useful.

The legal dimension: What does AI documentation protect?

HR professionals operate in a legal environment in which the quality of documentation is frequently the difference between a defensible employment decision and an indefensible one. AI-powered conversation documentation has specific implications for this environment that HR must understand clearly.

• Creates a contemporary record: In employment disputes, a contemporary record, one created at the time of the relevant event rather than reconstructed after the fact, carries significantly more weight than a record created in response to a claim. An AI-generated transcript of a performance conversation, created immediately after that conversation, is a contemporary record. A manager's written recollection of what they discussed three months ago is not. This distinction matters enormously in formal proceedings.
• Removes the credibility battle: Many employment disputes come down to a credibility contest: the manager says one thing, the employee says another, and it is up to the HR teams to determine which account to believe. AI documentation removes this contest from a significant proportion of cases, because the record of what was said exists independently of either party's recollection. This protects both the employer and the employee.
• Creates accountability for both parties: HR professionals sometimes assume that AI documentation primarily protects the organisation against employee claims. In practice, it protects employees equally. For instance, a manager who knows that conversations are being documented is less likely to make informal commitments they do not intend to keep, less likely to deny commitments they did make, and more likely to conduct difficult conversations with the care and consistency that the employment relationship requires.
• Must comply with applicable privacy law: The legal benefits of AI conversation documentation are real. So are the legal risks if the documentation is implemented without regard to applicable privacy legislation. In most jurisdictions, recording a conversation requires the informed consent of the parties to it. HR must ensure that any AI documentation programme is implemented with a clear consent framework, transparent communication to employees about what is being recorded and how it is being used, and governance structures that comply with the relevant data protection requirements.

The ethical dimension: When AI documentation becomes surveillance

There is a line between documentation that serves the employment relationship and documentation that surveils it. HR must know where that line is, and must design AI documentation programmes that stay on the right side of it.

The positive case for AI conversation documentation rests on its ability to protect both parties in a manager-employee relationship by creating an accurate, accessible record of what was agreed. This is genuinely valuable. It protects employees from managers who deny commitments. It protects managers from employees who misrepresent what was said. It protects HR from being asked to adjudicate disputes with no evidence. And it protects the organisation from legal liability that accurate documentation would have prevented.

The surveillance risk emerges when documentation moves beyond capturing agreements and outcomes and begins capturing every word of every interaction, hesitation, and informal comment. Employees who believe that every word they say to their manager is being recorded and potentially used against them do not have honest conversations with their managers. They suppress concerns and do not communicate their true feelings. The documentation infrastructure that was supposed to support the employment relationship ends up undermining it.

How can HR teams effectively implement AI documentation?

AI conversation documentation is a powerful tool. It is also one that requires careful, deliberate implementation to deliver its potential without producing the risks described above.

• Define the scope before deploying the technology: HR must be explicit about which conversations will be documented, how the documentation will be stored, who can access it, and under what circumstances it will be used. This scope definition should be completed before any technology is deployed, not after employees start asking questions about what is being recorded.
• Build consent into every process: Consent to AI documentation must be genuine and freely given. Employees who feel that consent is a condition of employment rather than a genuine choice are not consenting in any meaningful sense. HR must design consent processes that give employees real information about what they are agreeing to, and that do not create any penalties for declining.
• Train managers in how to use documentation outputs: The AI-generated summary of a performance conversation is a tool. It is not a substitute for the manager's judgment about how to use it. Managers must be trained to review documentation outputs for accuracy, to correct errors before records are finalised, and to use the follow-up materials the system generates in ways that strengthen rather than mechanise the relationship with their team member.
• Review and audit the programme regularly: AI documentation programmes must be subject to regular review. HR teams need to ask questions like: Is the documentation serving its intended purpose? Are employees experiencing it as protective or as intrusive? Is the consent process working as designed? Are the privacy and data governance commitments being upheld in practice?
• Design AI documentation programmes: Not every conversation needs to be documented. The programme should focus on the specific categories of conversation where documentation serves a clear purpose, such as performance discussions, commitment-setting conversations, disciplinary meetings, and formal one-to-ones. It should not extend to every casual interaction between a manager and a team member.

AI conversation documentation is an infrastructure that supports the human relationship at the centre of the employment contract. Getting the implementation right determines whether it serves that relationship or undermines it

Key Takeaways

• Undocumented verbal agreements between managers and employees are one of the most persistent and costly sources of workplace conflict. Grievances, performance management disputes, broken trust, and inconsistent policy application all trace back, regularly, to conversations that were never recorded.
• Memory is not a reliable substitute for documentation. When a manager and an employee remember the same conversation differently, both can be telling the truth as they experienced it.
• AI documentation tools solve this by creating an accurate, structured record of what was actually said. Verbatim transcription, intelligent summarisation, automated follow-up generation, and integration with existing HR systems together close the gap between what was discussed and what was confirmed, without adding significantly to the manager's workload.
• The legal value of AI documentation is specific and significant. A contemporary record created at the time of a conversation carries far more weight in formal proceedings than a written recollection produced after a claim has been raised. 
• AI documentation protects both parties equally. A manager who knows conversations are being documented is less likely to make informal commitments they do not intend to keep and less likely to deny commitments they did make. An employee has the same protection in reverse.
• There is a clear line between documentation and surveillance, and HR must know where it is. Capturing agreements and outcomes from formal conversations serves the employment relationship. Recording every word of every interaction undermines it. Employees who feel constantly monitored suppress honest communication, which is the opposite of what the documentation programme is designed to produce.
• Scope, consent, and governance must be defined before the technology is deployed. Which conversations will be documented, who can access the records, and under what circumstances they will be used are questions that must be answered before employees start asking them.
• Consent must be genuine. Employees who feel that agreeing to AI documentation is a condition of employment rather than a real choice are not consenting in any meaningful sense. HR must design consent processes that give employees genuine information and genuine options.
• AI documentation supports the manager's judgment. It does not replace it. Managers must be trained to review outputs for accuracy, correct errors before records are finalised, and use the tool in ways that strengthen the employment relationship rather than mechanise it.