You Asked a General AI Tool to Help With Your Incident Investigation. Here's What You Actually Got.
You needed something fast. You opened a chat window, typed out what happened, and got back something that looked thorough. A summary. A list of probable causes. Maybe some recommendations. It was coherent. It sounded professional.
Then a few weeks later, your insurer asked a question you couldn't answer. Or a regulator requested documents you hadn't collected. And you started wondering: was that output actually enough?
That instinct is worth listening to.
What a General AI Tool Is Actually Doing
A general AI tool is designed to respond helpfully to any question. It draws on an enormous range of training data and generates text that is fluent, organised, and plausible. That is genuinely useful for a lot of tasks.
But when you describe a workplace incident in a chat window, that tool has no idea what a complete investigation looks like. It has no structured process for what evidence to collect. It doesn't know what your jurisdiction requires. It can't check whether your investigation is missing anything critical. And once the chat ends, everything is gone.
What you got was a well-written response to the description you provided. Not a structured investigation of what actually happened.
This article gives you the high-level picture of what was missing. Each article in this series then dives into one specific gap in detail.
Independent safety consultants use MyInvestigationCoach too — not just to generate investigation outputs, but to validate and review investigations more efficiently across multiple clients. The gaps described here apply equally whether you're investigating your own incident or reviewing someone else's.
The Four Gaps That Matter
Gap 1: It didn't know what evidence to collect
A structured workplace incident investigation follows a framework for evidence — typically called PEEPO, which stands for People, Environment, Equipment, Procedures, and Organisation. It is a way of ensuring you look beyond the obvious and collect evidence across every category that might be relevant.
A general AI tool doesn't apply this framework. It responds to what you described. If you didn't mention equipment maintenance records, it won't ask for them. If you didn't describe the training history of the people involved, it won't flag that as a gap. It reflects your description back at you in an organised way — it doesn't independently identify what's missing.
See the full breakdown in the companion article: A General AI Tool Doesn't Know What Evidence to Collect. Your Investigation Does.
Gap 2: It gave you one answer, not root cause analysis
Root cause analysis — identifying the underlying reasons an incident occurred, not just the immediate trigger — requires tracing multiple causal chains. A serious incident typically has several contributing threads: a systems failure, a supervision gap, a procedure that wasn't followed. These threads interact.
A general AI tool gives you one response to one prompt. It may sound like analysis, but it is a single pass at the description you provided. There is no mechanism for identifying multiple independent causal chains, cross-checking them against evidence, or testing whether the causes identified are genuinely grounded in what happened.
More on this in: One Prompt, One Answer. That's Not Root Cause Analysis.
Gap 3: It couldn't tell you whether your investigation was complete
One of the most important functions of a structured investigation process is quality assurance — checking whether the evidence collected actually supports the conclusions being drawn, and flagging where gaps remain.
A general AI tool has no mechanism for this. It responded to what you gave it. It has no way of knowing that a maintenance log was never collected, or that a witness statement covers only part of the timeline, or that a key procedure document is missing. It produced an output. Whether that output is supported by sufficient evidence is a question it cannot answer.
Detail in: A General AI Tool Can't Tell You If Your Investigation Is Complete. We Can.
Gap 4: It has no memory
If the incident developed over several days — if new evidence emerged, if witnesses came forward later, if you had follow-up questions — each new chat session started from zero. A general AI tool has no ongoing record of the investigation. Everything you'd already established had to be re-explained each time.
An investigation is a living record. It accumulates evidence, updates findings, and builds toward a documented conclusion. A chat window doesn't do any of that.
Read more: A Chat Window Has No Memory. Your Investigation Register Does.
What a Structured Investigation Actually Produced
Here's a worked example that runs through every article in this series.
A warehouse worker slips on spilled oil near a loading dock in a small manufacturing business and loses time from work. The owner types a description of the incident into a general AI tool and receives a coherent, professional-looking summary: probable cause (wet floor), recommended actions (clean spills immediately, install warning signs), and a short narrative of events.
It looks complete. Then the insurer calls and asks whether equipment in the loading dock area had current maintenance records. Whether the injured worker had completed manual handling training. Whether there was a documented procedure for spill management in that area. Whether any supervisor had flagged the floor condition before the incident.
The general AI tool's response mentioned none of these things — not because they weren't important, but because the owner's description hadn't included them. And because the tool has no framework for knowing what a complete investigation of a slip-and-fall incident requires.
What a PEEPO evidence checklist surfaces: The structured checklist generated for this incident type flags maintenance records, training documentation, and spill procedure records — across People, Equipment, and Procedures categories — none of which appeared in the general AI tool's response.
What Was Actually Missing
The general AI tool produced something that looked like an investigation. But an investigation that can't answer an insurer's questions — because key evidence was never identified and never collected — isn't a complete investigation. It's a document that held up fine until someone looked closely.
If a regulator or insurer asks to see your investigation and key evidence is missing, that gap works against you. Not because the AI tool gave you bad advice, but because it gave you a response to a chat prompt. That's a different thing.
Want to see how a structured investigation handles this? Start your free trial
In This Series
- A General AI Tool Doesn't Know What Evidence to Collect. Your Investigation Does.
- One Prompt, One Answer. That's Not Root Cause Analysis.
- A General AI Tool Can't Tell You If Your Investigation Is Complete. We Can.
- A Chat Window Has No Memory. Your Investigation Register Does.
- Generic AI Doesn't Know OSHA. Your Investigation Tool Does.