AI programming and your insurance: the questions to ask before you sign a client

Search for a straight answer on AI-assisted programming and professional indemnity cover and you will find almost nothing useful. A handful of insurer blog posts that circle the topic without landing on it, some forum threads that argue past each other, and a lot of confident-sounding claims from people who are not insurers.

So this piece does not give you a legal answer, because we are not qualified to give one and neither is anyone else writing about it. What it does is set out the questions worth asking, and the process that makes the answers less frightening.

Why the question is genuinely unsettled

Most professional indemnity policies were drafted long before a coach could generate a training block in minutes. The wording usually turns on whether you acted within your competence and followed a reasonable process. Nothing in that test mentions tools, and nothing in it excuses you for using a bad one carelessly.

That leaves a gap. Your insurer may treat AI drafting as ordinary tooling, in the same category as using software to write the plan up. Or they may treat the generated output as something you did not personally author, which is a different question about who is responsible for it. The two readings are a long way apart, and the only way to know which applies to you is to ask.

Two things make this urgent rather than academic. The first is that when FitBudd surveyed coaches in 2026, 54% named output safety and quality as their biggest AI concern, ahead of job loss. That is what the profession is worried about, and it is worth reading alongside what coaches actually do with AI before you decide where your own process sits. The second is that the low-touch, fully online, template-heavy end of the market is the part under real pressure, and that is precisely the end where programmes get sent with the least review.

The five questions to put to your insurer in writing

Email is better than a phone call here, because you want a written answer you can point at later. Ask them all in one message so the reply covers the whole picture.

Does my cover respond to a claim arising from a programme I produced with AI-assisted software, where I reviewed and approved the final plan before delivery? That is the central question and the one most likely to produce a hedged answer, which is itself informative.

Does the answer change if the plan was generated rather than adapted from my own templates? Some insurers distinguish between AI as a drafting aid and AI as the author. If yours does, you need to know where the line sits.

What documentation would you expect me to be able to produce if a claim were made? This is the most useful question of the five, because the answer tells you what to keep. Usually it is the intake, the plan as delivered, and evidence you reviewed it.

Does anything need to be disclosed to you at renewal? Tool use rarely needs declaring, but you want that confirmed rather than assumed.

Is there anything in my policy that restricts or excludes automated or generated content? Worth asking directly, because a clause like that is easy to miss in a document written for a world without this software. Keep the reply. If your insurer declines to answer in writing, treat that as a signal to shop around, and say so when you do.

What a defensible process looks like

Whatever your insurer says, the same practical question sits underneath it: could you show that a competent professional reviewed this plan before a client saw it? That is a process question, and it is answerable. The coaches in the strongest position are the ones who can produce four things.

A documented intake. Age, experience, goals, equipment, training days, and the health information that shaped the programme. If a client mentioned a bad knee, you want to be able to show it was recorded rather than remembered. Programming around an injury is a skill in its own right, and it starts with the intake.

A plan with a review trail. Not just the final document, but evidence of what changed and why. This is where hand-editing everything in a chat window becomes a liability: the original output and your edits end up in the same undifferentiated text, and you have lost the thread of your own judgement.

Named decisions on anything outside your scope. If a health condition or a rehabilitation protocol influenced the plan, that is a judgement you made as a professional, and it should be visible as yours.

And a delivery record. When the plan went, what the client was told, and what you said about how it was produced. That last part connects straight back to the disclosure question, which turns out to be the same question wearing different clothes.

Where the tool should be doing the work for you

Here is the part most coaches get backwards. They treat the review as an administrative chore to get through, then wonder why it does not feel defensible. The review is the professional act. It is the thing your insurance is actually covering. So the useful question is not "how do I avoid reviewing", it is "what makes the review fast enough that I actually do it properly every time".

In practice that means wanting the output to tell you where to look. A plan that arrives with a note saying an exercise was removed because it conflicted with the stated knee injury has done half your review for you, because it has pointed at the exact line where your judgement is needed. A plan that arrives as a clean document with no commentary has hidden that from you, and you now have to hunt for it.

That is the design decision we made with Forge AI, and it is worth being precise about what it is and is not. It is not a claim that the software makes a plan safe. It is that the system screens allergies at ingredient level, screens exercises against stated injuries and conditions during generation, validates the structure, runs an independent review pass over the finished plan, and removes anything contraindicated rather than substituting it quietly. Then it tells you what it did.

The judgement stays yours. That is where the defensible part lives: a professional reviewed it, and the record shows they did.

You can read how the safety pipeline works, or generate a free plan and look at what the review notes actually say before you decide whether you would want them on a client file.