Somewhere in your business, right now, there is a reasonable chance someone has already opened ChatGPT, Copilot or Gemini to help with something at work. Not because you asked them to. Not because anyone rolled it out. Because it was faster than doing the thing the old way and faster than asking you.
This is not a guess dressed up as a trend. A survey of office workers at large companies in the UK, the US, Australia and Japan, published by PagerDuty in June 2026, found that 51 percent of UK respondents had put work correspondence into a public AI tool, against 43 percent across all four countries. Separate research commissioned by Microsoft UK the previous October found that 71 percent of UK employees had used AI tools their employer had not approved. Asked why, 28 percent said their company did not provide a work-approved option. Read those two findings together and the picture is not employees going rogue. It is employees solving a problem nobody at the top gave them a proper way to solve.
Worth being honest about where this evidence comes from. Most of what has been measured on this comes from larger organisations, the kind with enough staff and enough IT policy for a "shadow AI" problem to show up in a survey in the first place. Almost nobody has properly measured what is happening inside a business with ten, twenty or fifty people. That absence of data is not proof nothing is happening. If anything, it is the more interesting fact. In a smaller business, this entire question, who on the team is using what, for what and whether it is actually helping, tends to be genuinely invisible, not because it does not matter, but because nobody has ever thought to ask.
The two wrong assumptions
Ask most owners what their staff do with AI and the answer usually falls into one of two camps and both are usually wrong.
The first is "nobody here really uses that." Given the numbers above, this is very often untrue. Someone almost certainly has, even if only once, even if only to draft an awkward email.
The second is "everyone's probably fine, they'll sort themselves out." This is the more dangerous assumption, because it can be true and false in the same office at the same time. One person may have quietly found a genuinely good use for it, saving themselves real time every week and never mentioned it because nobody asked and it did not occur to them it was worth sharing. Another person, in the same room, may be pasting client details or pricing information into a free public tool with no idea that matters. Both of those things can be happening under one roof, unnoticed, right now.
Why this is worth more than curiosity
There is a genuine risk side to this, stated plainly and without drama. Information going into a public AI tool without anyone having thought about what is and is not appropriate to put there is a real, ordinary business risk, the same shape as any other data handling question, not a reason to panic, but a reason to have an actual position on it rather than an assumed one.
There is also a genuine upside sitting in the same gap. If someone on your team has already found a shortcut that works, that is free, ready made value that is currently benefiting one person instead of the whole business, purely because nobody has ever asked the question out loud.
What this actually sounds like in practice
None of this needs to be complicated to be useful. In practice it is usually a handful of short, informal conversations, not a workshop and not a form. Ask the person who does your invoicing whether they have tried using anything to help draft the awkward chasing messages. Ask whoever answers your enquiries whether they have ever pasted a tricky reply into a chatbot to get a faster first draft. Ask the person who writes your quotes whether there is a step in that process they have quietly found a shortcut for and never mentioned to anyone.
What tends to come back is a mixture. Somebody will admit to something small and slightly sheepish, using a public tool to tidy up a message, worried it might look like they were not doing the job properly rather than doing it smartly. Somebody else will describe a genuinely useful habit that has been sitting entirely inside their own week, helping nobody else, because sharing it never occurred to them and nobody ever asked. At least one person, in most businesses this size, will say they have not tried any of it and are not sure where they would even start. None of those three answers is a problem to fix. They are simply the actual picture, which nobody had before asking.
Why an outside number cannot tell you this
National surveys are useful for context and cannot tell you what is happening inside your own business. A figure showing most UK office workers have tried putting work correspondence into a public AI tool tells you something is common. It cannot tell you whether it is happening in your business, who is doing it, what they are putting into it or whether it is actually helping. Averages describe a country. They do not describe your team on a Tuesday.
That is the real argument for the conversation over the statistic. It is also the argument for having it now rather than waiting for a clearer national picture to emerge, because that picture, even once it arrives, will still be an average with your business sitting somewhere inside it, unidentified. There is a natural next step once the picture is clearer: working out what, if anything, needs a simple policy and what is worth sharing more widely because it is genuinely useful. That is a smaller, calmer piece of work once you know what you are actually looking at, rather than a guess dressed up as a plan. It is also where the training side of what we do usually begins, with the people who are already experimenting.
This is not an audit and it is not about anyone's job
To be clear about the frame here, because it matters: this is not about catching anyone out and it is not a step toward doing anyone's job with a machine instead. The businesses that get real value from AI adoption are not the ones that quietly replace people. They are the ones where the boring, repetitive layer gets lighter and the person is freed up for the part that was always the actual job, the judgement calls, the conversation with the customer, the thing a machine still cannot do. Finding out what is really happening with AI inside your business is a step toward that, not away from it.
What good actually looks like
A business that has done this well does not look dramatically different from the outside. Nobody has announced an AI strategy. There is no new dashboard on the wall. What has usually changed is smaller and more durable: the person who found a genuinely good shortcut has shown two colleagues, the person who was pasting sensitive information into a public tool now has a straightforward, sanctioned way to do the same job safely and the owner has stopped guessing and started knowing. That is a modest sounding outcome. It is also the actual foundation everything else gets built on, because you cannot sensibly decide what to invest in next until you know what is already happening.
The actual first step
You do not need a system, a policy document or a piece of software to start finding this out. You need a handful of honest conversations, with the people who actually do the work, about what they have tried, what helped, what did not and what they wish they had a straightforward way to do. That is a smaller ask than it sounds and it usually surfaces something useful within the first few conversations.
This is also, genuinely, the actual starting point for anyone thinking about where AI fits into their business at all, whatever sector you are in, whatever size you are. Before any tool, any spend, any training plan, there is this one question: what is actually happening right now, that nobody has looked at yet. If you have already run the bottleneck audit, these conversations sit comfortably alongside it.
That is also, honestly, the more interesting conversation to have than most people expect walking in. The usual pattern is a double surprise, once at what somebody had already quietly figured out and once at how quickly the awkward, sheepish admissions stop feeling awkward the moment it is clear nobody is being caught out.
If that sounds like your business, get in touch. This is not a form to fill in and it is not a sales page to read through. It is a conversation and it starts with a phone call.