By Alberto Alatorre ·
How do I stop staff pasting client data into chatbots?
This is a training problem first and a software problem second. A rule plus a safe alternative plus practice is what actually stops the next paste.
Staff paste client data into chatbots because it is fast, not because they are careless. A client record goes in, a clean summary comes out, and the work gets done. Telling people to stop will not work unless you give them something just as fast to do instead. A block alone moves the habit somewhere else.
How big is the problem?
The numbers we have come from large companies, not small businesses, so read them as a warning sign, not as your rate. A LayerX enterprise security report from 2025, reported by Akamai in October 2025, found that 45 percent of enterprise employees already use generative Ai tools, and 77 percent of those users paste data into prompts. Most of that pasting, 82 percent, comes from personal, unmanaged accounts, at an average of 14 pastes a day, with at least 3 containing sensitive data.
The cost side is also enterprise data. IBM's Cost of a Data Breach Report 2025, summarized by vendor blog CurrentWare, found that 20 percent of data breaches involved shadow Ai, with about $670,000 in extra containment cost on average. It describes large companies. A small business will not face the same dollar figure, but the shape of the risk is the same: data you cannot see, in accounts you do not control.
No small business paste rate has been verified. You do not need one. If your team uses chatbots at all, assume the pasting is happening and fix the habit.
What actually stops it?
Four things, in this order.
First, name the rule in plain words. Write down what never goes into a chatbot: names with account details, financial information, health information, contracts, and anything a client would not want read by strangers. One page, not a policy document. When in doubt, leave it out.
Second, give people a safe alternative that is just as fast. A business plan for the chatbot, where data is excluded from training by default. A template with placeholders instead of real client details. A teammate to check with when the job needs real data. If the safe path takes ten minutes longer than the risky one, people will take the risky one.
Third, practice with real work. Show the team a real example: a client email with the names and numbers swapped for placeholders, then the same job done right. People learn the habit by doing it once, not by reading a rule once.
Fourth, make someone the owner. One person on the team who answers the question, can I paste this, and who checks in once a month. Without an owner, the rule fades in a few weeks.
What about blocking the tools?
A block has a place, but it is not the fix. Blocking the chatbot on work computers stops one path. It does not stop a personal phone, a personal account, or a different tool. Most pasting in the enterprise studies came from personal, unmanaged accounts. A block pushes the habit onto a phone you cannot see.
Use a block to buy time while the training lands, not as the whole answer. The goal is a team that knows the rule and has a faster safe way to do the job, not a team that works around you.
What to do next
Start with the one page rule and one practice session. Pick the three most common jobs where pasting happens, and work through each one with the safe alternative in the room.
That is the kind of training we do: hands-on work with real workflows, plus the guardrails and habits a team keeps using after training ends. The Build step of our Ai upskilling work is exactly that, guardrails that survive after the session is over. For an overview of our team training services, visit the Ai upskilling page. It describes our approach, rather than providing the rules, templates, or follow-up plan discussed here.