Asking an AI assistant about your business: what to trust
AI chat assistants answer business questions in seconds. What the security agencies warn about, which answers to trust, and why the assistant must respect who is asking.
"Who owes me money?" "What should I reorder before Friday?" "How did September compare with August?" A business owner used to pay an accountant or wait for a report to answer these. An AI assistant answers in seconds. The useful question for 2026 is not whether to use one but which answers to trust and what never to type into a public chatbot.
Two different kinds of assistant
A public chatbot has read the internet and can write a supplier letter or explain a tax term. It has not read your sales. Anything it says about your business is either a guess or a repetition of what you typed in. A business assistant is connected to your own records: it looks up the figure, then explains it. The first kind is a writing tool; the second is a reporting tool. Confusing the two is how owners end up making decisions on invented numbers.
What the security agencies say
The UK National Cyber Security Centre's note on large language models is blunt: they "can get things wrong and hallucinate incorrect facts", they "can be biased, are often gullible", and they are "prone to injection attacks". On privacy, it corrects a common fear and raises a real one. A public assistant does not automatically learn your question for other users to see, but "the query will be visible to the organisation providing the LLM", and stored queries "will almost certainly be used for developing the LLM service or model at some point". Its recommendations are "not to include sensitive information in queries to public LLMs" and "not to submit queries to public LLMs that would lead to issues were they made public".
For a shop that means: never paste a customer list, a payroll or a bank statement into a public chatbot. Ask it to explain withholding tax; do not ask it to compute yours from your data.
The US National Institute of Standards and Technology publishes a voluntary AI Risk Management Framework, released in January 2023. Its purpose is "to better manage risks to individuals, organizations, and society associated with artificial intelligence". The framework is written for large organisations, but its first lesson scales down: know what the system is for, know what data it touches, and have someone responsible for it.
Three tests for an answer
- Where did the number come from? A trustworthy business assistant shows the records behind the answer, or at least names the report. If it cannot, treat the figure as a draft.
- Who is asking? An assistant connected to business data must answer within the permissions of the person typing. A cashier should get today's sales count, not the gross margin. If the assistant answers everyone the same way, it has quietly become the widest-open door in the business.
- What did it cost? Assistants consume capacity, often counted in tokens. A meter under each answer keeps the cost visible and stops a curious afternoon becoming a surprise bill.
Questions worth asking first
The best early questions are the ones you would otherwise pay for or put off:
- Which customers have balances over thirty days, and how much in total?
- Which items sold out this month, and which have not moved since July?
- Which supplier's deliveries were late most often this quarter?
- How did each branch do on Saturday compared with the previous four Saturdays?
Each has a precise answer in the records. Each is tedious to assemble by hand. None requires the assistant to invent anything.
From the counter to the taxi
Kit AI answers in every Kit web page, in Kit Business on Android and iOS, and on the counter machine through the Kit AI desktop app. The desktop download page currently lists version 1.19.22, dated 26 September 2026, for Windows, macOS and Linux, with scheduled reports under Settings and, since 1.19.19, an Activity view that shows the assistant's file reads and record checks as they happen. Seeing the steps is the point: an answer you can check is an answer you can act on.
A house rule for staff
Write one line and pin it beside the till: business data goes only into the business system's assistant, under your own login; public chatbots are for writing and general questions. Allow people to experiment, as the NCSC suggests, "in a way that doesn't place organisational data at risk". The assistant then does what it is good at, and the records stay where they belong.
Sources
Run the whole business from one login.
Point of sale, stock, CRM, accounting free in every plan, payroll and Kit AI. Start on the web today and add the till, the phone app and the desktop app as you grow.
No card needed · 14-day trial