What is missing before your AI writes nonsense to a customer?

Monday morning. The first customer request is on the desk. One colleague asks AI to draft a reply. Another enters the same prompt and gets something that sounds completely different. Both are sure they were clear. The problem is often what they left unsaid: Who is the customer? What tone fits? What can you promise? Team prompting becomes reliable when these assumptions are made explicit. But how much context is enough, without turning every request into a form?
The missing word is rarely in the prompt
A request such as “Write a friendly reply to this complaint” sounds clear. In practice, it often isn't. For a hotel, friendly might mean addressing the guest personally and offering a solution. For a wholesaler, a concise, matter-of-fact reply may work better. The AI can see the task, but it won't automatically know what your team expects.
This is especially clear with words that seem obvious internally: “brief”, “professional”, “as usual” or “for our most important customer”. They often stand for experience and unwritten rules. If you don't provide them, you leave the AI to guess what they mean.
Then the rework begins. You change a few sentences, add information and explain what you meant by “our tone”. If the same thing happens with similar tasks, ask a different question: what information is regularly missing before someone submits the prompt?
The same task isn't always the same task
Two colleagues can enter the same prompt and expect different results. One wants a finished email that can go straight to the customer. The other needs three wording options to choose from. Both write, “Reply to the enquiry in a friendly way.” The sentence means something different in each person's working context.
In a small business, this knowledge may sit in the owner's head. In a medium-sized firm, it may be spread across sales, customer service and specialist teams. Larger organisations have approvals, brand guidelines and different areas of responsibility to consider. Adding more people doesn't automatically make the context clearer.
A good team prompt doesn't have to sound elaborate. It needs to settle the decisions each person would otherwise make from scratch. What should the final result look like, who will read it, and how will you know it works for this purpose?
Give the AI a useful brief
Start by describing the task and the specific situation. “Draft a reply to an existing customer's question about our delivery time” gives more direction than “Write an email”. Then add what the AI needs to know about your offer, the customer and the conversation so far. Without those details, it can only work with what you have actually given it.
Next, set boundaries. Can the reply give a delivery date? Is a discount off the table? Should the AI ask for missing information rather than fill in the gap? These instructions won't prevent every unsuitable phrase, but they make clear which decisions remain yours.
Finally, specify the format you want, such as a short email with a subject line, greeting and next step. An example from your own work can also show what “friendly and direct” sounds like in your business. Make sure the example fits the task and contains no confidential details.
A team template should help, not slow people down
When everyone starts from scratch, personal variations soon build up. A short template can capture shared expectations without making every reply sound the same. Useful fields include the task, audience, relevant background, desired result and boundaries. Colleagues then add only the details that matter for the specific case.
Keep the template deliberately short. A long mandatory form gets skipped in a busy working day, especially when the phone is ringing or several customers are waiting for a reply. Ask the people who handle recurring enquiries: what do they most often have to add to the AI prompt? That's where to start improving it.
Try the template on real tasks. Does it work for an initial reply but not a complaint? You may need two versions. An example the team agrees works well is more useful than an abstract rule such as “always write precisely”. Which task would make a good first test?
What this means for your business
Pick a recurring task that regularly needs reworking. It might be a reply to a question about an appointment, a summary after a customer call or a draft quote. Use a few real examples to identify what the person handling each one knew before writing the prompt.
Turn that information into a team template and test it with the people who actually use the process. Don't judge the replies only by how well they read. Check whether they fit the customer, avoid unconfirmed promises and suggest the right next step.
Also decide who maintains the template and who checks the results before they are sent. For a sole trader, that might mean a short personal checklist. In a larger business, responsibility should sit with the people who know the relevant rules. That way, it's clear who makes the call when the AI lacks information.
Next Monday can start more calmly
Picture another customer enquiry landing on the desk on Monday morning. This time, your colleague opens a template with the right guardrails, adds the conversation history and checks the draft against agreed rules. They don't have to guess what “friendly” or “brief” means in your business. They still own the professional judgement.
For this to work, you need feedback from day-to-day use. When an answer regularly needs editing, note what was missing. Was the audience unclear? Did the AI make a promise it wasn't allowed to make? Update the template rather than quietly creating another personal version.
That makes team prompting a shared effort to build clearer processes. Not every task needs a detailed template, and not every reply can be used without changes. But when the same misunderstandings keep cropping up, sharing the right context can make a real difference. Which task is causing your team the most rounds of corrections today?
Frequently asked questions
What does team prompting mean?
Team prompting means agreeing how to give AI requests so colleagues can share important information and expectations. This includes the task's purpose, its audience, relevant background, the desired format and clear boundaries. That way, everyone doesn't have to explain the same rules from scratch. The team is still responsible for checking the answer.
How do you write a good prompt for work?
Describe the specific task and situation. Explain who the result is for, what information the AI needs and what it must not decide or promise. Specify the format you want and, if useful, provide a relevant example. Then check that the answer is accurate and fit for its intended purpose.
Why does the same prompt produce different answers?
A prompt often leaves important expectations unstated. Words such as “brief” or “professional” can mean different things to different colleagues. The information people have in their heads isn't automatically available to the AI either. The wording and context provided can therefore change the result. A shared template makes expectations clearer.
How do you create a prompt template for a team?
Start with a recurring task that often needs reworking. Note what information is needed for a good result and put it into a short template. Test it with the people who do the task, add any missing guidance and decide who will update it when processes change.
What information should you never put in an AI prompt?
Don't enter confidential or personal information into an AI tool unless your organisation has explicitly approved it for that use. Check your internal rules and use anonymised examples where needed. A good prompt needs relevant context, but that doesn't automatically mean including names, contact details or other sensitive information.
Interested?
Want to know what this means for your company?
This article was researched and written with AI assistance and automatically checked against the sources listed before publication. More on our AI transparency
Free guide · PDF
The AI Compass for small and mid-sized businesses
9 jobs your team can hand over by Monday.
Get it free
Klarframe AI Readiness Check
How ready is your company for AI?
About ten minutes, instant result with your stage, a 90-day plan, suitable solutions and notes on the EU AI Act.


