Task statement 1.1 · Prompting and task execution
1.1 · Create Effective Prompts
In Wrenfield Housing, the course example, the communications team asks Claude for letters and reports that go to tenants. When a prompt, a message asking Claude for work, leaves out who will read the result and what it must say, Claude fills the gaps with guesses, and someone rewrites the draft by hand.
share
The tempting habit is to send a short request and fix the draft afterwards, because typing a longer prompt feels slower than editing. Claude knows nothing about your organisation, your readers or your facts beyond what the conversation gives it, so a short request makes it guess each of them. The edits then take longer than the missing sentences would have, and a guessed fact can slip through to the reader. The opposite habit fails too, because a prompt is not better for being long, only for answering the questions the task depends on.
KnowledgeWhat Claude knows when it reads your prompt
A prompt is the message you send Claude: the request, plus everything you give it to work with. Claude works from that message, the conversation before it and any file you attach. It cannot see your inbox, your organisation's habits or the facts in your head, so it fills every gap with a likely guess. Anthropic's prompting guide asks you to treat Claude as a brilliant but new employee who lacks context on your norms and workflows. Its test for a prompt is to imagine a colleague with minimal context on the task following it: if they would be confused, Claude will be too. This rules out any answer that expects Claude to work out a missing fact, reader or standard from a vague request.
An effective prompt therefore answers, in plain sentences, the questions a new colleague would ask before starting. Anthropic's guidance and help articles point to the same parts, which you can add in this order:
- The task: what to produce, in one sentence.
- The purpose and audience: who will read the result, what it is for, and the reason behind any instruction.
- The facts and inputs: every name, date, figure or document the result depends on, typed in or attached.
- The limits: tone, length, and what to leave out.
- The format: the shape of the result, with an example when the shape is hard to describe.
Interactive · what Claude must decide as the prompt grows
Step through the boiler letter's prompt, from one line to the full brief, and see what Claude is left to decide.
One line: Claude knows only that a letter about boiler servicing is wanted. It guesses the readers, the dates and the tone, so the draft is generic, full of blanks, and promises a £20 credit nobody offered.
Purpose and audience added: Claude now writes to Millbrook households and asks them to let the contractor in. The dates, slots and contact are still guesses, so the blanks stay.
Facts added: the dates, time slots and repairs line replace every blank. Tone, length and what the letter may offer are still Claude's choice.
The full brief: every decision the letter depends on is in the prompt, so the draft needs checking, not rewriting.
Two habits make these parts work harder. The first is giving the reason behind an instruction, not only the rule. Anthropic's prompting guide uses the example of an instruction never to use ellipses: once the prompt adds that the text will be read aloud by a text-to-speech engine, Claude understands the goal and can apply it to cases the rule never named. The second is keeping material apart from instructions. When you paste a long document, wrapping it in tags such as <document> and </document> tells Claude where the material ends and your request begins. Numbered steps help in the same way when the order of the work matters.
KnowledgeBusiness tasks and technical tasks
The exam guide asks for effective prompts for both business and technical tasks. The parts stay the same, but a different part carries the weight. A business task, such as a letter, a summary or a briefing, has no single right answer, so it depends most on the audience, the purpose and the tone. A technical task, such as a spreadsheet formula, a cleaned-up table or instructions for a tool, has a result that either works or fails, so it depends most on exact inputs and the exact shape of the result. Associates do not write code, but they do ask Claude for formulas and step-by-step instructions, so both kinds are part of daily work:
Business task
The weight is on purpose, audience and tone: who reads it, what they should do, and how it should sound.
Left out: Claude writes for an average reader in a neutral voice.
Technical task
The weight is on exact inputs and output: the program in use, column names, a few sample rows and the expected result.
Left out: Claude guesses the layout, and the answer fails on your real data.
SkillWorked example: the boiler letter that needed rewriting
Wrenfield's communications team had to tell the 1,200 households on the Millbrook estate that a contractor would service each home's boiler between 3 and 14 March. The first prompt was one line. The draft came back formal and generic, with blanks where the facts should be and a promise Wrenfield never made:
# before Write a letter to tenants about boiler servicing. → "on [DATE] … call [CONTACT NUMBER] … a £20 credit for any missed visit" # after Write a one-page letter to the 1,200 households on our Millbrook estate. Purpose: each tenant must let our contractor in to service their boiler. Facts: visits run 3–14 March, mornings 8:00–13:00 or afternoons 13:00–18:00. Tenants who cannot be in rebook through the repairs line. Plain English, friendly but firm. Offer nothing that is not listed here. → every fact filled in, no blanks, no credit offered
The cause was missing context, not a weak model. The one-line prompt gave Claude the task and nothing else, so it wrote for a neutral reader, left blanks for facts it could not know, and added a credit that sounded plausible. The new prompt adds the purpose and audience, the facts and the limits, and its last line closes the gap the credit came through. The communications lead still checks the letter before it goes out, as Wrenfield's AI policy requires, but the check is now a read-through rather than a rewrite.
Longer is not the goal. Anthropic's prompting advice warns that longer, more complex prompts are not always better. Start simple, add only detail that answers a question the task depends on, and check that each addition improves the result.
- Claude knows only what the conversation gives it, so every missing fact, reader or limit becomes a guess.
- An effective prompt gives the task, the purpose and audience, the facts, the limits and the format, with the reason behind each instruction.
- Business tasks lean on audience and tone; technical tasks lean on exact inputs and the exact shape of the result.