Show AI What Good Work Looks Like

You ask AI to write a professional update. It gives you three paragraphs of polished language, a formal greeting and a closing you would never say. You shorten it, remove the grand phrases and rewrite the opening. Eventually, it sounds like your business again.

For a task like this, try giving AI a small example of work you already consider good. Then explain what makes that example useful. You are giving it something concrete to follow: the level of detail, the order of information and the way you speak to your reader.

This can be useful for project updates, product descriptions, short reports and other writing you repeat. The goal is to make your standards easier to communicate, while keeping the new work accurate and recognisably yours.

Choose one task and one strong sample

Start with a task you actually expect to do again. “Help with marketing” is too broad for this exercise. “Turn my project notes into a short client update” gives you something you can try and review.

Choose a short sample written for a similar reader and purpose. A friendly social caption may be a poor guide for an internal incident report. An old proposal might contain useful structure but outdated offers. You should be able to say why the sample deserves to be followed.

Use writing you own or have permission to use. Public business copy or a fictional sample is a sensible starting point. Removing a customer's name may still leave an identifiable situation, confidential terms or private details. If you cannot safely share the material with the AI service you are using, make a fictional version that demonstrates the same writing pattern.

Explain what the example gets right

“Write like this” leaves room for the assistant to copy the wrong thing. Add a few observations in everyday language.

For a short project update, your instructions might be:

• Open with the most important change since the last update.
• Use plain words and a calm, helpful tone.
• Finish with one specific next step.

Those observations become a small, reusable standard. You can also tell a colleague what matters without expecting them to guess your preferences.

Anthropic's prompting guidance recommends relevant, varied examples to help steer Claude's format, tone and structure. The exercise here starts with one sample to keep the first attempt manageable. Add more when a second example teaches a genuinely different situation.

A fictional example: the three-sentence project update

Imagine a small design studio preparing an update for a client. These are invented details, used only to demonstrate the method.

The owner likes that it gives the status first, keeps each sentence short and ends with a clear request. The owner wants those qualities in the next update.

Now the task changes. The new facts are: a website homepage layout is ready; the service descriptions are still needed; the client should send the approved descriptions; no launch date has been agreed.

The pattern transfers. The brochure, printer and Tuesday deadline stay behind. That distinction matters whenever your examples include prices, names, dates or promises that belong to an earlier job.

Keep the example separate from today's facts

Label the parts of your request clearly. One section holds the reference sample. Another says which qualities to follow. A third contains the facts for the new task.

Try this prompt, replacing the bracketed sections:

You can include a short preference such as “avoid formal greetings” or “use the words our customers normally use”. Keep the instructions focused on this task. A long list of preferences from unrelated work can make the standard harder to apply.

Test whether the pattern travels

Read the new draft beside the reference. Ask three questions:

• Did it preserve the useful qualities? Check the opening, level of detail and next step.
• Did it use only the new facts? Look for a stray date, copied offer or promise from the reference.
• Does it fit this reader? A tidy structure is still wrong if the tone or explanation misses the person's needs.

If it falls short, give one concrete correction. “Put the request in the final sentence” is easier to act on than “make it better”. Keep the corrected draft if it becomes a useful second example.

Try another ordinary case before turning this into your default process. For example, test an update with no new progress as well as one with a completed milestone. A single strong result does not show how well the instructions handle both.

Build a small reference you will actually reuse

Save the sample, the reasons it works and the reusable prompt together. Add a date and a note about the task it suits. When your services or preferred language change, review the examples too. An old sample should not quietly become the source of a new business promise.

Keep reviewing what AI produces. Examples can guide the work, but the responsibility for accuracy, judgment and what you publish remains with you.

For your next task, choose one short piece of work you are proud of and write down three qualities worth keeping. You already make decisions about what good work looks like. This exercise helps you put those decisions into words.