Format is the shape and style of the answer, separate from the facts inside it.

A prompt is the message you send to an AI assistant. Inside that message there are usually two different kinds of requests. The first is about content: what you want to know or what you want done. The second is about format: how you want the answer to look and feel when it comes back. Many people write only the first kind and forget the second.

Format covers many small choices. Should the answer be one sentence or one page? Should it be a list, a table, a few paragraphs, or a set of numbered steps? Should the tone be friendly or formal? Should it be written for a child, a manager, or an expert? Should it follow a certain layout so that another program can read it? Each of these is a format decision, and each can be stated in plain words.

The assistant cannot see the place where your answer will land.

When you read an answer, you know where it is going. Perhaps it is for a slide, a message to your boss, a spreadsheet, or a school essay. The AI does not know this unless you say so. It sees only the words you typed. When details are missing, it fills the gaps with a sensible default guess, which often means a medium-length answer with a friendly tone and some headings or bullet points.

That default is fine for many casual questions. But it is only a guess about an average reader. Your situation is specific. You may need a very short reply because it will be read on a phone. You may need a plain paragraph because bullet points look odd in an email. You may need a strict layout because a tool will process the result. None of that can be guessed reliably.

flowchart LR
    A["Your prompt
with no format details"] --> B["AI fills the gaps
with a default guess"]
    B --> C["Answer in a
generic shape"]
    C --> D["You reshape it by hand"]
    E["Your prompt
with clear format details"] --> F["AI follows
your description"]
    F --> G["Answer ready to use"]
        

A format request is built from a few simple pieces. You can pick the ones you need.

You do not have to describe everything every time. Think of these parts as dials. Turn only the dials that matter for the task, and leave the rest to the assistant’s good judgment.

Knowing the names of common formats helps you ask for exactly what you want.

You do not need technical vocabulary. Ordinary words work well. Here are the shapes people most often request, with the kind of job each one fits best.

Length is the easiest format detail to state and one of the most useful.

Without guidance, an assistant may write far more than you need, or less than you hoped for. A length request removes that uncertainty. You can state length in several ways: a number of sentences, a number of paragraphs, a rough word count, or a comparison such as “short enough to read in thirty seconds.”

The skeleton of an answer decides how easy it is to read and reuse.

Some content naturally fits one structure. Steps in a process fit a numbered list, because order matters. A comparison of three products fits a table, because the same features are checked for each. A set of loosely related ideas fits bullet points. A story or a persuasive argument usually reads best as flowing paragraphs.

It is also fine to ask the assistant to avoid a structure. If you will paste the answer into an email, you might say, “Write in plain paragraphs, with no bullet points or headings.” Many assistants love lists by default, so saying what you do not want is just as valuable as saying what you do want.

The same facts feel very different depending on who is speaking and who is listening.

Tone and audience are quiet but powerful format controls. A description of electricity written for a ten-year-old uses simple words and everyday comparisons. The same topic written for an engineer uses precise vocabulary and skips the basics. If you do not say who the reader is, the assistant picks a middle path that may be too hard for a beginner or too simple for an expert.

Sometimes a sample explains the format better than a paragraph of rules.

When the format you want is hard to describe, show a small example. You can paste a sample of the kind of answer you like, or describe one in a sentence, and ask the assistant to follow the same pattern for your new content. Examples work because they carry many details at once: the length, the layout, the tone, and even the small habits of style.

A good practice is to give one or two examples that are clearly different from the content you want, so the assistant copies the shape and not the words. If you also mention what you do not like, the picture becomes even sharper. Some people find it helpful to give one example of a good result and one example of a result they would reject.

Positive instructions usually work better than long lists of bans.

It is natural to describe what you dislike. “Do not use bullet points. Do not make it long. Do not sound stiff.” Such rules can help, but they leave the assistant guessing what to do instead. A stronger approach is to name the target directly: “Write in two short, friendly paragraphs.” The target tells the assistant exactly what success looks like.

You can still mention an avoid rule when it matters, especially for a habit you keep seeing. The best prompts often pair the two: describe the desired shape first, and then add one short note about the thing to skip.

Placement and clarity both matter, especially in longer prompts.

In a short prompt, a format sentence anywhere is usually fine. In a longer prompt with pasted documents, notes, or data, it helps to keep your instructions clearly separate from the material. Many people state the task and the format first, then provide the material, and sometimes repeat a one-line reminder of the format at the end.

Clear separation also prevents confusion. If your pasted text contains lists or headings, the assistant may copy that style by accident. A line such as “The text below is source material. Please write your answer as a short summary in plain paragraphs” keeps the roles apart.

flowchart TD
    A["1. The task
What do you want done?"] --> B["2. The format
Length, structure, tone, audience"]
    B --> C["3. The material
Notes, text, or data to use"]
    C --> D["4. Short reminder
of the key format rule"]
        

Sometimes the reader of the answer is another program, and the rules become stricter.

Not every answer is read by a human. In many tools and workflows, the assistant’s output is passed straight to another piece of software, which expects a precise layout. Examples include data in a fixed field-by-field arrangement, a comma-separated list for a spreadsheet, or labeled sections that a program will split apart. If the assistant adds friendly extra sentences, the program may break.

When your answer is going to software, be very explicit. Name the exact layout, list the fields you want, say what to do when information is missing, and state that nothing else should be added before or after. It is wise to test with a few real cases, because even small variations can cause problems later.

The place where the answer will land is the best clue for choosing the format.

Before you send a prompt, ask one simple question: where will this answer go? A chat window can display neat formatting. A plain-text email may show stray symbols if the answer contains special formatting marks. A slide needs very short phrases. A spreadsheet needs rows and columns. A text message needs to be tiny. When you name the destination, the assistant can choose a fitting shape without further instruction.

flowchart TD
    Q["Where will the answer be used?"] --> E["Email or message"]
    Q --> S["Slide or poster"]
    Q --> T["Spreadsheet or table"]
    Q --> P["Another program"]
    E --> E1["Plain paragraphs
and a clear tone"]
    S --> S1["Very short phrases"]
    T --> T1["Rows and columns"]
    P --> P1["Strict layout,
nothing extra"]
        

Follow these steps and your prompts will be clearer almost every time.

Seeing the difference in real situations makes the idea easy to remember.

You do not have to start over. A short follow-up usually fixes the shape.

Even a good prompt can produce an answer in the wrong shape. This is normal and easy to handle. Because the assistant remembers the conversation, you can simply ask for a change: “Make that shorter,” “Turn this into a table,” “Rewrite it as plain paragraphs,” or “Keep the content, but use a friendlier tone.” Content and format are separate, so changing one rarely damages the other.

If the same problem keeps appearing, look at your original wording. Was the instruction vague? Did you describe what to avoid but not what to produce? Did a pasted sample accidentally set the style? A small tweak to the original prompt often fixes the issue for every later request.

flowchart LR
    A["Send prompt with
format details"] --> B["Read the answer"]
    B --> C{"Right shape?"}
    C -- "Yes" --> D["Use it"]
    C -- "No" --> E["Ask for a specific change
to length, layout, or tone"]
    E --> B
        

More control is helpful, but too many rules can also cause trouble.

The best approach is proportion. Use light guidance for casual questions and stricter rules only when the output has an important job to do, such as feeding a system or being sent to a customer.

These habits cause most format problems, and each has a simple cure.

If you ask for the same shape again and again, save yourself the typing.

Many people repeat the same format wishes in every conversation: short answers, plain language, a summary first. Instead of repeating them, look for ways your assistant lets you set standing preferences. In Claude, for example, you can describe your general preferences in your profile settings, give project-specific instructions inside a project, or choose a response style. Exact menu names can change, so check the current help center for where these live.

A useful rule is to keep always-true preferences in one place and task-specific format rules in the prompt itself. That way your everyday habits are handled automatically, while a special request, such as a strict layout for one report, can still override them when needed. If a standing preference and your current request disagree, your current, specific request is usually the clearer signal, so state it plainly.

A few popular ideas about format requests are not quite right.

Quick answers to the questions beginners ask most often.

Name the reader or the destination. Once the assistant knows who will read the answer and where it will be used, it can choose a sensible length, tone, and layout.

No. Everyday words such as short, plain, step by step, or in a table work very well.

Either works for short prompts. For longer prompts with pasted material, state the task and format first, then provide the material, and optionally repeat a short reminder at the end.

Ask positively for what you want, such as plain connected paragraphs, and optionally add a short note that you do not want lists or headings.

Yes. A sample is often the clearest way to show length, layout, and tone. Say whether the assistant should copy the style, the facts, or both.

Ask for the change directly, such as shorter, as a table, or in plain paragraphs. The content usually stays the same while the shape changes.

Name the exact layout, list every field, say what to do when information is missing, and state that nothing else should be added. Test it on varied cases and check the result automatically if you can.

A few clear rules usually work best. Too many rules can conflict, so keep only the ones that truly matter for the task.

Often yes. Many assistants let you set standing preferences or instructions. Keep general habits there, and put task-specific rules in the prompt.

It should not. Format changes how the answer is shown, not what is true. Still, it is always wise to review the content, especially after heavy shortening.

Yes, the idea is general. Clear descriptions of length, structure, tone, and audience help with nearly every assistant, although small details of how each one responds can differ.

An AI assistant can write brilliant content and still hand it back in a shape you cannot use, simply because it was never told what shape you needed. Format is the separate half of a good prompt. It covers length, structure, tone, audience, order, and destination, and each can be described in plain words. By naming the reader, choosing a shape, setting a length, showing an example when needed, and asking for the desired result rather than only listing bans, you get answers that are ready to use. When the first answer misses, a short follow-up reshapes it. Over time, saving your favorite habits as standing preferences makes every conversation smoother.