How do you control response length and tone?

Ask ChatGPT a simple question and you often get five paragraphs, a bulleted summary and a cheerful closing line. Ask it to "be professional" and the result may sound stiff. Neither is a flaw in the tool. It fills in whatever you leave unsaid with its own defaults, and length and tone are the two things people most often leave unsaid.

OpenAI's own prompting advice makes the same point: be specific about the length, format and style you want, and replace vague words with precise ones. "Fairly short" is vague. "A three-to-five-sentence paragraph" is precise. The same goes for tone, which is simply the attitude a text takes toward its reader, such as formal, friendly, firm or playful.

Length: pick a unit the model can hit

ChatGPT writes in tokens, small chunks of words, and doesn't count words as it goes. Exact word counts are therefore approximate. Limits stated in sentences, bullet points or paragraphs tend to land much closer.

Two more habits help. Tell ChatGPT what to leave out: "No introduction and no closing summary" removes the padding at both ends. And if a reply is still too long, don't start over; say "Cut this by half and keep the three main points." Revising inside the same chat is faster than rewriting the prompt.

Tone: name it, then aim it at a reader

A single tone word leaves a lot of room. Pair two or three words with a description of the reader, and the model has far less guessing to do. Reading level is part of this. "Write for a 12-year-old" or "plain language for adults who are new to the topic" changes sentence length and vocabulary at once.

OpenAI also advises saying what to do rather than only what to avoid. "Don't be stiff" gives the model nothing to aim at; "write the way a friendly colleague talks across a desk" does.

A worked example: one message, three settings

Here is a prompt that sets both length and tone:

A customer's order of six chairs will arrive five days late because of a supplier delay.
Write the email we send them.
Length: three short paragraphs, under 120 words in total, no subject line.
Tone: warm, direct and apologetic without grovelling. The reader is a small-business
owner who is busy and mainly wants the new date.
Reading level: plain language, short sentences.
End with one concrete offer, not a general "let us know if you need anything."

Change only the tone line and the same facts come out very differently. Here are illustrative opening lines for three settings:

How do you keep a consistent voice across chats?

Each new chat starts from scratch, so whatever you set in one conversation is gone in the next. The tool for this is custom instructions, which OpenAI introduced in July and opened to free users on August 9 and to users in the EU and UK on August 21. You find it in your settings menu. It asks two questions: what ChatGPT should know about you, and how you would like it to respond. Whatever you write there is considered in every new conversation.

Put your length and tone defaults in the second box. For example:

Default to short answers: one paragraph or up to five bullet points, unless I ask for more.
No introductions, no closing summaries.
Tone: clear, warm and plain, like a patient teacher. Canadian spelling.
When you write for me, use the first person and short sentences.
If you are unsure of a fact, say so.

Three habits keep that voice steady over time. First, write instructions you would be happy to see applied everywhere; for one-off jobs, set length and tone in the prompt itself, where it overrides the defaults. Second, keep a short "style card" in a note (tone words, reading level, a few phrases you like and a few you avoid) and paste it into a chat when you work on something important, such as a newsletter. Third, check now and then. When it launched custom instructions as a beta, OpenAI warned that ChatGPT may sometimes overlook them, or apply them when you didn't intend it. If replies start drifting back to long and generic, a quick "Follow my custom instructions on length" usually brings them back.