Shaping the answer, format, tone and length

In the earlier steps I learned to ask well and to correct on the fly. Here I discovered something that put order into the way I work: besides choosing what I ask the AI for, I can choose how it hands it back to me. And one sentence is almost always enough.

This step closes the block on talking to it well. If you climb it, you come out knowing how to say three things that hold for everything that comes next.

Form is something you ask for too

For a while I put up with answers exactly as they reached me. I'd ask for something, get back a dense paragraph and, if that form didn't work for me, I'd rewrite the whole question hoping the second time would come out better. It took me a while to realise I was fighting what wasn't worth fighting.

Because the form of the answer isn't a closed thing you have to put up with. It's another lever, and you're the one who moves it. If you get a tight block of text and you wanted it as a table, you don't rewrite anything: you say "put it in a table" and that's that. The material the AI had inside is the same; only the mould it hands it to you in changes.

This ties back to what we saw at the start of the staircase. An AI chat builds the answer out of what you put in front of it. If you say nothing about the form, it's not that there's no form: it's that it picks the form by default, and the default almost never matches your particular need. So you're better off telling it.

Format: the mould of the answer

The first lever is format: which container you want the content in. The same information can come out as a running paragraph, a numbered list, a table, an email or steps one through five. It isn't an aesthetic whim; each mould serves a different purpose.

The phrasings I use are as simple as they sound. "Make me a numbered list." "Lay it out as an email." "Give me a table with two columns, concept and explanation." "Sum it up in five steps." The APD, in its run-through of how to fine-tune a prompt, says it without hedging: if you want the answer in a particular format, you have to say so in the request, and it gives as examples exactly that —list, table, report, three paragraphs—. It doesn't guess; you tell it.

The nice part is that the same content travels between moulds without your having to regenerate it from scratch. If you asked for a paragraph and then you'd rather have it as a table, you don't repeat the question: you chain a "now put it in a table" and the AI reorders what it already had. Ask, look, reshape, like in the step on dialogue.

Tone: how it sounds

The second lever is tone: the register it speaks to you in. The same explanation can come out formal or warm, technical or simple, dry or kind. And here the change is sometimes more useful than it looks, because a correct text with the wrong tone is no use to you.

The phrasings, again, are easy. "Explain it to me like I'm a child." "Put it in a professional tone, it's for a client." "More casual, it's for a group of friends." "Drop the waffle and keep it technical." Xataka itself, in its round-up of ChatGPT uses, reminds you that you can ask it to write something in a more formal or more informal way without touching the substance. You change the voice, not the information.

For me this solved an everyday problem. The same idea I sometimes needed clear for someone starting from scratch and sometimes dense for someone in the trade. I used to write two different questions; now I write one and then say "and now the version for someone who already knows the subject." Same material, two registers.

Length: how much it gives you

The third lever is length: how far you want it to go. By default the AI tends towards a medium length that doesn't always fit you: sometimes it falls short of what you need, sometimes it rambles on when you only wanted the bare idea.

You head it off with phrasings of the same kind. "Sum it up in three sentences." "A hundred words at most." "Too terse, develop it more." "Just give me the headline, no explanation." You set the bar and the answer adjusts to it. It's the lever you notice fastest, because the result shows at a glance: either it fits what you asked for or it doesn't.

With this you close the game. The same information can come out as an email to your boss, a short list to jot down, or a long explanation for someone starting from zero. The raw material doesn't change; you decide the mould, the voice and the size.

The three combine

You don't have to move the levers one at a time. The usual thing is to ask for all three at once in the same sentence: "make me a table, in a simple tone and short." Format, tone and length in a single shot. The clearer you are with the three, the fewer rounds you go afterwards.

And since no request comes out perfect first time, this leans on the dialogue you already know. You ask with the three levers set, look at what arrives and readjust the one that came out weak: "the table's fine, but shorten it", "the tone's perfect, but lay it out as a list". You don't start from scratch; you fine-tune.

The price of shaping

It's worth knowing that shaping isn't free. When you ask for "shorter" or "simpler", the AI cuts, and sometimes what it cuts is a nuance that did matter. And when you ask for "longer", it pads, and part of that padding can be filler that adds nothing. The form you gain you pay for, sometimes, in content.

So the rule I came away with is not to trust the shaped result blindly. I read it properly: if I asked for a three-sentence summary, I check it hasn't dropped the essentials; if I asked it to develop something, I look at whether what it added says anything. Shaping the form is a powerful lever, but like any lever, it's worth watching where you move it.

With this we run out of what can be done talking to it with text alone. From here on, the conversation stops being only written words, and that's where the next stretch opens.

Definitions

- Format: the mould you ask the answer in —list, table, email, steps, paragraph—. Changing it doesn't change the information, only how it's presented. - Tone: the register the AI speaks to you in —formal, warm, technical, simple—. You ask for it with one sentence and it adjusts the voice without touching the substance. - Length: how far the answer goes. You set it with requests like "in three sentences" or "develop it more", and it's the thing you notice fastest. - Shaping the answer: telling the AI not just what you want, but how you want it, moving format, tone and length together.

Further reading

- APD, Cómo optimizar tus prompts para ChatGPT — a run-through of task, tone and format with examples, in Spanish. https://www.apd.es/prompts-para-chat-gpt/ - Xataka, 22 funciones y cosas que puedes hacer con ChatGPT para exprimirlo al máximo — includes adjusting the tone of the answer, in Spanish. https://www.xataka.com/basics/22-funciones-cosas-que-puedes-hacer-chatgpt-para-exprimir-al-maximo-esta-inteligencia-artificial - Xataka Colombia, Sabías que puedes pedirle a ChatGPT información en formato de Excel: así puedes hacerlo — a concrete example of shaping the output format. https://www.xataka.com.co/otros/sabias-que-puedes-pedirle-a-chatgpt-informacion-formato-excel-asi-puedes-hacerlo