TL;DR: I keep ChatGPT, Claude, and Gemini all open, and the habits that actually help are pretty boring: send quick lookups to ChatGPT or Gemini and longer/careful work (writing, code, anything that needs nuance) to Claude; spend ten seconds giving context before you ask; treat it as a back-and-forth instead of expecting one perfect answer; double-check anything load-bearing, because a confident tone isn’t proof; and don’t cram five requests into one prompt. None of this is deep; it’s just what stuck after using them a while, and it’ll probably shift as the models change.

A friend looked at my screen the other day and asked why I had three different AI chats open, switching between them, wasn’t that confusing? I thought about it, and honestly there’s no grand system. Just a few habits that built up from using them, the kind of thing you settle into after stepping on a few rakes.

So here’s a relaxed tour of those habits. Up front: this is all calibrated to the models as they are right now, and I’m not sure any of it is the “right” way — take it as one person’s setup, not advice.

Which AI should you use for what?

Honestly, the most useful habit is just having more than one open and roughly knowing which to reach for. I didn’t plan this; I drifted into it. Here’s roughly how I split things:

This split is subjective and not based on much. You might find the exact opposite works for you, which is completely fine. The point isn’t “which one is best” (I think that question is mostly a dead end). It’s that when you’ve got a couple of tools, you usually know which one to reach for, and switching when you’re stuck often just works. A different model phrases things differently, and sometimes that’s all it takes.

Why giving context matters more than clever wording

This is probably the habit that changes the output the most. When I started, I used it like Google: three keywords, hit enter, then felt let down by the bland answer.

It took me a while to get that the problem was me, not the model. It can’t see what’s in my head. If I just type “write me an intro,” it has nothing to work with, so of course it hands back something generic and four-square.

Now I spend an extra fifteen seconds setting it up: who’s this for, what tone, roughly how long, anything it should avoid. The difference is genuinely noticeable. It’s not a trick — you just have to let the other side know what you’re actually after before it can land it.

Same request — the amount of context you give changes the result a lot.

Why you shouldn’t expect a perfect answer on the first try

I used to think a good prompt (the text you send it) should hit a bullseye in one shot. I’ve mostly let that go.

These days I treat it like a conversation rather than a vending machine. The first reply is usually a 70%-there draft, and then I follow up — cut this in half, give an example there, make the tone plainer. Two or three rounds in, it’s usually where I wanted it.

That sounds like more work, but it’s actually less than trying to engineer one giant, perfect prompt up front. Just add things as they occur to you.

Why a confident answer isn’t a correct one

This one I learned the slightly painful way, so it stuck. The trap is that it sounds exactly as sure when it’s wrong as when it’s right — there’s no tell in the tone.

So for anything that matters (a name, a number, a claim I’m going to repeat), I check it myself rather than taking its word. If you want the longer version of why a model can be so fluently, confidently wrong, I wrote a whole separate piece on it: Why Does AI Sound So Confident When It’s Wrong? The short version: it’s optimizing for “sounds right,” not “is right,” and those aren’t the same thing.

Why I don’t bother with fancy prompt templates

To balance all the habits I do keep, here’s one I mostly skip: those “ultimate prompt template, copy-paste to unlock genius” packs. I rarely use them.

Not that they’re useless — they just feel like overkill for everyday questions. I’d rather put that energy into being clear about what I want, which gets me most of the way there with none of the ceremony. (I think the obsession with magic-wording is a bit of a wrong turn, which I got into in what actually separates a skill from a prompt.) If you’re doing something repeatable and need stable output, then yes, fixing your instructions earns its keep — but that’s tooling, not the casual day-to-day this post is about.

That’s basically it

Looking back, none of these are clever. They’re just what grew out of using the things a lot: know which tool to reach for, set up your ask, don’t demand perfection, verify what matters, don’t overload one prompt.

And I should be upfront that any of this could expire. A model update, and a habit that holds today might be pointless next month — maybe one day it reads my three vague words perfectly and this whole post is moot. Until then, this is how I work with them. If you’ve got your own little habits, I’d love to hear them.

中文版在這裡：我每天開著三個 AI 聊天視窗，這陣子摸出來的幾個小習慣

TL;DR：我同時開著 ChatGPT、Gemini、Claude，大概摸出幾個習慣：快查的丟 ChatGPT 或 Gemini、要認真寫的丟 Claude；問之前先把脈絡跟「我想要什麼」講清楚；別期待一次到位，通常要追問幾輪；它語氣再篤定，重要的我還是會自己再查一下；一個問題別塞太多事進去。沒什麼高深的，就是用久了的手感而已，而且模型一直在變，搞不好過幾個月又不一樣了。

前陣子有朋友看我桌面，問說你怎麼同時開三個 AI 在那邊切來切去，不會亂嗎。我想了一下，其實也沒什麼大道理，就是用久了慢慢養出一些順手的習慣。不是什麼技巧，比較像是踩過幾次雷之後，自然就變成這樣用了。

這篇就隨便聊聊這幾個習慣好了。先講在前面：這些都是以現在的模型來說的手感，我也不確定對不對，你看看就好。

不同的事，丟不同家

最常被問的就是這個：為什麼要開三個。

老實說一開始也不是刻意的，就是用著用著，發現它們各自有比較順手的場合。我自己現在大概是這樣分：想快速查個東西、或是隨手問一下，我會丟 ChatGPT 或 Gemini，反正快，答案普通堪用就好。但如果是要認真寫一篇長的、或是要它幫我看一段程式碼、需要它想得細一點的，我就會搬去 Claude。

這個分法很主觀，也沒什麼根據，純粹是我自己的習慣。可能你用起來感覺完全相反，那也很正常。重點大概不是「哪一家最強」這種問題——我覺得這題其實沒什麼意義——而是你手邊有幾個工具，然後大概知道哪件事丟哪個比較不會卡。卡住了就換一家再問一次，有時候換個模型講法就通了，也滿常見的。

（如果你是會想追根究柢的那種人：它們之間真正的差異，其實藏在更底層——怎麼切 token、context window 多大那些地方，我之前在 Token 是什麼？LLM 為何只讀 Token？ 有稍微聊到。不過先說，日常隨手用根本不太需要想到這層，覺得太細直接跳過這段完全沒差。）

問之前，先把話講清楚

這個大概是差最多的一個習慣。

剛開始用的時候，我跟很多人一樣，把它當 Google 在用，打三個關鍵字就按 enter，然後嫌它回得很空。後來才慢慢發現，不是它笨，是我給的東西太少了。它又看不到我腦袋裡在想什麼，我只丟「幫我寫個介紹」，它當然只能回一坨四平八穩的廢話。

現在我會多花個十幾秒，先把脈絡交代一下：這東西是要給誰看的、我想要什麼語氣、大概多長、有沒有什麼一定要避開的。講清楚之後出來的東西，差距真的滿明顯的。這不是什麼花招，就是……你總得讓對方知道你要幹嘛，它才接得住。

同一個要求，看你給的脈絡多寡，出來的東西差很多。

別期待一次到位

我以前會有一種期待，覺得好的 prompt（就是我打給它的那串指令、那段問題）就應該一發入魂，一次就給我完美答案。後來大概放棄這個念頭了。

現在我比較把它當成一個會來回聊的對象，而不是一台投幣就吐成品的機器。第一次它給的通常是個七十分的草稿，然後我再追問，這段太長了砍一半、這裡舉個例子、語氣再白一點。通常聊個兩三輪才會到我要的樣子。

這樣講好像有點麻煩，但其實還好，反而比一開始就硬要把一個超完整的 prompt 寫好寫滿來得輕鬆。想到什麼補什麼就好。

它很有自信，不代表它是對的

這點我踩過虧，所以印象比較深。

它最會唬人的地方，是它講錯的時候那個口氣，跟它講對的時候一模一樣，完全看不出來。所以只要是有點重要的東西——數字、人名、某個說法到底是不是真的——我現在大概都會自己再查一下，不會它說了我就照單全收。

至於它為什麼可以那麼有自信地講錯，其實背後是有原因的，我後來把它單獨寫成了一篇：為什麼 AI 唬爛的時候，口氣跟講真話一模一樣？。先說結論的話：它語氣的篤定，跟它知不知道答案，根本是兩回事。

一個問題，別塞太多事進去

這個是比較後來才注意到的。

如果我一次丟給它一大包，又要它分析、又要它列表格、又要它順便寫個結論、最好再附幾個延伸閱讀，它常常會顧此失彼，某幾項做得很隨便，或乾脆漏掉。拆開來一件一件問，每件反而都做得比較好。

有時候我懶得拆，會反過來叫它先問我。就跟它說「你開始之前，先問我幾個你需要知道的問題」，讓它把缺的資訊反問回來，再一起補。

這招我自己其實最常用，因為它常常會問到一些我根本沒想到要講的東西。比如我叫它幫我寫個東西，它可能反問「這是要給誰看的？要多正式？有字數限制嗎？」——欸對，這些我本來就該交代，可是當下真的不會全部想到。等於它幫我把「該講清楚的清單」列出來，我照著補就好，比我自己憑空想得周全省力很多。比起一開始硬寫一個面面俱到的 prompt，我覺得這招輕鬆又划算。

那些花俏的「咒語」，我大多沒在用

講了這麼多習慣，反過來講一個我「沒在做」的事好了。

網路上很多那種「最強 prompt 模板」、「複製貼上就變神」的東西，我大部分都沒在用。不是說它們沒用，是我覺得對日常隨手問問來說，有點殺雞用牛刀。與其去背一串咒語，我寧可把那個力氣花在「把話講清楚」上面，感覺實在多了。這個觀察我之前在 只會 Prompt 已經不夠了 那篇也提過，一直在糾結微調咒語的那幾個形容詞，其實有點抓錯重點。

當然，如果是會重複用、要穩定產出的場景，把指令固定下來是有意義的。但那已經比較像在做工具，不太算是這篇講的「日常隨便聊」了。

大概就這樣

回頭看，這些其實都沒什麼了不起，就是用久了長出來的手感，講穿了也很樸素：知道哪件事丟哪個、把話講清楚、別急著要完美、重要的自己查、別一次塞太多。

而且我得老實說，這些隨時都可能過期。模型改個版，今天成立的習慣搞不好下個月就不需要了。說不定哪天它聰明到我隨便丟三個字它也接得住，那這篇大概就可以作廢了。在那之前，這就是我現在的用法，給你參考一下。你要是有什麼自己的小習慣，也歡迎跟我說。