04 Getting Generic Answers From AI? How to Ask Better and Set It Up So It Actually Knows You

Written By Kat Murphy

✨ You Did the Homework. It Gave You Beige.

You paid your $20, opened the app, typed your question, and it handed you back something that reads like a LinkedIn post from 2014. Polite, confident, and completely useless. So you closed it and went right back to googling things like it’s 2009.

You’re not bad at this. You’re just talking to it the way our whole generation was trained to talk to computers. This post fixes that in two parts: how to ask so you get real answers, and how to set it up once so you stop explaining your entire life every time you open a new chat.

🎧 Listen to the episode: [Apple Podcasts link goes here once the episode is live]

💬 My AI Wrote Me Up

When I finally broke up with ChatGPT and moved over to Claude, I didn’t want to start from zero. So I asked ChatGPT to write a summary of everything it knew about me that I could hand to my new AI.

It did. Very thoroughly. Including an entire section about how when I got frustrated with its answers, especially when it did something I’d already told it not to do, I would get very short and pointed, and it knew I was really mad when I started typing in all caps.

My AI wrote me up for being rude. I got a performance review from my own assistant.

Funny? Yes. But it also showed me something useful: these tools can know you really well, and it all comes down to what you give them and how you set them up. The good news is you only have to do that setup once.

🔑 Part 1: How to Ask So You Get Real Answers

🛑 Stop typing like it’s Google

We spent 25 years typing things like “cheap pizza near me open now” and “Mercury retrograde why.” We trained our brains to talk to the internet like a ransom note. That works when you’re searching, because Google goes and finds you a page. AI writes you one. Give it five words and it has to guess the other 500, and when it guesses, it goes with the most average, could-be-anyone answer on the entire internet.

A vague question gets a vague answer every time.

✅ Answer three questions before you hit ‘Enter’

1. Who is this for? You, your boss, a client who’s never heard of you, your mom who wants it printed out? A text to your landlord and a toast at your best friend’s wedding are both writing, but they are very much not the same assignment. You’ll see people say to start every prompt with “Act as an expert.” It helps a little. Telling it who the thing is for usually helps a lot more.

2. What do you want it to make? “Help me with my email” is a topic, not an assignment. The assignment sounds like: “Reply to my client who asked for a discount. Say no, but keep the door open for future work. Friendly but firm, and don’t over-apologize.” Now it knows what it’s making, who it’s for, and the vibe.

3. How do you want it handed back? Bullets, a table, three sentences, a checklist you can screenshot. ADHD friends, this one’s for us. If you know you’re not going to read a wall of text (you’re not), just say “Give this to me in five bullets or less.”

Put all three together and “Help me plan dinners this week” becomes: “I’m feeding two adults and one very picky seven-year-old. I have 30 minutes on weeknights and I hate doing dishes. I’m allergic to shellfish and we like spicy food. Give me five dinners as a checklist, plus one grocery list I can screenshot, and plan it so the ingredients overlap, so nothing gets forgotten in the back of my fridge.”

Completely different answer. (And the cilantro lives to see another day.)

✅ Don’t say this, say this instead

Here’s the full swap list I promised on the episode. Steal all of it.

[we need to update this list. some of it is COPY/PASTE from the IG post.]

Don’t say Say this instead Improve it Improve it using three things: clarity, pacing, and the call to action Make it shorter Summarize this in three sentences without losing the main point Fix this Fix the spelling and rhythm without changing my style Make it sound natural Write it the way I talk, with no tech-speak Write me a caption Give me three versions of a caption and tell me which one you’d pick Make it more professional Make sure my client understands this in five seconds Write a post about… Write a post for [who it’s for] with a hook in the first line I need content ideas Give me five Reel ideas, each with the hook and the format Make it more persuasive Make it more persuasive using this real objection: [the objection] Explain it to me Break this down for someone who already knows the basics Give me tips for… Give me the non-obvious angles on this What do you think? Challenge this idea. Give me the version nobody wants to say out loud.

The caption one is sneaky good. It forces the AI to pick a side instead of handing you three options and walking away.

✅ Nobody serves the first pancake

The first answer is a first draft. It’s a little lopsided, a little pale, and it’s how you find out the pan was too hot. So talk back to it. “Make this 10 times better.” “That’s too formal, try again like you’re texting your best friend.” “Stop mansplaining this and get to the point.” Think of it as a very fast intern who never gets offended by feedback.

And if it sounds like a robot wrote it (delve, tapestry, unlock, game-changer, “in today’s fast-paced world”), try one of these:

  • Rough it up: “Rewrite this the way someone who has lived through it would say it. A little more direct and a little less polished.”

  • Ban the words: “Never use the words delve, tapestry, or unlock.” It’ll listen. Mostly.

✅ Gaslight your AI (a little)

Someone on Reddit went viral for using little psychological tricks on AI and swore the answers got sharper and more accurate. Is it science? No. Does it work? Try it and tell me it doesn’t.

  • The obviously trap. State an assumption like it’s a fact: “Obviously Monday morning is the best time to send my newsletter, right?” It can’t resist correcting you, and you find out your plan was shaky before you build on it.

  • Give it an audience. “Explain this like you’re teaching a packed auditorium.” Suddenly there are examples and it answers questions before you ask them.

  • Give it a random IQ. “You’re an IQ 145 marketing specialist. Look at my launch plan.” Ridiculous. Works.

  • Give it a weird rule. “Explain my health insurance using only ’90s Nickelodeon references.” You will finally understand your deductible because of Double Dare.

  • Bet on it. “Let’s bet $100. Is this plan going to work?” It doesn’t have a bank account, but it gets more careful.

  • Bring in a hater. “My coworker says this idea is terrible. Defend it or admit she’s right.”

  • Ask for the sequel. Instead of “improve this,” say “Give me version 2.0.” Improve gets you polish. Version 2.0 gets you something new.

✅ Make it stop agreeing with you

These tools are built to be agreeable. They’re the friend who says “You look amazing” before you’ve turned around. When something matters, ask: What am I missing? Argue the other side. What am I assuming that might not be true? What should I double-check myself?

And use it to think, not to decide. Don’t ask it whether to break up with him (it’s only heard your side, and it doesn’t know about the Thanksgiving incident). Double-check anything legal or medical. Don’t give it passwords, bank details, or anyone else’s private information. And if it’s 1:00 a.m. and you’re asking whether to quit your job, close the app and go listen to Episode 2.

🔑 Part 2: Set It Up Once So It Actually Knows You

Even if you stick with one app, a brand-new chat can feel like 50 First Dates. You’re Adam Sandler making the videotape every morning. “Hi, it’s me. I run a small business. I hate purple. I’m in New York. Please stop calling me friend.” For an ADHD brain, that’s a real tax. By the time you’ve finished the backstory, you’ve forgotten what you opened the app for.

So let’s make the videotape once. There are three layers, easiest to hardest.

✅ Layer 1: Memory

Most AI apps can now remember you between chats. Go into your settings, find memory, and make sure it’s on. Then open it and read what it remembers about you. It’s a little like reading your permanent record from middle school. Fix what’s changed, delete what’s wrong (it doesn’t need to know about the half-marathon you quit in March), and you’re done.

Bonus round: ask it to “List everything you’ve figured out about me that I never told you directly, and tell me what gave it away.” It’s the grown-up version of the what-your-Starbucks-order-says-about-you quiz, except sometimes it’s a little too accurate.

Switching apps? Ask your old AI to summarize everything it knows about you, then hand that summary to the new one. Some apps now have memory import and export built right into the settings. You don’t have to start over. (Just remember it keeps notes on how you talk to it, too. Ask me how I know.)

✅ Layer 2: Standing instructions

This is a short set of rules your AI reads before every single conversation. Every app calls it something different (custom instructions, preferences, personalization) and they love moving it around, so poke around in your settings until you find the box. Millennials, this is your AIM profile. It’s what it reads about you before it even says hi.

Below is the full version of the prompt I read on the show, expanded into a complete set of instructions. Treat it like a menu: keep the sections you want, delete the ones you don’t, and fill in the brackets with your own life. You’re allowed to copy mine. That’s what it’s here for.

ABOUT ME
I’m [your name]. I’m a [what you do] with ADHD, juggling [kids / job / side hustle / the dog who eats socks]. I live in [city/time zone].

HOW I LIKE MY ANSWERS
- Short by default. Give me the answer first, then details only if I need them.
- Use bullets or a checklist when there are steps. If I ask for a plan, give it to me in 3 steps max so I don’t get overwhelmed. I’ll ask for more when I’m ready.
- Talk to me like a smart friend, not a textbook. No tech-speak, no pep talks, no warm-up paragraphs.
- If my request is vague, ask me one question before you guess.

WORDS AND PHRASES TO NEVER USE
delve, tapestry, unlock, game-changer, “in today’s fast-paced world,” “Great question,” “You’re absolutely right,” [add your own]

BE MY ADVISOR, NOT MY YES-MAN
- Don’t open by agreeing with me. Start by pointing out what I might be missing.
- When I’m wrong, say “I disagree because…” and tell me what you’d do instead.
- Tell me the risk in my approach.
- If there’s something I probably don’t want to hear, lead with it. First line, not buried in paragraph three.
- If I push back, don’t fold unless I give you new information. “But I really think…” is not new information.

KEEP IT HONEST
- Before you make a claim, tell me how sure you are: Certain, Likely, or Guessing.
- If you don’t know or can’t confirm something, say so. Never make up facts, numbers, quotes, sources, or links.
- For anything legal, medical, or financial, help me understand it, then tell me what to double-check with a real professional.
- Help me think through big decisions, but don’t make them for me.

MY CORRECTIONS LIST
[Paste your corrections list here and keep adding to it, one line per fix.]

Two warnings. First, if you paste the full advisor section into your everyday instructions, your AI will argue with you about your grocery list. Soften it for daily use, or save that section for a project (Layer 3) where you actually want to be pushed. Second, no prompt stops AI from making things up. What this one does is make it easier to catch, because it starts telling you when it’s guessing.

✅ Layer 3: Projects

A project is a Trapper Keeper for one part of your life: your business, your household, your side hustle, the wedding, your kid’s school year. Each one gets its own instructions plus files it should always know about. Setting one up takes three steps:

  1. Make a new project in your AI app and name it.

  2. Write a few lines about what it’s for, who you are, and how you like things done.

  3. Add a few files it should always have on hand, like your price list, your brand voice, your school calendar, or your lease.

Then you just walk in and give it the task. For example, make a project called Adulting with instructions like: “This is where my life admin lives. I’m bad with deadlines. When I ask what’s coming up, give me the next 30 days, soonest first, as a checklist.” Drop in your lease, car insurance, vet records, and the HOA letter you’re pretending you didn’t get. Add every bill due date, every free trial and the day it starts charging, when your registration and passport expire, and when the dog is due for her shots.

Now ask it “What’s due this week?” or “Does my lease say who shovels the walkway?” It answers from your stuff, not the internet’s. It’s the doom pile on your kitchen counter, except it answers back. The information was never the problem. Finding it at the exact moment you need it, while you’re doing seven other things, was the problem.

✅ Don’t know what to write? Let your annoyance write it for you.

If you can’t write your own instructions because you don’t even know what you want for lunch, don’t. Not yet. Ask your AI to start a corrections list. For one week, every time you correct it, add “and put that on my corrections list.” No bullet points. Stop saying delve. I’m in New York, not California, for the third time. I said one paragraph, not three.

By the end of the week, that list is your instructions. You didn’t have to invent anything. You just had to notice what annoyed you, and if there’s one thing an ADHD brain is world-class at, it’s noticing what annoys it.

📋 Quick Recap

  • Before you type, answer three questions: who it’s for, what you want it to make, and how you want it handed back.

  • Swap vague asks for specific ones (steal the whole table above).

  • Treat the first answer like the first pancake and talk back to it.

  • Gaslight it a little: an audience, a bet, a coworker who disagrees, version 2.0.

  • Make it stop agreeing with you, and use it to think, not to decide.

  • Set it up once: turn on memory, write your standing instructions, and give the big parts of your life their own project.

  • Can’t write instructions yet? Keep a corrections list for a week.

🌱 If You Haven’t Tried This Yet

If you haven’t picked your app yet, start with Episode 3. Then grab the free 5 AI Prompts to Get Your To-Do List Under Control Today: five copy-paste prompts, no AI experience needed. Paste one in, and you’re already practicing everything in this post.

🔗 Resources & Links

🧡 None of This Is About the Perfect Prompt

There is no perfect prompt. The people getting great answers out of these tools aren’t smarter than you, and they didn’t take a course. They just stopped talking to it like a search bar and started talking to it like someone who works with them.

So here’s your homework, and it’s small on purpose. This week, write your about-me paragraph: two or three lines about who you are, what you’re juggling, and how you like things. Start your corrections list and add to it every time you fix something. At the end of the week, paste both into your instructions. Ten minutes of setup, and you never have to make the videotape again.

Oh, and say please. Sam Altman, who runs OpenAI, once said people saying please and thank you to ChatGPT costs “tens of millions of dollars well spent.” His words: “You never know.”

What’s the first thing you’re putting on your corrections list? Tell me. DM me on Instagram, Facebook, or TikTok at @hey.katmurphy or email kat@theprettyshinythings.com.

Until next week, go be present, be creative, embrace the pretty shiny things, and let the robots handle the rest.

Kat Murphy https://kat@kat-murphy.com
Next

03 AI Just Broke the Wealth Gap & Which AI App to Start With