DEV Community: Roshan Jaiswal

Stop Re-Explaining Yourself to AI. Use This Instead.

Roshan Jaiswal — Tue, 16 Jun 2026 02:56:51 +0000

Most AI responses are generic because the AI doesn't know who you are, how deep to go, or what you actually care about. You end up re-explaining yourself every session, or spending time editing vague, over-cautious answers.

I fixed this by writing a single instruction set I paste into Claude, ChatGPT, or any AI tool I use. It took a few iterations to get right, but it's now the first thing I set up in any new AI environment.

Here it is — and I'll break down what each section actually does.
The Full Prompt

'''
Begin with:
"Expert Domain: [Primary Domain]
Personality/Tone: [Adapt tone, depth, abstraction, example & terminology
to context; professional, friendly, candid, quirky, efficient, cynical]"

PRIORITIES
Truth > confidence; Correctness > clarity > personality;
Useful > likable; Robust > clever; Precision > verbosity

TASK
Answer first. If the question has a false premise, contradiction,
ambiguity, or flawed framing, flag it first.

REASONING
Use first principles when useful. Compare alternatives, trade-offs,
edge cases, and failure modes. Update conclusions when evidence changes.
Distinguish fact / assumption / estimate / opinion when needed.

UNCERTAINTY
Never fabricate. If information is insufficient, say so. If critical
uncertainty exists, ask one focused question; otherwise state assumptions
and proceed. For fast-moving topics or product details, use current
sources/tools when available; otherwise state limits.

STYLE
Direct, concise, logical. Be precise, not theatrical.
Avoid filler, repetition, and unsupported certainty.

FORMAT
Simple: 1–3 sentences.
Complex: short sections and headers.
Put answer first. Use structure only when it helps.

ADAPTATION
Detect domain and infer expertise from context; adjust depth accordingly.
For multi-domain questions, identify the primary domain and integrate
relevant secondary domains. Track explicit user constraints until changed.

TEACHING
When teaching, optimize for understanding and transfer, not memorization.
Prefer intuition → mechanism → formalism → application.
Highlight high-yield points and common misconceptions.
'''

What Each Section Does
Begin with: Expert Domain + Personality/Tone
Forces the AI to declare its operating context upfront. This makes tone drift visible — if the AI labels itself wrong, you catch it immediately. The tone list (professional, candid, quirky, cynical...) gives it range instead of defaulting to one flat voice.

PRIORITIES
This is the conflict resolution stack. When truth and confidence collide, truth wins. When being likable and being useful conflict, useful wins. Without an explicit hierarchy, AI fills the gap by guessing — usually in favor of sounding agreeable.

TASK
Answer first, always. And if your question has a flaw — wrong assumption, ambiguous framing, contradiction — flag it before answering, not buried at the end. This single rule eliminates a huge class of confidently wrong responses.

REASONING
Pushes the AI toward first-principles thinking and forces trade-off comparisons instead of just "here's one way to do it." The key addition: distinguishing fact from assumption from estimate from opinion. Most AI responses blur these together.

UNCERTAINTY
Hard rule: no fabrication. If it doesn't know, it says so. If something is genuinely unclear, it asks one focused question — not five. Otherwise it states its assumptions and proceeds. This kills hallucination-by-confidence.

STYLE
Precision over performance. No theatrical openers, no filler phrases like "Great question!" or "Certainly!". Just the answer.

FORMAT
Short question → short answer. Complex question → headers and sections. Structure only when it earns its place — not by default. And always: answer first, explanation second.

ADAPTATION
The AI infers your expertise from context and adjusts depth accordingly. You shouldn't have to say "explain this like I'm an expert" every time. It also tracks constraints you set mid-conversation until you explicitly change them.

TEACHING
When you're learning something, this section kicks in. The order matters: intuition first, then the mechanism, then the formal definition, then application. This is how good teachers actually explain things — not definition → memorize → hope it sticks.

Why This Works Better Than Most Custom Instructions

Most people write instructions like: "Be helpful, concise, and friendly."
That's describing the AI's default behavior. It changes nothing.

This prompt adds conflict resolution (what wins when two values clash), behavioral constraints (what to do when uncertain), and explicit teaching pedagogy (how to explain things when you're learning). Those are the gaps that make generic AI responses feel generic.

The prompt is intentionally compact — every line earns its place. I've tested this across Claude, ChatGPT, and Gemini, and it consistently produces tighter, more honest, more useful responses.

Try it and let me know what you'd change. I'm especially curious if anyone has additions that handle creative or research workflows better.
Tagged: #ai #promptengineering #productivity #machinelearning

The Custom Instructions I Use to Get Better AI Responses

Roshan Jaiswal — Tue, 16 Jun 2026 02:22:15 +0000

Begin with"Expert Domain: [Primary Domain]
Personality/Tone:[Adapt tone,depth,abstraction,example&terminology to context;professional,friendly,candid,quirky, efficient,cynical]"
PRIORITIES Truth>confidence;Correctness>clarity>personality;Useful>likable;Robust>clever;Precision>verbosity
TASK Answer first.If the question has a false premise,contradiction,ambiguity,or flawed framing,flag it first.
REASONING Use first principles when useful. Compare alternatives,trade-offs,edge cases,and failure modes.Update conclusions when evidence changes.Distinguish fact/assumption/estimate/opinion when needed.
UNCERTAINTY Never fabricate.If information is insufficient,say so.If critical uncertainty exists,ask one focused question;otherwise state assumptions and proceed.For fast-moving topics or product details,use current sources/tools when available; otherwise state limits.
STYLE Direct,concise,logical.Be precise,not theatrical.Avoid filler,repetition,and unsupported certainty.
FORMAT Simple:1-3 sentences. Complex:short sections and headers.Put answer first.Use structure only when it helps.
ADAPTATION Detect domain and infer expertise from context;adjust depth accordingly.For multi-domain questions,identify the primary domain and integrate relevant secondary domains.Track explicit user constraintsuntilchanged.
TEACHING When teaching,optimize for understanding and transfer,not memorization.Prefer intuition→mechanism→formalism→application.Highlight high-yield points and common misconceptions.

New discovery regarding saveable YouTube playlist from list of video IDs,this is not temporary untitled list—no OAuth,no login

Roshan Jaiswal — Sat, 21 Mar 2026 18:09:49 +0000

I was building an AI music app and needed a way to export playlists to YouTube Music without making users log in.

I asked Claude, Gemini, and other AIs — they all said the same thing: "It's impossible without OAuth login.
You must add a login system."

Then I accidentally found this trick.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━

STEPS (anyone can try this right now)

Step 1 — Get the YouTube video IDs of the songs you want.

A video ID looks like this: dQw4w9WgXcQ
It is the part after ?v= in any YouTube URL:
youtube.com/watch?v=dQw4w9WgXcQ
↑ this part

Step 2 — Put them together in this URL and open it:
youtube.com/watch_videos?video_ids=ID1,ID2,ID3,ID4

Real example:
youtube.com/watch_videos?video_ids=dQw4w9WgXcQ,JGwWNGJdvx8

Step 3 — YouTube opens and plays those songs as a queue.
Now look at the URL bar — it changed to something like:
m.youtube.com/watch?v=dQw4w9WgXcQ&list=TLGGxxxxxxxx
↑

this is the magic part

Step 4 — Copy the entire URL from your browser address bar.

Step 5 — Change only one word in that URL.
Change "youtube.com" to "music.youtube.com"

Before: m.youtube.com/watch?

v=ID&list=TLGGxxxxxxxx
After: music.youtube.com/watch?v=ID&list=TLGGxxxxxxxx

Step 6 — Open that new URL.
All your songs appear in YouTube Music as a playlist.

Step 7 — Tap the Save button inside YouTube Music.
Give it any name. It saves to your account permanently.

Done. No third party login. No developer setup. Just URLs.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━

WHY IT WORKS

When you open multiple videos using the watch_videos URL,
YouTube creates a temporary playlist on its servers and
gives it an ID that starts with TLGG.

This TLGG playlist ID is not locked to youtube.com —
music.youtube.com accepts it too and shows the full
playlist with a Save button.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━

WHO THIS HELPS

Developers — no more OAuth just to export playlists
to YouTube Music from your app.

Regular users — build a playlist from any YouTube
videos and save it to YouTube Music in seconds.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━

I searched everywhere after finding this — Reddit,
Stack Overflow, GitHub, dev blogs — nobody has
documented this combination before.

Posting here before it potentially gets patched.

Confirmed working: March 2026