The Best ChatGPT Custom Instructions Template (2026)

You ask ChatGPT how to handle a difficult conversation with your manager. It gives you a five-bullet framework that reads like it was pulled from a 2012 HR manual.

You ask for a productivity system. It recommends time-blocking. Again.

You ask for career advice. It tells you to "network strategically and build transferable skills."

None of it is wrong, exactly. It's just useless — because it has nothing to do with you.

That's the real problem with default ChatGPT. It's not bad at thinking. It's blind to context. It doesn't know your personality, your communication style, how you make decisions, what's already failed for you, or what kind of advice actually lands. So it gives you the statistical average of all possible answers, hedged into meaninglessness.

Custom instructions fix this. Here's exactly how to use them — and a template that actually works.

What custom instructions actually do

Custom instructions are a persistent text block that gets prepended to every new ChatGPT conversation. They live in Settings → Personalization → Custom Instructions, and you fill out two fields:

1. "What would you like ChatGPT to know about you?" — background context, personality, situation
2. "How would you like ChatGPT to respond?" — tone, format, depth, communication preferences

Every time you start a new chat, ChatGPT reads this context before it reads your message. It's the closest thing to a standing briefing you can give an AI.

A few things worth knowing about the mechanics:

• Token limit: Custom instructions count toward your context window. Keep the total under ~1,500 words. Ruthless prioritization matters — everything in there should earn its place.
• They don't auto-update: Whatever you write stays until you change it. If your situation changes significantly, update your instructions.
• New conversations only: Instructions don't retroactively apply to existing chats. Start a fresh conversation after updating them.
• ChatGPT Memory is separate: Memory is dynamic and builds over time. Custom instructions are static and intentional. Both are useful; they do different things.

The goal isn't to give ChatGPT more text. The goal is to give it the right text — the kind that changes how it thinks about your problems.

The custom instructions template

Here's a structured template you can copy, fill in, and paste directly into your custom instructions. The sections are ordered by impact — don't skip the first two.

Field 1: What ChatGPT should know about you

## About Me

**Personality & Thinking Style**
- [Describe how you process information — big picture first or details first?
  Analytical and data-driven, or values-based and intuitive?]
- [How do you handle ambiguity — comfortable with uncertainty or need clarity before acting?]
- [What's your default mode under stress — withdraw, push harder, seek input?]

**Communication Style**
- [How direct are you? How direct do you want others to be with you?]
- [Do you prefer concise answers or thorough explanations?]
- [What frustrates you in communication — hedging, over-qualification, vague advice?]

**Decision-Making**
- [Do you decide quickly and iterate, or research extensively before committing?]
- [What makes advice actually useful to you — logic/data, or values alignment and human impact?]
- [Known blind spots or patterns you want challenged: e.g., "I overthink" / "I avoid conflict"]

## My Context

**Current Situation**
- [Role, field, major life situation — the minimum context needed to give relevant advice]
- [What you're actively working on or trying to figure out]

**Constraints That Matter**
- [Time, budget, energy, family situation — whatever limits what's actually actionable for you]

**What Has Failed Before**
- [Approaches, systems, or types of advice that don't work for you, so ChatGPT stops recommending them]

Field 2: How you want responses

## Response Preferences

**Tone**
- [Direct / collaborative / encouraging — pick what actually helps you]
- [Push back when I'm off-track vs. support my direction — which do you want?]

**Format**
- [Bullet points vs. prose vs. tables vs. depends on the question]
- [Short and scannable vs. thorough and explained]
- [Lead with the answer, or build up context first?]

**Depth**
- [When should ChatGPT go deep vs. give a quick take?]
- [Do you want caveats and nuance, or just the most likely answer?]

**Don'ts**
- [Specific things to avoid: excessive hedging, motivational language, generic frameworks,
  unsolicited life advice, excessive bullet points — whatever grinds your gears]

Filled-in example

## About Me

**Personality & Thinking Style**
- Big picture first — I need the concept before the steps. Details without context lose me.
- Intuitive decision-maker. I use data to pressure-test decisions, not to make them.
- Under stress I go quiet and avoid. Flag it if you see avoidance creeping into my reasoning.

**Communication Style**
- Blunt is better. I find over-qualified answers condescending.
- Concise by default. Go deep only if the question requires it.
- I hate motivational padding. No "great question!" or "that's a really important point."

**Decision-Making**
- I decide fast and course-correct later. Help me think through second-order consequences
  I might be glossing over, but don't slow me down with excessive research checklists.
- Known pattern: I overweight novelty. If I'm choosing something just because it's new,
  call it out.

## My Context

**Current Situation**
- Independent consultant, strategy and operations. Work with early-stage startups.
- Currently building a side project in the evenings; limited to ~8 hours/week on it.

**Constraints That Matter**
- No budget for paid tools right now. Evenings only for side project work.
- Two kids under 5 — solutions that require sustained daily effort rarely stick.

**What Has Failed Before**
- Rigid daily routines. They work for two weeks and then collapse.
- Anything requiring extensive upfront setup before getting value.

## Response Preferences

**Tone**
- Direct. Push back if my framing seems off, but don't lecture.
- I want honest assessments, not cheerleading.

**Format**
- Lead with the answer. Background context after, not before.
- Bullets for lists, prose for analysis. No bullet points for things that aren't actually lists.
- Skip the summary paragraph at the end — I just read what you wrote.

**Depth**
- Quick take by default. Flag when a question deserves more depth.
- Skip caveats unless they're actually load-bearing.

**Don'ts**
- No generic frameworks (SMART goals, 80/20, etc.) unless they're actually the right tool.
- No "it depends" as a final answer. Give me your best read and caveat it lightly.
- Don't recommend productivity systems designed for highly disciplined people.

This is the kind of context that changes everything. Not because it's clever — because it's specific.

The problem with writing your own instructions

Here's where most people get stuck: they sit down to write their custom instructions and don't know what to put.

Not because they're bad at writing. Because self-knowledge is genuinely hard.

You've been living in your own head your whole life, which paradoxically makes you one of the worst observers of your own patterns. Confirmation bias filters out the feedback that would sharpen your self-description. You highlight the version of yourself you aspire to, not the one that shows up under pressure. You forget the failed systems and remember only why they failed in theory.

So the instructions you write from scratch tend to be:

• Too surface-level. "I'm detail-oriented and like clear communication." Cool — so does 60% of the population.
• Too aspirational. "I prefer to research thoroughly before making decisions." Do you actually, or do you wish you did?
• Missing the nuance that matters. The difference between "I'm introverted" and "I'm introverted but I perform well in meetings when the agenda is clear in advance" is the difference between advice that's marginally relevant and advice that's actually useful.
• Blind to the patterns you can't see. You don't know your blind spots are blind spots. That's the whole point.

The result: custom instructions that technically describe someone, but not very accurately, and not in a way that meaningfully changes AI output.

The better approach: personality-science-backed instructions

Instead of guessing at your own personality, you can measure it.

This is what personality science was built for. Validated frameworks like the Big Five — the most research-supported model of personality structure — map where you actually fall on traits like conscientiousness, openness, extraversion, agreeableness, and emotional stability. Not where you think you fall. Not your ideal self. Where you actually are, based on how you respond to carefully designed questions.

That data produces a very different kind of custom instruction — one grounded in how you actually think, decide, and communicate, rather than how you imagine you do.

InnerForge's Forge Blueprint is built on exactly this principle. You complete a series of science-backed personality quests — short, structured assessments that map your traits across multiple dimensions — and the platform generates a custom instructions block ready to paste directly into ChatGPT, Claude, or Gemini.

No guessing. No aspirational self-description. No vague paragraphs that apply to half the population. A precise, AI-optimized personality profile based on your actual data.

The difference between hand-written instructions and a personality-science-backed blueprint is the difference between telling ChatGPT you're a good driver and handing it your actual driving record.

Before and after: what this looks like in practice

The prompt: "I keep starting new projects and not finishing them. What should I do?"

Without custom instructions:

This is a common challenge. Here are some strategies to help you follow through:

1. Break projects into smaller milestones
2. Set clear deadlines for each phase
3. Use accountability partners to stay on track
4. Identify your "why" to stay motivated
5. Eliminate distractions and create a dedicated workspace

Generic. Inoffensive. Applicable to no one in particular.

With a personality-science-backed blueprint (high openness, low conscientiousness, intuitive decision-maker, pattern of overweighting novelty):

The pattern you're describing is consistent with high openness combined with lower conscientiousness — you're genuinely energized by starting things, which means finishing has to compete with the pull of whatever new thing just became interesting. Discipline-based solutions (stricter deadlines, accountability partners) tend to fail because they treat this as a motivation problem. It's not — it's a wiring issue.

What tends to work better: ruthlessly shrink the scope of each project before you start, so "done" arrives before novelty fades. Give yourself one active project with a hard rule — no new starts until the current one ships, even badly. And flag honestly: is this project actually worth finishing, or are you keeping it alive out of sunk cost? Permission to kill it cleanly might be more useful than a system to finish it.

Same question. Completely different answer — because the AI knows who it's talking to.

Works with Claude and Gemini too

The same blueprint that works in ChatGPT works everywhere.

Claude (Anthropic): Go to Settings → Profile and paste your blueprint into the personal description field. Claude tends to be particularly good at adapting tone and reasoning style to personality context — worth testing.

Gemini (Google): Go to Settings → Extensions & Personalization and add your blueprint to the personalization instructions. Gemini's workspace integrations make personality-aware instructions especially valuable if you're using it for professional work.

Any API or tool: If you're using Cursor, Raycast AI, custom agents, or building with the API directly, paste the blueprint at the top of your system prompt. It works anywhere that accepts custom instructions.

One blueprint. Every AI tool you use. The investment in getting it right pays off compounding.

---

Get your Forge Blueprint free — it takes 5 minutes and works with ChatGPT, Claude, and Gemini. Complete a personality quest, get a blueprint built on real data, and paste it wherever you use AI. You'll notice the difference in the first conversation.

Start your free Forge Blueprint — no guessing required.