The Repetition Tax
Every time you open a new AI chat and type "I run a small marketing agency" or "I'm a freelance designer working with Figma" — you are paying a tax. A context tax. It costs you time, it costs you output quality, and it guarantees the first response will be generic because the AI is still catching up to who you are.
Custom instructions eliminate this entirely. They are a block of persistent context that loads before every conversation. Think of it as a briefing document your AI reads before you walk in the room.
The insight: Custom instructions are not a feature. They are an unfair advantage. Someone with good custom instructions gets better output from the same model, the same prompt, the same everything — because the AI already knows what matters.
Where Custom Instructions Live
Every major AI tool now supports persistent context. Here is where to find it:
- ChatGPT: Settings → Personalization → Custom Instructions (two fields: about you, and how it should respond)
- Claude: Settings → Profile → Custom Instructions (or use Projects for per-project context)
- Gemini: Settings → Extensions → Saved Info
- Any API tool: System prompt field — same concept, full control
The 4-Block Custom Instructions Template
Fill in these four blocks and paste them into your custom instructions. This template works on every platform.
Block 1 — Identity
Who you are, what you do, what stage you are at.
I run a 3-person digital marketing agency. We focus on
e-commerce brands doing $500K–$5M/year. I handle strategy
and client communication. My team handles execution.
I have been in marketing for 8 years but I am relatively
new to using AI in my workflow.Block 2 — Tools and Stack
What you actually use day to day so AI can give relevant advice.
Tools I use: Notion for project management, Figma for
design handoffs, Slack for team communication, Google
Workspace for docs and sheets, Shopify for client stores,
Meta Ads Manager and Google Ads for paid campaigns.Block 3 — Communication Preferences
How you want AI to talk to you. This shapes every response.
Be direct. No filler, no "Great question!" openers.
Use short paragraphs. When I ask for options, give me 3
with a clear recommendation. Default to practical over
theoretical. If something is risky, say so plainly.
Use bullet points for lists. Bold key terms.Block 4 — Anti-Patterns
What you explicitly do not want. This is surprisingly powerful.
Do not use corporate jargon like "leverage," "synergize,"
or "circle back." Do not give me 10 options when I asked
for 3. Do not hedge everything — if you think an idea is
bad, say it is bad and tell me why. Do not write emails
longer than 150 words unless I specifically ask.Set Yours Up Right Now
Open your AI tool of choice. Find the custom instructions or system prompt setting. Fill in these four blocks with your real information:
- Identity: Who you are, what you do, what stage
- Tools: What you use daily
- Communication: How you want AI to respond
- Anti-patterns: What to never do
Then start a new chat and ask something you have asked before. Notice the difference immediately.
Advanced Patterns
Role-Specific Presets
If you wear multiple hats — founder, marketer, hiring manager — create different instruction sets and swap them based on what you are working on. ChatGPT lets you toggle between sets. Claude Projects give you per-project instructions automatically.
// Hiring mode
I am currently hiring a junior media buyer. When I discuss
candidates, evaluate based on: portfolio quality, platform
experience (Meta preferred), communication skills, and
culture fit. I value self-starters who can work async.
// Client mode
I am preparing a quarterly review for a client doing $1.2M
on Shopify. They care about ROAS, new customer acquisition
cost, and email revenue percentage. Frame everything in
terms of these KPIs.Output Templates
If you repeatedly need the same format, bake it into your instructions so you never have to specify it again.
When I ask you to write a client update email, always use
this structure:
1. One-line status summary
2. What was completed this week (bullets)
3. What is planned next week (bullets)
4. Any blockers or decisions needed from them
5. Sign off with "Let me know if you have questions."
Keep it under 200 words.Knowledge Anchors
Pin specific facts so AI does not make assumptions or hallucinate details about your business.
Facts about my business:
- We charge $3,000–$8,000/month per client
- Our average client retention is 14 months
- We do not offer SEO services
- Our target market is DTC e-commerce, not B2B
- We are based in Austin, TX (CST timezone)
Never suggest SEO as a service offering. If a question
involves pricing, use my actual ranges, not generic ones.The Compound Effect
Good custom instructions do not just save you time per conversation. They change the baseline quality of every interaction. Over weeks and months, the compound effect is significant:
- Fewer corrections: AI stops suggesting tools you do not use, formats you do not like, and approaches that do not fit
- Faster drafts: First outputs are closer to final because the context is already loaded
- Better recommendations: AI can factor in your actual constraints instead of giving generic advice
- Less fatigue: You stop fighting the AI and start collaborating with it
Think of it this way: Custom instructions are the difference between hiring a stranger every morning and working with someone who already knows your business. Same person. Same skills. Wildly different output because of context.
Common Mistakes
- Too vague:"I like concise responses" tells AI nothing. "Keep paragraphs under 3 sentences, use bullets for lists, bold key terms" is specific and actionable.
- Too long: Custom instructions have token limits. Be dense, not verbose. Every sentence should carry weight.
- Set and forget: Update your instructions as your role, tools, or priorities change. A quarterly review takes 10 minutes and keeps everything current.
- Contradictory rules:"Be detailed" plus "be brief" confuses the model. Pick a default and specify exceptions.
Get new lessons free
We publish free AI lessons weekly. Drop your email and we will send them directly — no spam, no sales sequences, just signal.
What Comes Next
You now have three foundational skills: writing effective prompts (R-C-O), building a daily AI workflow, and setting up persistent context. Together, these three things put you ahead of 90% of people using AI.
The next level is connecting multiple AI models and tools into a single workflow. That is where efficiency becomes real leverage — and it is exactly what Premium covers in Multi-Model Workflows.