7. Stop Explaining Yourself Every Time
Twenty minutes of setup so you never retype your context again. Standing instructions, a project per recurring job, and a note called prompts.
The WJS Desk
Sep 3, 2026 · 6 min read
By now you have prompts that work. The problem is that tomorrow you will open a new chat and type all of it again from scratch, and the day after that you will not bother, and you will slide back to typing one vague line and being disappointed.
This lesson is about the setup that stops that happening. It is 20 minutes of work, once, and it is the difference between AI being a thing you occasionally try and a thing you actually use.
The problem, concretely
Every new conversation starts from zero. It does not know your job, your company, who you write to, that you spell things the American way, that you hate exclamation marks, or that "the report" means a specific weekly document. So every session you either re-explain all of it or accept output that ignores all of it.
There are 3 fixes, and they stack.
Fix 1: the standing instructions
Every major AI app now has a settings field for instructions that apply to every conversation. The name varies by product, some call it custom instructions, some personalization, some a profile. Find it once. It is in settings, usually within 2 clicks.
What to put in it. Be concrete and short, roughly 100 to 200 words:
I run a 6 person design studio. Clients are mostly
US based small businesses.
How I want you to write:
- Plain and direct. No corporate filler.
- US spelling.
- No exclamation marks.
- Never open with "I hope this finds you well."
- If I ask for an email, default to under 150 words
unless I say otherwise.
How I want you to behave:
- If a question depends on facts you do not have,
ask me rather than assuming.
- Flag anything in your answer I should verify.Those last 2 lines are worth more than the rest combined. They change the default behavior from confident guessing to asking, which is the single biggest quality improvement available in a settings box.
Write it as rules with reasons, not adjectives. "Be professional" does almost nothing, because it already thinks it is being professional. "No exclamation marks" is a rule it can follow. And where a rule has a reason, give it, the same way lesson 2 did: a reason generalizes to cases you did not list.
Fix 2: a project or space per recurring thing
Most AI products now have some version of a project: a container that holds a set of chats plus reference documents plus its own instructions. Names differ, the idea does not.
Set one up for anything you do more than twice a month. Ours would be:
- Client work. Holding the brand guidelines, a rate card, and 3 emails in the house voice.
- Admin. Holding the standard contract and the invoice template, so "does this new client contract differ from ours" is a 1 line question.
- Learning. Just a place where the instruction is "explain to someone with no background, always give a worked example".
The payoff is that the documents are already there. Lesson 4 was about pasting a document in. A project is that, permanently, so you stop pasting.
Fix 3: the file of prompts that work
This one costs nothing and needs no features, and it is the one we would keep if we had to drop the other 2.
Open a note. Title it "prompts". Every time something works well, paste it in with a 3 word label. That is the entire system.
MEETING NOTES → SUMMARY
Turn these raw notes into a summary for people who
were not there. Group by topic, pull decisions and
action items out separately. Do not add anything I
did not write.
CONTRACT CHECK
Answer only from this text. Say "not stated" rather
than guessing. Then: what would cost me money and
when, what are my obligations with deadlines, what
is unusual, what is NOT covered that I might assume is.
HARD EMAIL
Facts below. Voice: plain, direct, friendly, no
corporate filler. Under 150 words. Only apologize
once. Write 3 versions if the tone is unclear.Within a month you will have 10 to 15 of these and you will stop writing prompts from scratch for anything routine. Reuse is what turns the techniques in this course into a habit instead of a set of tricks you remember having read.
What to keep in one conversation, and when to start a new one
A single conversation carries everything said so far, which is why follow-ups work so well. That is also its failure mode.
| Keep going in the same chat | Start a new one |
|---|---|
| Refining a draft | Switching to an unrelated task |
| Follow-up questions on the same document | It has misunderstood twice in a row |
| Building on an answer you liked | You want a genuinely fresh opinion |
| Anything where earlier context helps | The chat has run very long and got vague |
The one people get wrong is row 2. When a conversation has gone off the rails, correcting it inside that conversation usually does not work, because everything it already said is still sitting in front of it, pulling it back. Start again with a better prompt. It costs 30 seconds and it works.
When the standing instructions get ignored
This happens and it is not a bug. Three reasons, in the order to check them:
- The rule was an adjective. "Be concise" is a mood. "Under 150 words unless I say otherwise" is a rule. Anything you cannot check by looking at the output is not going to hold.
- Your in-chat request contradicted it. What you type now beats what you saved once, which is correct behavior. If you ask for a detailed breakdown, you will get one, word limit or not.
- You wrote 600 words of instructions. A long list dilutes itself. Ten specific rules beat 40, and the 40-rule version is usually 10 real rules plus 30 restatements.
When a rule keeps slipping, put it back in the message for that one task. Standing instructions set the default. They do not override what you just said.
What to actually put in a project
The mistake is treating it as a folder and dumping everything in. What earns its place:
- The documents you would otherwise paste. Brand guidelines, the standard contract, the price list, last quarter's report.
- Three examples of your output in that context. Per lesson 6, this does more than any instruction you can write.
- The vocabulary. What "the report" means, what your product names are, which acronyms are yours. This kills a whole class of confusion.
- What not to do. "Never quote a price, always tell me to check the rate card." Rules with teeth.
What does not earn its place: anything out of date. A project quietly feeding it last year's prices is worse than no project, because you will stop checking.
A 20 minute setup, in order
- Minutes 1 to 5. Open settings, find the custom instructions field, write the 2 blocks from the template above: how to write, and how to behave. Save.
- Minutes 6 to 8. While you are in settings, find the privacy or data controls and decide about training. Once, then never again.
- Minutes 9 to 15. Create 1 project for the thing you do most. Drop in 2 documents and 3 examples of your own work.
- Minutes 16 to 20. Open a note called "prompts". Paste in the 3 above. Add a line at the top: "when something works, paste it here."
That last line is the whole system. The file only grows if adding to it is a 5 second habit, and it is only worth anything once it has 10 things in it.
One thing to decide once
If you are putting standing instructions and reference documents into a project, spend 5 minutes in the privacy settings. Most consumer AI products let you turn off training on your conversations, and business tiers generally do not train on your data at all. Decide once, then stop thinking about it.
Do this one now
Two things, about 20 minutes total. Write your standing instructions and save them. Then open a note called "prompts" and paste in the 3 above as a starting set.
That is the whole setup. Everything after this is just using it.
Next
The last lesson is the one that makes all of this safe to actually rely on: how to check its work, including the trick that caught the mistake in lesson 5.
