Six brief fields, without which any assistant produces a template
Two people open the same assistant with the same task: the first gets a document ready to use, the second a set of generalities. The difference is not luck but six fields one supplied and the other did not. Below are all six and which one weighs most.
Two people open the same assistant with the same task: the first gets a document ready to use, the second a set of generalities. The difference is not luck but six fields one supplied and the other did not. Below are all six and which one weighs most.
---
The same tool, different results
Two people open the same ready assistant. They set what is essentially the same task.
The first gets a document that can go straight into work. The second gets a set of generalities that would suit anybody.
The second concludes: "the assistant is mediocre". The first concludes nothing, because it works for them.
The difference is not the assistant and not luck. It is that the first supplied inputs and the second supplied a task.
Below are six fields usually missing. They are universal: they work with any assistant and on any task.
Why a template comes out without them
The mechanics are simple and worth understanding.
A model cannot produce specifics it does not have. You asked for landing page copy and did not say who for. It does not know your audience — so it writes for an average one.
The average is the template. Not because the model is lazy but because it is the only thing that can be built with no direction.
Second: a good assistant asks clarifying questions itself. But it asks them by its own frame rather than by your situation, and if you answer in monosyllables it will still fill in the rest itself.
The practical conclusion: it is quicker to supply the inputs at once than to gather them through conversation.
The six fields
1. Who for
Not "for an audience" but specifically: who these people are, what they already know, what they have already tried, what stage they are at.
Bad: "for entrepreneurs".
Good: "owners of car repair shops in towns under half a million, no website, never done any marketing, believe customers turn up by themselves".
What that changes: the assistant stops explaining basics to people who know them, and vice versa.
Where to get it: from your audience research, from reviews, from the questions you get asked most often.
2. What the person should do afterwards
One concrete action. Not "become interested" but "leave an enquiry", "send a message", "book a meeting".
Without this field the material ends in nothing — and that is the most common reason good copy does not work.
3. What must not be said
The most underrated field and the strongest.
A list of prohibitions: which words not to use, which promises not to make, which subjects not to touch, which tone to avoid.
Prohibitions work harder than wishes. "Write it well" produces a mediocre result. "Do not use 'unique', 'in today's world', do not open with general reflections, do not promise sales growth" produces something else entirely.
The compulsory minimum for commercial copy: do not promise income, do not promise a business result, do not make medical or financial claims.
4. In what voice
Not "professionally" — that means nothing. Give an example.
The fastest way: attach two of your own texts and write "write like this". Showing a voice is easier than describing one.
What to specify in words if you have no examples: sentence length, formal or informal address, whether irony is acceptable, favourite turns of phrase, what never appears in your speech.
5. What already exists
The material to work from: your data, your examples, your price list, your documents.
This is the key field for reliability. When all the material has been supplied by you, there is nothing to invent — the safest mode of working there is.
The request "here is a document, go through it against these points" is more reliable than "tell me what there is".
6. What has to come out
Format and length. A document, a list, a table, text of a given length, how many sections.
Without this field you will get whatever the model finds convenient, and something different every time. For processes where the result feeds onwards, that breaks the next link.
How this looks in practice
The same task, two approaches.
The usual way: "Write a post about our services".
What comes out: smooth text for nobody, listing advantages and inviting people to get in touch.
With six fields:
> Who for: owners of car repair shops, no website, believe they do not need one. The real pain is customers ringing with the same question about prices.
>
> The action: send the word "PRICE" as a message.
>
> Not allowed: the words "unique", "innovative", "in today's world". Do not promise sales growth. Do not open with general reflections.
>
> Voice: short sentences, formal address, no irony. An example of my own writing attached.
>
> Material: here are ten real customer questions.
>
> Format: up to 900 characters, short paragraphs, no list of advantages.
The second version is several times longer. But it gets written once and reused from then on: only the material changes.
A stored brief instead of every time
Hence a practical technique that saves the most.
Assemble the brief once and keep it. The first four fields — who for, the action, the prohibitions, the voice — rarely change for you. The material and the format get substituted per task.
From then on every new request begins by pasting that stored brief in.
If you work with clients, one brief per client. A single document per client: their audience, their voice, their prohibitions. A second job for the same client costs you several times less than the first.
And that is also what turns into a knowledge base if you build an assistant for a company.
How to check the inputs are enough
A simple test before sending.
Read your request and ask: could a competitor substitute their own name in here and have it all remain true?
If yes, the inputs are insufficient and the result will be a template. Not because the request is bad but because there is nothing of yours in it.
A second test: does the request contain a single one of the six fields that others definitely do not have? Usually it is the material — your data, your examples, your clients' questions.
What is available on the platform
More than 140 ready assistants across 14 categories. Each has its own frame and its own clarifying questions — which is exactly what takes some of the input work off you. But the material and the prohibitions are still yours.
Refinement modes: assessing the work, harsh assessment, focus group, criteria-based refinement. A useful technique is to submit the first result for a harsh assessment and see what the model itself thinks was missing. Usually the missing field turns up there.
Prompt Master in Gen AI — the "improve prompt" button rewrites the phrasing to a methodology. It does not replace inputs but it helps with form.
"Freelancer positioning" — if your first field, who for, is not filled in at all: a chain of seven assistants producing a niche and a client avatar. $9, included in Basic.
AI Workflow — if a request repeats, it becomes a chain step: the inputs get set once and only the material changes.
Registration is free and opens three days of full Basic access.
Where to start today
One action. Take your last poor result — the one that came out as a template — and check your request against the six fields.
Usually two are filled in: what has to be done and in what format. The model built the other four itself.
Add the third field, the list of prohibitions. That one field alone usually makes the most visible difference, and filling it in takes a minute.
---
Where the phrasing itself comes from
Six fields is what to supply. A separate question is how to phrase it so the model understands.
Module 3, "Basic prompting", has two lessons on that.
Tone of voice — how to set a voice so the result sounds like you rather than like averaged text.
Role prompts — how to assign a model a role so it holds it to the end rather than sliding into a general tone after three paragraphs.
And the module's fourth lesson is cascading prompts. It explains what to do when six fields are filled in and the task is still too large: split it into a chain where each step solves its own sub-question and passes the result on.
The phrasing from the lesson: you lead the model along a clear plan, like a scriptwriter.
Module 3 — the Basic plan.
---
What to read next
[One assistant against three](/en/blog/one-assistant-vs-three) — what to do when one is not enough.
[What prompt engineering is](/en/blog/what-is-prompt-engineering) — the fundamentals of setting a task.
[How Claude differs from other models](/en/blog/claude-vs-other-models) — why structure works.
[How to build your own AI assistant](/en/blog/build-your-own-assistant) — how to fix the fields once.
[AI agents: what they are and what people pay](/en/blog/ai-agents-guide) — the full guide to agents, with market rates.