How to brief an AI writing tool so the first draft is usable

By Thomas Lovas Lokoy · Published 16 October 2026 · 6 min read

The difference between a draft you can publish and a draft you delete is usually decided before the model writes a word. A tool like AI Orbit asks for very little, a topic, a voice, a length, a reader stage, and that is exactly why each input carries so much weight. This guide goes through them one at a time and shows what a weak brief and a strong brief look like for each.

The topic is a sentence, not a keyword

A topic field that accepts up to 300 characters is an invitation to write a sentence, and the sentence should contain the thing the article will argue or answer. Compare:

• Weak: "solar batteries"
• Better: "solar battery payback in a UK semi-detached house"
• Strong: "Why a home solar battery rarely pays for itself in the UK unless you are on a time-of-use tariff"

The weak version leaves the model to pick the angle, and it will pick the most generic one available. The strong version has already made the editorial decision: there is a claim, a place, and a condition. Everything downstream, the research, the headings, the examples, is now constrained to that claim. If you find yourself unable to write the strong version, you do not yet know what the article is for, and the model cannot know it for you.

Two habits help. Write the topic as the headline you would be happy to publish. And include the one word that signals the reader's situation (a country, a budget band, a job title), because that word is what the research step will search for.

The audience note is where you put what the topic cannot hold

Series and batch runs in AI Orbit accept an audience or niche note of up to 500 characters. Use it for context that would make the topic sentence clumsy: who the reader is, what they already know, what they are not interested in. "Small UK landlords with one or two properties, already familiar with Section 21, not interested in commercial property" does more for a draft than another 200 words of topic. For a single article, the same information belongs in the topic sentence itself, so keep the sentence long enough to carry it.

Length is a commission, so ask for what you need

The length slider runs from 300 words to your plan's ceiling in steps of 100. The writer is instructed to treat the figure as a commission: anything under 90 per cent of it has not met the brief, and it should plan roughly one section per 350 words. That has a practical consequence. If you ask for 2,000 words on a topic that honestly supports 900, the model is told to add substance rather than pad, and the substance it reaches for will be examples and counter-arguments you did not ask for. Decide the length from the question, not from a target.

A rough rule that holds up: a single how-to question is 800 to 1,200 words; a comparison with real criteria is 1,200 to 2,000; a piece that has to carry someone from "what is this" to "which one" is a series, not one long article. The series tutorial explains how a seed topic becomes three to five linked pieces.

The reader's stage changes the structure, not just the tone

On Starter and above you choose a funnel position. This is the input most people leave at the default, and it is the one that most changes the shape of the draft, because each stage has its own four-part structure in the writer's brief:

• Awareness. The reader has the problem but not the vocabulary. The piece opens on the symptom as they would describe it, names what is going on, lays out the two or three broad approaches fairly, and closes with what to work out next. The close is a next read, never a purchase.
• Consideration. The reader knows the options and is weighing them. The piece names the decision in one sentence, gives three to five criteria, works each option against them honestly, and ends with a "choose X if…, choose Y if…" split rather than a single winner.
• Decision. The reader has decided in principle. The piece confirms the decision is sound, gives the steps in order with what each costs, answers the sceptic's objections including when not to proceed, and ends with the first action to take today.

Pick the stage by asking what the reader has already decided when they arrive. If they searched "what is", it is awareness. If they searched "vs" or "best", it is consideration. If they searched "how to set up", it is decision. The reader never sees those labels; the brief forbids them from appearing in the text.

The call to action is optional and should usually be specific

The optional "what should the reader do next" field takes up to 200 characters and is passed to the writer as the desired next step. "Sign up" is a poor instruction; "book a 15-minute call to check eligibility before applying" is a good one, because the writer can build the closing section towards it. On an awareness piece, make the call to action a next read, because the structure will not sell for you and should not.

The voice is a decision, not a decoration

Five personas are available on every account, and the voice can be switched off entirely for a neutral editorial register. Choose by destination. A persona with a strong point of view suits an opinion column and a newsletter; it does not suit a help-centre article or a client's corporate blog, where the neutral register is the right call. Switching the voice off is not a downgrade. It is the correct brief for most commercial writing.

Research on or off

On Pro and above, keyword research is on by default. Leave it on when the article is meant to be found through search, because the brief it produces, one primary phrase, supporting phrases, the questions people actually ask, and the gaps in what currently ranks, is folded into the writer's instructions. Switch it off for pieces nobody will search for: an internal explainer, a newsletter issue, a draft you are going to rewrite heavily anyway. Research adds minutes to the run and gains you nothing on those.

You can also preview the brief before you spend a credit. If the primary phrase that comes back is not the article you meant to write, fix the topic sentence rather than fighting the draft.

A worked example

Weak brief: topic "credit cards for students", default everything.

Strong brief: topic "Which features actually matter on a first credit card for a UK student, and which are marketing", stage Consideration, length 1,400, voice off, research on, call to action "compare the three cards named in the article on the issuer's own site before applying".

The second brief has made six decisions the first left to the model. Every one of them is cheap to make and expensive to undo.

Before you press Generate

Read your topic sentence back as a headline. Check the stage matches what the reader has already decided. Set the length from the question. Then let it run, and when it finishes, check the draft against the brief rather than against your hopes. The pipeline tutorial shows exactly how each of these inputs reaches the writer, and the pricing page shows which inputs your plan unlocks.