Best practices

The difference between a thin answer and a usable one is almost never the model. It is the brief. This page collects the habits that separate the two, in the order they matter.

None of it is long. If you read one section, read the first — briefing the matter properly is most of the technique, and every other habit here is a variation on it.

Brief the matter, not the task

Brief it the way you would brief a senior colleague, not the way you would type into a search box. Give it the facts, the posture and where you are stuck: who the client is, what the other side has done, what you are trying to achieve, and what you want back — a clause, a memo, a risk list, a reply. A thin prompt gets a thin answer, and it is the single most common reason a session disappoints.

The briefed version names the side you are on, the governing law, the stage of the negotiation and the decision you are trying to make. That is why it comes back as a negotiating position rather than a summary of the contract.

Worked examples by kind of work: AI Document Review, AI Drafting and the Legal AI overview.

Name the jurisdiction, every time

A legal question without a jurisdiction has no single right answer, so the AI gives you the general one. This is the most common reason a research result comes back generic, and it is the cheapest thing on this page to fix: close the prompt with the governing law.

Research is also the narrowest thing the engine does — a slice of what it is built for rather than the main event. Four things decide the answer: the governing law, the instrument or statute in play, the specific question, and the jurisdiction. Name them and the answer comes back as conditions to satisfy rather than a description of the area. AI Legal Research works this through with a real example.

One matter, one thread

Keep a matter in its own conversation while you are working on it, and open a new chat when you move to the next one. Each message you send is read in the context of the whole thread, so a conversation that has been running since January is carrying every unrelated question you asked in between — which makes the answers vaguer, not just slower.

A thread that has run all year is the most common reason an account behaves unpredictably. Starting a fresh chat per matter is the fix, and it costs nothing.

Attach what the question turns on

Context is not free, and it is not neutral either. Ten documents attached to a question that turns on one of them is ten documents' worth of reading, and nine documents' worth of things the answer can wander into. Point it at the clause, the contract or the file that actually matters.

• One contract, one question — attach the agreement, ask about the indemnity.
• A comparison — attach both versions and say which is yours.
• A data room — say what you are looking for before you attach it, not after.

Iterate in the thread instead of starting over

When an answer is close but not right, say what is wrong with it. Follow-up questions refine the output in place, and the AI already has the matter in front of it. Re-writing the original prompt from scratch in a new chat throws that away and usually lands somewhere else entirely.

1. Say what is wrong, specifically. “Too generic” gets you a different generic answer. “This assumes English law, redo it under UAE law” gets you the answer.
2. Ask for the change, not the document. “Tighten clause 8 to a twelve-month tail” beats “draft it again.”
3. Press on the reasoning. Asking why it reached a conclusion is a legitimate prompt, and on a delegated piece of work it is the one worth asking.

Start from the library when you can

The prompts that work have already been written down. The library holds several hundred, structured by use case, practice area, deliverable and complexity, and it is free to browse without an account — see the Prompt Library or browse it on the Academy.

There is no model to pick, and that is the point

Some AI tools open with a dropdown: fast model, smart model, cheap model. HAQQ does not, and it is deliberate. Which model is strongest is not a stable fact — it moves by task and it moves month to month as new ones are released. Putting that choice in front of a lawyer asks them to make an engineering decision before the work starts, with none of the information needed to make it well.

What it means for you is simple: moving up a plan buys volume, not better answers. Sizing a plan is a question of how much you expect to use it, not how good you need the output to be — Credits and usage says the same thing from the billing side, and which features a plan unlocks is a separate question covered on Subscription plans.

So there is no setting to tune for quality, and all of the leverage sits in the brief. The controls you do have — the default agent, reasoning mode, and what the AI is allowed to read — are on AI settings.

Getting more out of every prompt

Credits measure work, not words, and the habits above are the ones that decide how much work a question turns into. That is not a discount and it is not a trick: a well-briefed question gets a usable answer on the first pass, and a vague one sends you round three more times before it lands on the same place.

What actually moves usage

Complexity and volume. A short, well-scoped ask is a small piece of work. Drafting a full agreement is a larger one. Analysing a long contract, or several documents at once, is larger again — because the work genuinely is. Credits and usage covers the mechanics: how credits arrive on each plan, whether they roll over, and how a firm's shared pool works.

Why nobody can quote you a number

There is no fixed price per question, and any figure you are given would be wrong. The same question asked by two people, with different documents attached and a different matter behind it, is two different amounts of work. The honest guide is the shape of it: the more complex the ask and the more material attached to it, the larger the job.

Optimise the finished piece of work, not the message

This is the part worth internalising. The unit that matters is the deliverable that leaves your desk — the executed agreement, the advice you gave, the filing. A vague prompt is not the cheap option; it is the one that sends the work down the wrong path and gets redone. Get the brief right and everything downstream is shorter.

Read it before it leaves your desk

The AI gets you to a working draft in minutes instead of hours. The professional judgement on the way out the door is still yours, and nothing on this page changes that. Check the citations, check the clause numbers, check that the jurisdiction it answered on is the one you meant.

Turning on reasoning mode makes that check faster rather than slower: it exposes how the answer was reached, which is what turns an output into something a supervising partner can audit. It is worth leaving on wherever work is delegated — AI settings.

Where the detail lives