Why the prompt matters more than the platform, and how to build habits your whole team can use.

Generative AI has been part of in-house legal life for two to three years now. In that time, most of us picked up a handful of prompting habits, leaned on them for a while, and then let them drift. Prompting sessions are consistently among the most popular webinars we run at ITGC, and that told me the community was ready for a refresher.

So for our Back-to-School Webinar Series, I sat down with Gus Neate (Co-founder and CEO at Claren), a former Clifford Chance lawyer who works on this with in-house teams every day. Here's what came out of that workshop.

The real bottleneck sits in the prompt

When AI output misses the mark, the instinct is to blame the tool and start looking for a new one. Gus's starting point is simple: for most in-house teams, the biggest gains come from writing better prompts.

Lawyers are trained to draft carefully, and to draft with a lot of assumed context. A good prompt needs that context spelled out, with clear instructions and a clearly defined output. The discipline is the same one good legal drafting already demands.

“Prompting is a learnable skill, and you don't need a technical background to get good at it. Better still, the habits are portable. Once you've built them, they work in whichever tool you use today and whichever one comes next.”

Gus Neate - Co-Founder & CEO, Claren 

Seven habits that travel

These are the habits Gus walked through live on real contracts. They hold up in any platform.

1. Don't let the AI be lazy. Tell it to hit every single clause and give you a clause reference for each risk. Gus's advice is to think of the AI as a new junior on your team who hasn't done this task before. Ask them to be really thorough and the quality goes up. For legal work, go for the smarter model over the faster one. Start with the full output, and once you're confident it's covered everything, trim back to the points you care about.
   
   
   

2. Be precise. The AI only knows what you tell it. If you want clause numbers or the exact wording, ask for them. Tell it which side of the contract you're on. If you've just come off a negotiation call, drop your call notes in under the prompt, because the more context you give it, the more precise the answer gets. And give it the Word version if you have one, so it can work straight on the document.
   
   
   

3. Break the task down. Use one prompt to pull out the key information, a second to assess it and a third to redline or draft the email. Check each one before you move on. When the task is narrow, it's much easier to spot where things went off track. Once those smaller prompts are giving you consistent results, you can stack them into one bigger prompt or a playbook.
   
   
   

4. Avoid leading questions. Generic AI tools can be trained to be very agreeable, so if your prompt hints at the answer you want, you'll probably get it. Ask whether to accept or negotiate, get the strongest argument for each, and then ask for a recommendation.
   
   
   

5. Specify what good looks like. If you know you want a table with particular columns, put that in the prompt up front. The same goes for formats like dates and currency. If your team already reports risks in a set template, load it into the tool and ask for the output in that template. You'll spend far less time going back and forth.
   
   
   

6. Let it learn. Save your preferences into the tool so you're not typing the same instructions every time. After three good chats, ask the tool to build you a playbook or skill from how you worked. House style can go in there too, right down to how you sign off your emails. Eventually you can just say "review this contract" and the tool already knows how you like it done.
   
   
   

7. Steal from what works. Keep an eye on what your best users are doing and use their prompts. Start from your tool's prompt library if it has one, and hit the prompt improver when your draft is a bit rough. Save your winning prompts and share chat links around the team. Over time, those shared prompts become a library your whole team can draw on.

Why legal prompting needs extra care

Most generic prompting advice is written for marketers and generalists. The stakes in legal work are higher. A weak output in a marketing brief is an annoyance. A weak output in a contract or a regulatory filing has real consequences.

Context is essential every single time. A generic AI tool knows nothing about your playbooks, your jurisdiction or the commercial sensitivity of the deal in front of you until you tell it.

Completeness matters more than brevity. In most settings, shorter prompts work well. In legal work, if the output needs to flag specific clause types or apply specific standards, say so explicitly.

Your judgment stays in the loop. AI is a drafting aid. The decision, and the sign-off, stay with you.

Legal-specific tools give you a head start here, because your playbooks and standard positions can already be built into the platform. The prompting skill still matters. The tool carries your institutional knowledge, and you still have to tell it what to do with it.

The mistakes that come up most often

Gus sees the same handful of mistakes across in-house teams, and every one of them is fixable:

🔶 Assuming the AI already knows your context. Starting without context is the single biggest cause of outputs that miss the mark.

🔶 Asking for too much in one go. One clear task per prompt works best. Build from there.

🔶 Accepting the first output. The first response is a draft. The teams getting real value prompt again and push on the sections that matter.

🔶 Letting good prompts disappear. Every useful prompt that gets discarded is lost institutional value.

🔶 Using a generic tool for specialist legal work. Every session starts from zero when the tool has no access to your playbooks.

From one prompt to a team asset

Most legal teams use AI the same way every time: one person, one task, one prompt, starting from scratch. That approach works for individuals and stalls at team level.

A prompt becomes an agent when it turns into a documented, repeatable workflow. Picture an NDA review prompt that already contains your standard positions and your preferred output format, saved where everyone on the team can reach it. Building this requires no engineering support and no dedicated legal ops function. It requires the discipline to document what works and the habit of building on it.

A strong test for any prompt that's producing good results is whether a colleague could pick it up without explanation, a new hire could use it on day one, and it could run consistently across 50 contracts. A prompt that passes is ready to become a shared asset.

Every legal team also has one person who has quietly cracked AI and whose prompts live entirely in their head. That knowledge belongs in a shared prompt library, labelled and easy to find. A 15-minute monthly conversation where people share what they've tried builds more collective capability than a one-off training session. Rough prompts are worth sharing too, because a colleague can build on them.

A quick self-check for this week

Score your team honestly against these five statements:

1. We can name the AI tasks our team repeats every week.

2. Our best prompts are saved somewhere the whole team can find them.

3. A new hire could use our prompts on day one without asking for help.

4. We review and refine AI output before it goes anywhere.

5. We regularly talk as a team about what's working with AI.

If fewer than three of those ring true, start with habit seven and save what works.

The good news is that none of this needs a large budget or a transformation programme. It needs clearer prompts and a team habit of sharing what works. The written prompt pack from the workshop is available on request, and the full recording is available here: https://irishtechgc.com/blog#webinars.

This article was written in collaboration with Claren, an AI platform built for in-house legal teams. Learn more at claren.law