Welcome back to Legal Prompting, I'm Nicola Fabiano, and this is episode 10, the last of the first season.
Download transcript (.vtt)
0:00
0:08
0:16
0:24
0:31
0:38
0:54
1:01
1:07
1:14
1:22
1:30
1:37
1:45
1:54
2:01
2:09
2:16
2:24
2:32
2:38
2:47
2:53
3:01
3:11
3:19
3:27
3:34
3:41
3:48
3:53
3:58
4:03
4:09
4:17
4:25
4:30
4:37
In the previous episode, we addressed the AI Act and the obligations it imposes on legal professionals.
Transparency, human oversight, risk classification, responsibility in deploying AI systems in legal practice.
Today we close the loop. We will synthesize the method we have built episode after episode,
and we will try to look ahead. The central principle of this season is simple, and we have
repeated it in different forms. Legal prompting is not a trick, it is a method. It is the disciplined
Do not replace professional judgment, articulate it into verifiable instructions.
Everything else, the techniques, the use cases, the operational safeguards, flows from here.
Let us revisit three threads that have run through the season. The first concerns prompt quality.
We have seen, from the analysis of supervisory authority decisions to drafting privacy notices,
from contract review to using compliance workflows, that the prompt must declare role,
context, instructions, expected format, and constraints. A vague prompt produces a vague
output. A structured prompt is already half of the legal work. The second thread is verification.
No model output enters a brief, an opinion, or a corporate procedure without human review.
We have talked about regulatory hallucinations, incorrect cross-references, summaries that lose
argumentative nuance. Verification is not optional. It is the condition that makes the use of AI in
legal work sustainable. The third thread is infrastructure. The choice of tool is a legal
choice. Where the data travels, who processes it, with what guarantees, under which contractual
basis, consistent with which obligations, these questions precede use, they do not follow it.
Cloud or local, generalist or vertical, with retrieval or without. The technical decision
is inseparable from professional secrecy, GDPR, AI Act, and Italian Law 132 of 2025.
To these threads, we add three operational rules that apply across the board.
First rule, document. Keep prompts, versions, revisions. Traceability is the foundation of
responsibility. Second rule, update. Models change, rules evolve, techniques mature,
the method is stable, the tools are not. Third rule, share with caution. Do not put into a
prompt what you would not write in an unencrypted email. Professional secrecy travels with the data,
it does not stay in the office. Looking ahead, some directions are already visible.
Local models are becoming practicable for law firms and legal offices, with meaningful
advantages for confidentiality. Specialized tools for the legal domain will grow,
but the ability to evaluate them critically remains the professional's task.
The regulatory framework will consolidate in the coming months,
with authorities called to translate the AI Act into practice.
One underlying conviction remains, AI does not replace the lawyer. It does, however,
force the lawyer to make explicit what often remains implicit, premises, reasoning, sources,
evaluation criteria. In this sense, legal prompting is also a pedagogy of legal method.
We close the first season of Legal Prompting here, thank you for following it.
To stay in touch, subscribe to the newsletter at nickfab.eu, that is where the deeper discussions
continue between one season and the next. See you soon!