AGENCYSCRIPT
CoursesEnterpriseBlog
đź‘‘FoundersSign inJoin Waitlist
AGENCYSCRIPT

Governed Certification Framework

The operating system for AI-enabled agency building. Certify judgment under constraint. Standards over scale. Governance over shortcuts.

Stay informed

Governance updates, certification insights, and industry standards.

Products

  • Platform
  • Certification
  • Launch Program
  • Vault
  • The Book

Certification

  • Foundation (AS-F)
  • Operator (AS-O)
  • Architect (AS-A)
  • Principal (AS-P)

Resources

  • Blog
  • Verify Credential
  • Enterprise
  • Partners
  • Pricing

Company

  • About
  • Contact
  • Careers
  • Press
© 2026 Agency Script, Inc.·
Privacy PolicyTerms of ServiceCertification AgreementSecurity

Standards over scale. Judgment over volume. Governance over shortcuts.

On This Page

Scenario One: A Privacy Policy UpdateThe weak promptThe stronger promptScenario Two: A Vendor Contract ClauseThe weak promptThe stronger promptScenario Three: A Consumer DisclosureThe weak promptThe stronger promptWhy naming the standard workedScenario Four: A Multi-Jurisdiction QuestionThe weak promptThe stronger promptScenario Five: A Regulatory ResponseThe weak promptThe stronger promptThe lesson on invented factsWhat the Examples ShareGrounding plus stated context wins every timeThe reviewer's job gets easierBuilding Your Own Examples LibrarySave the prompts that workedLearn from the failures tooTurning examples into templatesFrequently Asked QuestionsAre these real cases with real numbers?What single change improved the most drafts?How did naming the plain-language standard help?Why not let the model average multiple jurisdictions?What made the regulatory response safe?Can I reuse these prompt patterns directly?Key Takeaways
Home/Blog/Concrete Legal Prompts and What Separated Good Drafts From Bad
General

Concrete Legal Prompts and What Separated Good Drafts From Bad

A

Agency Script Editorial

Editorial Team

·August 9, 2020·8 min read
prompting for legal and compliance writingprompting for legal and compliance writing examplesprompting for legal and compliance writing guideprompt engineering

Principles are easy to nod along to and hard to apply when you are staring at a real prompt box. This article works through specific scenarios in legal and compliance writing, showing the actual prompt choices that produced a usable draft or a dangerous one. The scenarios are composites drawn from common situations, not invented statistics, and each one isolates a single lesson you can carry into your own work.

For each example we describe the task, the weak version of the prompt and why it failed, and the stronger version and why it worked. The contrast is where the learning lives. A small change in how you frame a legal prompt often makes the difference between a draft a reviewer can refine in minutes and one they have to throw out.

Read these alongside the structured process they illustrate, and you will start to see the pattern: grounding, context, and honest uncertainty turn a model from a liability into a fast, safe first-drafter.

Scenario One: A Privacy Policy Update

The weak prompt

A team asked the model, "Write a privacy policy section about data retention that complies with privacy law." The model produced a confident, generic section citing requirements it pulled from training. It read well and was unanchored to any actual rule or to the client's real practices.

The stronger prompt

The improved version pasted the applicable regulation and the client's actual retention practices, then instructed: "Draft a retention section using only this regulation and these practices. Quote the controlling provision before applying it. Flag anything the materials do not cover." The result was anchored, honest about gaps, and ready for review. The grounding discipline this relies on is covered in Everything That Matters When You Prompt for Legal Writing.

Scenario Two: A Vendor Contract Clause

The weak prompt

"Draft a limitation of liability clause" with no further context. The model produced a clause that read professionally but used "should" where the parties needed "shall" and omitted a carve-out the client always requires.

The stronger prompt

The team supplied the client's standard clause as a template and instructed the model to adapt it for the new vendor, preserving all operative terms exactly and flagging any deviation. The draft kept the required carve-out and the correct operative language because the prompt protected both. This is the operative-language failure described in Seven Prompting Habits That Sink Legal and Compliance Drafts.

Scenario Three: A Consumer Disclosure

The weak prompt

"Write a clear disclosure about our cancellation policy." The model returned dense, lawyerly prose that technically described the policy but would likely fail a plain-language standard.

The stronger prompt

The improved prompt named the standard: "This is a consumer disclosure that must be plain and conspicuous, readable at a general audience level. Use short sentences and common words. Avoid defined-term jargon unless necessary." The model produced genuinely plain language because the comprehension standard was stated rather than assumed.

Why naming the standard worked

The model's default register is dense, so plain language has to be requested explicitly. Naming the standard and the audience level gave the model a concrete target instead of leaving it to guess what "clear" meant.

Scenario Four: A Multi-Jurisdiction Question

The weak prompt

A team asked the model to draft an employment policy "that works across our offices," listing several locations only in passing. The model produced a single policy that blended the jurisdictions into requirements that matched none of them precisely.

The stronger prompt

The stronger version named each jurisdiction explicitly and instructed: "Where requirements differ by jurisdiction, do not average them. Surface the differences and mark which provisions depend on location." The model produced a base policy plus a clear list of jurisdiction-specific variations for legal review.

  • State every jurisdiction by name, not in passing.
  • Forbid averaging away differences.
  • Ask for location-dependent provisions to be marked.

Scenario Five: A Regulatory Response

The weak prompt

"Help us respond to this regulator's inquiry" with the inquiry pasted but no source material on the company's actual practices. The model invented favorable-sounding facts about the company's compliance posture.

The stronger prompt

The team supplied the company's real records and instructed the model to draft a response using only those records, flagging any claim it could not support and marking where a human needed to confirm a fact. The draft made no unsupported assertions and pointed the team to exactly what needed verification before filing.

The lesson on invented facts

When the model lacks grounding material, it fills the void with plausible content, which in a regulatory response is dangerous. Supplying real records and forbidding unsupported claims converts the model from a fabricator into a careful drafter. The full sequence is in Drafting Compliant Clauses With AI, One Deliberate Step at a Time.

What the Examples Share

Grounding plus stated context wins every time

Across all five scenarios, the weak prompt asked the model to supply substance from memory, and the strong prompt supplied the substance and asked the model to apply it. The difference is consistent: ground the model, state the jurisdiction and audience, and demand honest gaps.

The reviewer's job gets easier

In every strong version, the draft arrived with flagged gaps and preserved operative language, which is exactly what lets a qualified reviewer work fast. The model did not replace the reviewer; it prepared the material so the review was efficient.

Building Your Own Examples Library

Save the prompts that worked

The fastest way to improve is to keep the strong prompts from your own tasks. When a prompt produces a draft a reviewer barely had to touch, save it as a pattern with a note on why it worked. Over time you accumulate a library of grounding instructions, jurisdiction lines, and gap-flag conventions tuned to the documents you actually produce.

  • Capture the prompt, the source material shape, and the outcome.
  • Note the single choice that made the draft usable.
  • Tag each example by document type for quick reuse.

Learn from the failures too

The weak prompts are just as instructive. When a draft has to be thrown out, write down what the prompt left to the model's memory or assumption. Most failures trace to the same root: the prompt asked the model to supply substance instead of applying provided text. Recording these keeps you from repeating them.

Turning examples into templates

Once you have several strong examples for a document type, the common elements become a template, as described in the structured how-to. The examples are the raw material; the template is the distilled, reusable form. This is how a handful of lucky good drafts becomes a reliable, repeatable process.

Frequently Asked Questions

Are these real cases with real numbers?

They are composites of common situations rather than specific clients, and they include no invented statistics. The point is to illustrate the prompt choices that reliably succeed or fail, which generalize across many real tasks regardless of the specific facts.

What single change improved the most drafts?

Supplying the governing material and instructing the model to use only it. In every weak example the model fabricated substance from memory; in every strong one it applied provided text. Grounding is the change that turned each failure into a usable draft.

How did naming the plain-language standard help?

It gave the model a concrete target. The default register is dense, so "clear" left to interpretation produces lawyerly prose. Stating the audience level and the conspicuousness requirement let the model write to a real standard instead of guessing what clarity meant.

Why not let the model average multiple jurisdictions?

Because an average matches no actual jurisdiction and can be wrong everywhere. The stronger prompt named each jurisdiction and asked for differences to be surfaced, producing a base document plus marked variations a reviewer could resolve. Averaging hides exactly what needs attention.

What made the regulatory response safe?

Supplying the company's real records and forbidding unsupported claims. Without grounding, the model invented favorable facts, which is dangerous in a regulatory filing. With real records and a no-fabrication instruction, it drafted only what the records supported and flagged what needed human confirmation.

Can I reuse these prompt patterns directly?

Yes, as starting points. Adapt the grounding instruction, the jurisdiction and audience lines, and the gap-flagging requirement to your task. The patterns are deliberately general because the underlying lessons, ground the model and state the context, apply across legal and compliance writing.

Key Takeaways

  • Across every scenario, grounding the model in supplied text beat asking it to supply substance from memory.
  • A liability clause stayed correct only when the prompt protected operative terms and required carve-outs.
  • Plain-language disclosures worked when the comprehension standard and audience were named explicitly.
  • Multi-jurisdiction tasks need each jurisdiction stated and differences surfaced rather than averaged away.
  • Regulatory responses are safe only when grounded in real records with unsupported claims forbidden.
  • Strong prompts deliver drafts with flagged gaps and preserved language, which makes qualified review fast.

Search Articles

Categories

OperationsSalesDeliveryGovernance

Popular Tags

prompt engineeringai fundamentalsai toolsthe difference between AIMLagency operationsagency growthenterprise sales

Share Article

A

Agency Script Editorial

Editorial Team

The Agency Script editorial team delivers operational insights on AI delivery, certification, and governance for modern agency operators.

Related Articles

General

Prompt Quality Decides Whether AI Earns Its Keep

Prompt quality is the single biggest variable in whether AI delivers real work or expensive noise. The model matters, the platform matters — but the prompt you write determines whether you get a first

A
Agency Script Editorial
June 1, 2026·10 min read
General

Counting the Real Cost of Every Token You Send

Tokens and context windows sit at the intersection of AI capability and operational cost—yet most business cases treat them as technical footnotes. That's a mistake that costs real money. Every time y

A
Agency Script Editorial
June 1, 2026·10 min read
General

Rolling Out AI Hallucinations Across a Team

Most teams discover AI hallucinations the hard way — a confident-sounding wrong answer makes it into a client deliverable, a legal brief, or a published report. The damage isn't just to the output; it

A
Agency Script Editorial
June 1, 2026·11 min read

Ready to certify your AI capability?

Join the professionals building governed, repeatable AI delivery systems.

Explore Certification