Context, Not Prompts
Why it Matters
A commercial lease for a Coral Gables café buries its personal guarantee near the end, under Miscellaneous. Hand the lease to a chatbot with a one-line request to find the risks, and it may flag the late-payment interest and miss the guarantee, because you left it to decide what mattered. Give the same chatbot a written method, the statute that governs the lease, and a fixed form for its report, and the odds change. The request barely changed, but what the machine could see did, and someone had to write the method and check the report. The three audits here differ only in what the machine can see, and you score them, then explain why the context, not the request, changed what came back.
Current Context
On August 20, 2026, Thomson Reuters relaunched CoCounsel Legal as a “fully agentic” assistant (opens in a new tab), which it says will plan and execute across a whole matter rather than answer isolated prompts. Each matter gets a workspace of the firm’s own documents and precedents, so the model works from those instead of what it recalls. On September 14, ILTA’s 2026 Technology Survey, drawn from 508 firms (opens in a new tab), found 94 percent using or exploring generative AI but only 13 percent using agents. The most common AI product it named was Microsoft 365 Copilot, a general tool rather than a legal one (opens in a new tab). Both the new agent and the ordinary chatbot most paralegals use put the same question to this activity: how much of what comes back depends on what the system was given to read.
Key Concepts
Prompt and Context
A prompt is the request you type. Context is everything the model can read when it answers: your instructions, the documents you upload, the examples you give, and the earlier turns of the conversation. The answer tracks the quality of the context far more than the cleverness of the request.
System and Project Instructions
Standing instructions that apply to every conversation in a project or a custom workspace: the role the model plays, the format it returns, the rule that it cite a section for each claim. They are how a firm makes one AI tool behave the same way for every user, rather than trusting each person to prompt it well.
Skill Document
A written method the model follows for one task: its purpose, when to use it, the steps in order, the categories to check, the rating scale, and the form of the output. A good skill document turns a general model into a specialist for a single job, and a firm can review and improve it like any other work product. The Commercial Lease Auditor you use here is one.
Grounding
Supplying the authoritative text, the lease itself and the statute that governs it, so the model quotes what is in front of it instead of recalling what it once read. Grounded output can be checked against the source; ungrounded output has to be researched from scratch before anyone can trust it.
Model Context Protocol
An open standard that lets an AI tool reach document systems, databases, and other software through one common interface. It is how a model inside a firm connects to the document management system, which is why deciding what a model sees now includes deciding which tools it may use.
Evaluation Rubric
A fixed set of criteria for scoring an AI output, such as coverage of the required categories, accuracy of the section citations, correct risk ratings, and whether missing provisions were caught. Without a rubric, comparing two audits is a matter of taste. With one, applied the same way each time, the comparison becomes a measurement.
Florida-Specific Provisions
Lease terms that Florida practice tells a reviewer to check by hand: force majeure for hurricanes and named storms, a governing-law and venue clause, and any confession-of-judgment clause. Security-deposit handling is another, since the residential rules of section 83.49, Florida Statutes, do not reach a commercial lease. A model that knows contract law in general misses these unless the context points it at them.
Resources
- Sample Lease Miracle Mile (PDF) (The commercial retail lease for a Coral Gables café that you audit; the version for reading.)
- Sample Lease Miracle Mile (Markdown) (The same lease as plain text, which AI tools read more reliably than a PDF; the version to upload.)
- SKILL: Commercial Lease Auditor (PDF) (The skill document that encodes the audit method: ten categories, four risk ratings, a Florida checklist, and a fixed report format. Read it before Part 2.)
- SKILL: Commercial Lease Auditor (Markdown) (The same skill as plain text, for the second and third audits.)
- Florida Statutes Chapter 83, Landlord and Tenant (opens in a new tab) (Part I governs nonresidential tenancies; read it for the grounded audit and use it to check the Florida findings. Part II, the residential act, does not reach a commercial lease.)
- Landlord-tenant law, Cornell LII Wex (opens in a new tab) (A plain-language overview of the doctrine behind the audit’s categories.)
- Equipping agents for the real world with Agent Skills, Anthropic (opens in a new tab) (The October 2025 post that defines the SKILL.md format the Commercial Lease Auditor follows: a short file that tells a model when a method applies and where the rest of it is.)
- Effective context engineering for AI agents, Anthropic (opens in a new tab) (The maker’s own account of why what fills the context window matters more than the wording of the request. Read it as one vendor’s view, then test it on the lease.)
- Model Context Protocol (opens in a new tab) (The open standard that lets a model reach a firm’s documents and tools. It is the plumbing a firm uses to do at scale what you do here by pasting.)
- 2026 Legal Context Engineering Benchmark Report, NetDocuments (opens in a new tab) (A study by a document-management vendor that held the model constant and changed only what the model was given, then measured accuracy and cost. Read the methodology.)
- Gemini Notebook (opens in a new tab) (A source-grounded tool that answers only from the documents you load; an option for the grounded audit.)
What to Do
The three audits hold the model and the request steady and change only what the model is given, so any difference among them comes from that and nothing else. What you practice is deciding what the model sees before it reads a word, then checking what came back. What you post is one comparison of the three audits, with your reconciliation, two verified findings, and a short explanation.
Part 1: Read the Lease, the SKILL, and the Statute
Start with the Background reading, then read the Miracle Mile lease once, before any machine touches it, so you know the document yourself. Read the Commercial Lease Auditor skill next, and notice what it imposes: ten categories, four risk ratings, a Florida checklist, and a fixed report format. Then read Part I of chapter 83, Florida Statutes, the sections on nonresidential tenancies, and note how little it dictates for a commercial lease. Section 83.49, Florida Statutes, the residential security-deposit rule, sits in Part II and does not reach a commercial lease, which is the kind of thing the grounded audit should catch.
Part 2: Run the Three Audits
Run the same lease through a general chatbot three times, each in a fresh conversation, changing only what you give the model. For the naive audit, upload or paste the lease and give one instruction: “Review this lease and identify any issues or concerns.” Take the output as it comes. For the skill-guided audit, give the model the Commercial Lease Auditor skill first, then the lease, and tell it to audit the lease by that method. For the grounded audit, add Part I of chapter 83. Give the model the skill, the lease, and the statute, and tell it to follow the skill and to cite a statute section for every Florida finding. A source-grounded tool, one that answers only from the documents you load, works well for this third run. Save all three outputs, and note how long each took.
Part 3: Score the Audits Yourself, Then Reconcile
Score each of the three audits yourself on five questions, one to five per line, before you open the Audit Comparator, and keep these first scores. Did it cover the skill’s ten categories? Are its risk ratings consistent with the skill’s definitions? Does it cite a specific section for each finding? Did it name the provisions that are missing, not just flag the language on the page? And did it catch the Florida-specific issues? Then run the Audit Comparator, which scores the same three audits on the same five questions and explains the gaps. Where its score differs from yours, go back to the audit and decide who is right, then record the reconciled score with a reason.
Part 4: Verify Two Findings
Pick two findings from the grounded audit, ideally ones you are unsure of, and check each against the source with the Finding Verifier. Paste the finding, the lease language it points to, and the statute section if the finding relies on one. For each, decide whether the lease section says what the finding claims, whether the risk rating fits the text, and whether the legal conclusion holds. A finding can name a real section and still misread it, so read the words yourself and quote the ones that settle it.
Part 5: Post the Comparison
Create a single new discussion thread with your comparison. It holds the scored table, showing your first scores and the reconciled scores side by side on all five lines for the three audits. Add a short note on each place you changed a score, with the reason, and your two verified findings with the language you quoted. Close with about 250 words on why the three audits differed: what the skill added over the bare request, what the statute added over the skill, and where even the grounded audit fell short. The Context Explainer can help you reason through the mechanism before you write, but the explanation is yours, tied to your own results.
Part 6: Share Chat Link
Include one AI chat link with a 1–2 sentence explanation of what the conversation shows and why you chose to share it.
Suggested AI Prompts
Use these prompts as a starting point, then adjust them to fit your goal. Strong prompting develops through trial, revision, and testing. It’s a foundational skill that grows into more advanced AI work such as context engineering and agent-based workflows.
Audit Comparator
Act as a supervising attorney evaluating three lease audits prepared by a paralegal: one from a simple prompt, one following a structured SKILL document, and one following the SKILL with the governing Florida statute supplied as grounding. Score each audit from 1 to 5 on these criteria: coverage of the SKILL’s required categories, consistency of risk ratings, specificity of section citations, identification of missing provisions rather than only problematic language, and Florida-specific issues caught. Present the scores in a table, then explain the differences by pointing to particular findings or gaps in each audit. Tell me which audit you would trust most and why. Wait for me to paste the three audits.
A numeric scale applied the same way to all three audits is structured output, and it turns a comparison that would be a matter of taste into a measurement. Asking for the finding behind each score keeps the reasoning checkable.
Finding Verifier
Act as a skeptical legal reviewer checking AI work product. I will paste one finding from a lease audit along with the actual lease language it references and, where relevant, the text of the Florida statute. Verify whether the lease section says what the finding claims, whether the risk rating fits what the text shows, and whether the legal conclusion is correct under the statute I supplied. Quote the lease or statute language that supports or undercuts the finding. Be critical; I need to know whether this finding can go to an attorney. Wait for me to paste the finding and the source text.
Handing the model the lease section and the statute grounds it in the source and confines it to reading and comparing, where it is reliable. The demand for a quotation leaves you a trail you can check yourself.
Context Explainer
Act as a legal technology consultant helping a paralegal understand why structured context changes AI output. I just compared a naive prompt, a SKILL-guided audit, and a grounded audit of the same lease, and I will describe the differences I saw. Explain what was happening inside the model in each case, why supplying the statute changed the findings, what a good SKILL document does that a prompt cannot, and where even well-designed context still fails. Then help me think about which other legal document tasks the approach would fit and which it would not. Ask me to describe my results to begin.
Explaining the mechanism after you have seen the difference ties the reflection to your own results, and that is what makes it stick. Asking where the approach fails keeps the lesson from turning into a sales pitch.