How to Draft a Contract: A Step-by-Step Guide

A contract is structure first, language second. Before you write a single clause, you decide what the deal is, who owes what, and what happens when it breaks. The polished sentences come after. Most people do it backwards: they grab a template, tweak the price and the names, and ship it without reading the parts that decide the outcome.

This guide covers the anatomy of a contract, a repeatable drafting process, the mistakes that cost real money, and the limits of AI. The throughline: the risk that hurts you hides in the boilerplate, not the headline terms.

This is general information for in-house teams, not legal advice for a specific deal.

Vaquill AI, the legal AI suite for in-house teams

TL;DR

  • Structure before prose. Define the deal (parties, obligations, money, exit, failure) before you write clauses. A clear deal in plain language drafts itself.
  • Know the anatomy. Title and parties, recitals, definitions, operative clauses, representations and warranties, covenants, conditions, term and termination, boilerplate, signature. Each block does one job.
  • A contract needs formation. Offer, acceptance, consideration, and mutual assent make it enforceable. A signed document with no real bargain is a wish.
  • The risk hides in the boilerplate. Governing law, assignment, limitation of liability, indemnity, and notice get copied without reading. Treat them as load-bearing, not filler.
  • Read clauses in pairs. Most boilerplate damage comes from two clauses that each read fine but contradict each other, like a cap that quietly swallows an indemnity.
  • AI drafts fast, you own the risk. It does not know your leverage, and it fabricates cross-references and misaligns carve-outs.
Quick check

Per the post, which term has ranked the most negotiated commercial term for over a decade?

For the AI side of drafting in depth, see our AI legal drafting guide. For standard positions on specific clauses, the clause library covers the language.


What makes a contract a contract

A document is not a contract because it has signatures and the word "Agreement" at the top. It is a contract because four things are present. Offer: one side proposes terms. Acceptance: the other side agrees. Consideration: each side gives up something of value, money, a service, a promise. Mutual assent: both sides intend the same deal. Capacity and a lawful purpose round out the basics.

Strip any one of those and you have paper, not an enforceable bargain. Formation is the floor. Get the bargain right first, then make it readable.


The anatomy of a contract

A contract is a stack of blocks, each doing one job. Learn the blocks and any agreement stops looking like a wall of legalese. Here is the standard order.

SectionWhat it does
Title and partiesNames the agreement and identifies who is bound (legal names, entity type, addresses)
RecitalsThe "whereas" background: why the parties are contracting (context, not binding terms)
DefinitionsSets the meaning of capitalized terms used throughout, so words mean one thing
Operative clausesThe core deal: scope, price, payment, delivery, IP, the obligations each side takes on
Representations and warrantiesStatements of fact each side relies on (and can sue over if false)
CovenantsPromises to do or not do something during the term
ConditionsTriggers that must happen before an obligation kicks in
Term and terminationHow long it lasts and how either side can exit
BoilerplateGoverning law, assignment, liability, indemnity, notice, and the rest of the "miscellaneous"
SignatureExecution block: who signs, in what capacity, when

Four of these get confused often. Representations assert a fact as true now ("the company owns the IP it licenses"). Warranties promise that fact stays true or give a remedy if false. Covenants are forward-looking promises ("the vendor will maintain insurance"). Conditions gate obligations ("payment is due upon delivery"). Mislabeling them changes your remedy when something goes wrong. And a sloppy definition quietly moves the scope of every clause that uses the term.


The boilerplate is where the risk lives

Here is the part most drafting advice skips. The clauses people fight over (price, scope, the headline cap) get read line by line. The boilerplate at the back gets pasted from the last deal and never reread. That is backwards. The back of the contract is where a dispute turns bad.

Five boilerplate clauses carry more weight than their placement suggests:

  • Governing law and forum. Which state's law applies, and where you litigate. Litigating across the country can cost more than the claim.
  • Assignment. Can the counterparty hand your contract to a competitor, or to whoever buys them? A silent or permissive clause means you can wake up bound to a party you never chose.
  • Limitation of liability. The cap and the exclusions decide what a breach is worth. A clause that excludes consequential damages and caps everything at fees paid can turn a six-figure failure into a refund.
  • Indemnity. Who pays when a third party sues. An uncapped or one-directional indemnity can dwarf every other number in the agreement.
  • Notice. How and where formal notice must be sent. Get this wrong and your termination or breach claim does not legally land.

None of these are headline terms. All of them can decide the outcome.

This is not a fringe worry. In World Commerce and Contracting's annual Most Negotiated Terms report, limitation of liability has ranked the most negotiated commercial term for over a decade, with indemnification close behind (WorldCC, 2022). The clauses people fight hardest over are the ones drafters paste fastest.


Three clauses lawyers under-review (bad vs better)

In my experience reviewing in-house paper, three clauses get the least attention per dollar of risk: limitation of liability, assignment, and notice. Here is a bad and a better version of each. These are illustrative snippets, not model clauses for your deal.

Limitation of liability

Bad: In no event shall Provider's aggregate liability exceed the fees paid in the twelve (12) months preceding the claim.

Better: Except for the Excluded Claims below, each party's aggregate liability shall not exceed fees paid in the twelve (12) months preceding the claim. Excluded Claims: breach of confidentiality, breach of data security obligations, and indemnification for third-party IP infringement, which shall be capped at [super-cap amount].

Why it matters: the bad clause is one-directional and caps everything, including a data breach, at a refund. The better clause makes the cap mutual and lifts the highest-risk categories above it.

Assignment

Bad: Either party may assign this Agreement.

Better: Neither party may assign this Agreement without the other's prior written consent, except that either party may assign to a successor in a merger or sale of substantially all assets, provided the assignee is not a competitor of the non-assigning party.

Why it matters: the bad clause lets your counterparty hand the contract to anyone, including a competitor who acquires them. The better clause permits a normal sale but blocks assignment to a rival.

Notice

Bad: Notices shall be sent to the addresses on the cover page.

Better: Notices must be in writing and sent to the address and named contact in Exhibit A by hand delivery, certified mail, or a tracked courier. Notice is effective on receipt. Each party shall keep its notice details current.

Why it matters: the bad clause does not say what counts as notice, when it is effective, or who receives it. A termination sent the wrong way may not legally land. The clause library has standard positions and fallbacks for each of these.


A short war-story: the cap that excluded the wrong thing

A SaaS customer signed a vendor agreement with a clean-looking liability clause. The cap was twelve months of fees. The carve-outs above the cap covered confidentiality and IP infringement. Indemnity was not carved out. Everyone read the cap, nodded, and signed.

A year in, a third party sued the customer over the vendor's component. The vendor's indemnity clause promised to defend such claims, and the defense costs ran past the annual fees. Then the vendor pointed at its own cap: indemnity was not an Excluded Claim, so it was capped too. The customer had an indemnity on paper and a cap that swallowed it. The fix was one line in the carve-out list. Each clause read fine alone, and the interaction is what broke.


A step-by-step drafting process

Good drafting is a sequence, not a single sitting at a blank page. Six steps, in order.

Loading diagram...

Step 1: Define the deal

Before any clause, answer five questions in plain English. Who are the parties? What is each side getting? What is the money and when does it move? How long does this last? What happens if someone breaches or wants out? If you cannot state the deal in a paragraph, you are not ready to draft.

Step 2: Pick a base or template

Start from a trusted precedent or your standard form, not a blank page and not a random web template. A base you understand saves hours. A base you do not understand imports risk you cannot see. Match it to the deal type and jurisdiction, then rework it instead of filling in blanks.

Step 3: Draft the operative terms

Write the core deal first: scope, price, payment, delivery, IP, the obligations. This part reflects what you negotiated, so it gets the most thought. Keep one idea per clause. Cross-check every defined term against your definitions section as you go.

Step 4: Add the protections

Now layer in representations, warranties, covenants, and conditions. These allocate risk: who promises what is true, who keeps doing what, and what triggers each obligation. Match each protection to a real risk in this deal. Generic reps copied wholesale add length, not cover.

Step 5: Pressure-test the boilerplate

Do not skim the back of the contract. Read governing law, assignment, liability, indemnity, and notice against the worst case. Ask: if this deal goes wrong, does this clause help me or the other side? Where does each carve-out point, and does that clause exist in the document?

Step 6: Review end to end

Read the whole thing once, start to finish. Check that defined terms resolve, cross-references point where they claim, numbers are consistent, and the recitals match the operative deal. Then a human signs off. This is where the contract becomes one coherent agreement.


Common drafting mistakes

The same errors show up across thousands of contracts. Most are structural, not stylistic.

  • Inconsistent defined terms. "Services" defined once and used to mean three things. Drift corrupts every clause that uses the term.
  • Broken cross-references. "Subject to Section 9.2" when 9.2 is about something else, or vanished after an edit. Renumbering breaks these silently.
  • Carve-outs that point at nothing. A clause excluding "indemnification under Section 11" when there is no Section 11. It reads fine and protects no one.
  • Vague obligations. "Reasonable efforts" and "promptly" with no standard or deadline. If you cannot tell whether a clause was breached, it is not doing its job.
  • No exit. A termination section that lets you in but not out, or a renewal that auto-extends with no clean way to stop it.

Most of these are caught by the end-to-end review in step 6. For how a structured review surfaces them at scale, see the in-house contract review playbook.


Where AI helps and where it does not

AI is good at the blank-page problem. Give it the deal terms and it returns a structured first draft in seconds instead of an afternoon. It assembles clauses from a precedent, redrafts a one-sided indemnity into a mutual one, and converts dense language for a business reader.

What it does not do is own the risk. The model does not know your leverage or that this counterparty is your only viable vendor. It fails most often in three spots. Cross-references to clauses that do not exist. Defined terms used before they are defined. Caps or carve-outs that do not line up, the pairing problem from the war-story above. Each reads fine in isolation.

The duty does not move when you use a tool. ABA Formal Opinion 512 (July 2024), the bar's first comprehensive guidance on generative AI, ties competence to understanding the tool's limits and puts verification on you. Our ABA Formal Opinion 512 guide covers the full set of duties.

The split is simple. AI compresses the drafting. You own the judgment, the verification, and the signature. Treat its output as a fast junior's first draft you always review. For the full AI-drafting workflow, see the AI legal drafting guide and our AI contract review lawyer's guide.


The verdict

Drafting a contract well is mostly a structuring problem, not a writing problem. Define the deal in plain language, build it block by block, and the prose follows. The biggest mistake is treating the boilerplate as filler. Governing law, assignment, liability, indemnity, and notice decide more disputes than the headline terms.

One more opinion, since people get it backwards. A short contract is often more dangerous than a long one. The short version usually drops the protections you only miss when the deal breaks: the cap, the indemnity, the notice mechanics. Length is not the enemy. Missing structure is.

A grounded drafting tool changes how fast you reach a first draft. It does not change who owns the risk. Structure the deal, treat the boilerplate as load-bearing, verify everything, and sign only what you have read. For a tool that drafts from your standard positions and verified law, Vaquill AI is built for in-house teams. For the risk side of the same skill, see the contract risk assessment framework.

FAQ

What are the basic parts of a contract? Title and parties, recitals, definitions, operative clauses, representations and warranties, covenants, conditions, term and termination, boilerplate, and the signature block. Each block does one job. The operative clauses carry the core deal, and the boilerplate at the back allocates risk if things go wrong.

What makes a contract legally enforceable? Four things: an offer, acceptance of that offer, consideration (each side gives up something of value), and mutual assent that both sides intend the same deal. The parties also need capacity to contract and a lawful purpose. A signed document with no real exchange of value can fail for lack of consideration.

Why is boilerplate the riskiest part of a contract? Because it gets copied from the last deal and rarely reread, while the headline terms get scrutinized. Governing law, assignment, limitation of liability, indemnity, and notice often decide the outcome of a dispute. Treat them as load-bearing structure, not standard filler.

Can I just use a contract template? A trusted template is a fine starting point, not a finished contract. Templates miss deal-specific risk, carry clauses that do not fit, and break when you edit numbers and names without checking cross-references. Use one you understand, match it to the deal type and jurisdiction, then rework it.

Is a shorter contract safer? Often the opposite. A short contract usually drops the protections you only miss when the deal breaks: the cap, the indemnity, the notice mechanics. Length is not the enemy. Missing structure is. Cut clauses that do no work, never the ones that allocate risk.

Which contract clauses get under-reviewed the most? In practice, limitation of liability, assignment, and notice. People scrutinize price and scope, then paste the back of the contract from the last deal. World Commerce and Contracting reports that limitation of liability is the most negotiated commercial term, yet it is among the most copied at the drafting stage.

Where does AI drafting fail most often? Three spots. Cross-references to clauses that do not exist. Defined terms used before they are defined. Caps or carve-outs that do not align. Each reads fine in isolation. The fix is checking every reference and reading related clauses as a pair.

Can AI draft a contract for me? AI can produce a structured first draft fast, assemble clauses from a precedent, and redraft to a position. It cannot make the judgment calls, know your leverage, or be trusted without verification. Treat its output as a first draft you check against the source and sign only after a full human review.

Last updated: June 2026.

Legal AI that reads your documents and knows the law.
Ask a legal question, review a contract, or search thousands of your files. Every answer shows where it came from. 7-day free trial, no card.
16 min read

New legal AI guides, weekly.

Arshita Anand

Arshita Anand

Co-Founder & CEO · Attorney

Arshita leads product and strategy at Vaquill, building the legal AI suite that solo, small-firm, and in-house US lawyers use to run a matter end to end.