Source material
The hardest document in the set, and why the answer is a second document
Terms and conditions are written to be defensible. Every property that makes them good at that, the numbering, the internal cross-references, the defined terms carrying capital letters, the deliberate breadth, makes them bad at being read one passage at a time. This is the one document in this set where the honest recommendation is not to fix it but to publish something beside it.
Why this one is harder than it looks
A contract is optimised for a dispute, not for a reader. Its sentences are long because they have to be exhaustive, its clauses are numbered because they have to be citable in an argument, and its structure follows legal logic rather than the order anybody wonders about things. None of that is a flaw in the document. It is a mismatch between what the document is for and what it is now being asked to do.
The cross-references are the sharpest edge. A clause reading that something applies subject to the limits in a numbered clause elsewhere is complete in a contract and empty in a passage, because the passage does not carry the clause it points at. Worse, every answer cites the passages it used by its own numbering, which has nothing to do with your clause numbering, so a reader sees a citation, goes looking for a numbered clause, and finds a completely different one.
Defined terms are the second edge. A contract that defines the Services on page one and then uses that capitalised word for forty pages is relying on the reader having page one available. A passage from page six does not, so the most load-bearing noun in the sentence means nothing, or worse, means whatever a plain reading suggests, which is usually broader than the definition.
Third, and this is the practical problem: the questions people bring to a business are almost never about the clauses that dominate the document. They ask what happens if they want to stop, whether they can transfer it to somebody else, what you are responsible for if something goes wrong, and who owns what they uploaded. Those are four short answers hidden inside forty pages of material that is mostly about other things, and every one of those forty pages competes for retrieval against them.
What it has to contain
Structure rather than wording. A passage pulled out of this document has to stand on its own, because that is the only form in which it will ever be read.
| A plain-language summary published as its own document | Written for a customer, organised by question, and clearly labelled as a summary of the terms rather than a replacement for them. This is the document you index. It should be short enough that somebody would read all of it. |
|---|---|
| A precedence sentence in both documents | One sentence in the summary saying the full terms govern and where to read them, and one in the terms acknowledging the summary exists and does not vary them. Have your advisers word both. Without this the summary is a liability rather than an asset. |
| Definitions restated wherever they carry weight | In the summary, no capitalised defined term should appear without the definition being present in the same section. If the Services means three specific things, say the three things, every time, rather than borrowing a definition from a page that will not be retrieved with it. |
| Cross-references replaced by the thing they point at | No clause number appears in the summary. Where the contract says subject to the limits below, the summary says what the limits are. This is the change that makes a passage self-contained, and it is impossible to do in the contract itself without breaking it. |
| One section per question people actually bring | Stopping, transferring, what you are liable for, what happens to their content or data, what happens if they breach, what happens if you change the terms. Headed in those words. Those six headings will absorb most of the traffic this document ever sees. |
The rewrite, in order
Write the summary and leave the contract alone
The instinct is to restructure the terms. Do not. The numbering is load bearing, the breadth is deliberate, and a contract edited for readability by somebody who is not a lawyer is a worse contract. The productive move is to write a second, shorter document that answers questions and points at the first one.
Work from the questions rather than from the clauses. List the things people actually ask your team about the agreement, then find the clause that answers each and write the answer in a paragraph anybody could read. You will end up with something between one and three pages, which is the right size.
Have the summary reviewed by whoever drafted the terms. It is a published statement about a contract, so it carries real weight, and the point of the exercise is accuracy at a different reading level rather than a friendlier tone over a vaguer meaning.
Expand every cross-reference as you write
Each time the contract sends you somewhere else, follow it and write what you find. If the liability clause is limited by an exclusions clause and a cap, then the liability section of the summary says what is excluded and what the cap is, in the same section, with no numbers in it at all.
This is tedious and it is the whole value of the exercise. A summary that reproduces the contract's habit of pointing elsewhere has reproduced the exact property that made the contract unanswerable.
Inline the definitions that do real work
Go through the summary looking for any word that means something specific because the contract said so. Replace it with the meaning, or state the meaning immediately beside it. Do this even where it makes a sentence longer, because a long sentence that is complete beats a short one that depends on a page nobody will see.
Pay particular attention to the words that have an innocent everyday meaning and a narrower contractual one. Those are the dangerous ones: a reader will resolve them from ordinary usage and never notice that the document meant something else.
Decide deliberately whether to index the contract at all
There is a real argument for indexing only the summary. The contract's passages compete with the summary's, they are longer, and the six passages retrieved for a question are capped in total size, so a page of dense clauses can crowd out the paragraph that would have answered well.
There is also an argument for indexing both, which is that the contract contains material the summary omits, and a question about that material is better answered from a clause than refused. If you do index both, keep the summary short and specific so it wins on the questions it covers.
Whichever you choose, make it a decision rather than an accident. Uploading a contract because it existed is how most businesses end up with an assistant that answers every commercial question in the voice of a lawyer.
What happens if you skip it
A clause quoted verbatim, entirely accurate, and of no use to anybody
Somebody asks what happens if they want to stop using the service. The retrieved passage is the termination clause, which is correct, which is cited, and which contains two references to other numbered clauses, one defined term, and a sentence structure built for a court. The customer reads it twice and asks a person.
Nothing was wrong. That is what makes this failure distinctive: the retrieval worked, the citation was honest, and the answer still failed, because the document cannot be read in fragments by a non-lawyer. It is the one case in this set where the fix is not a better arrangement of the same words but a second document written for a different reader.
Check it against this
Before you index it
- A plain-language summary exists as its own document
- Both documents carry a precedence sentence worded by your advisers
- No clause numbers appear anywhere in the summary
- Every defined term in the summary carries its meaning in the same section
- The summary has a heading for stopping, transferring, liability and data
- The summary is short enough that somebody would read all of it
- Whether the full terms are indexed at all was a decision, not a default
Questions
- Is a summary risky?
- It is a published statement about a contract, so it is not risk free, which is exactly why the precedence wording and the review matter. The alternative risk is worse and less visible: a contract answering customer questions in fragments, badly, with a citation attached.
- Can the assistant refuse legal questions outright?
- It answers only from the material supplied and says it does not know when the material does not cover the question, and below the match threshold it declines using wording you wrote. Terms are one of the strongest cases for making that wording point at a person by name.
- Our terms are one very long file. Does that matter on its own?
- It matters for how it is read. A long file is split into many passages that all sound alike, so they compete with each other for the same questions, and the material given to the model for any one answer is capped. Length is not fatal, but it is one more reason the short summary is the document that should be doing the work.
Keep reading
- How to write a returns policy an AI assistant can answer fromWhy a passage about the return window answers confidently without the exclusions, and how to restructure the document so it cannot.
- Writing a cancellation policy an assistant can quote accuratelyNotice periods are relative to something the passage does not carry. How to write a cancellation policy that survives being read in pieces.
- How to structure warranty terms for a retrieval assistantThree warranties get conflated by customers. How to separate them, state the void conditions, and write what parts and labour excludes.
- Every kind of source materialWhat to feed an assistant, and the surgery each document needs first.
Try it on your own material
Upload a document or point it at your site, paste one line of HTML, then ask it something only your business could answer.