Searching your firm's own precedents and know-how with AI
By the ChamberOne team 5 minute read
Every firm has the clause someone drafted brilliantly in a hard negotiation, the research note that settled a tricky point, the letter that defused a dispute. Almost nobody can find them. They sit in folders named after matters from years ago, known only to the lawyer who did the work, who may have left. Search tools that read language rather than file names promise to change that. They can, but only if you treat the confidentiality side as seriously as the finding side.
Key points
- Decide what counts as a precedent, mark its source and remove outdated material before indexing.
- Search must respect the permissions of the person asking, including restricted matters.
- Test the ethical wall with a distinctive phrase, and repeat the test after updates.
- Treat results as leads. Open the source, check context and check the date.
Why firm knowledge goes missing
Knowledge rarely disappears all at once. It leaks. A precedent is saved under a client's name. A good version is emailed round and never filed. A research note lives in a lawyer's personal folder. When that lawyer retires, the know-how goes with them.
The usual fix is a precedent bank run by a knowledge lawyer. That works in firms large enough to pay for one. In smaller firms the bank is a folder, updated when someone remembers. Search that understands meaning can make even a messy archive useful, because it does not depend on a tidy label.
Three ways people look for past work
Lawyers do not all search the same way, so an index built around only one of them will disappoint the others.
- By clause: show me every limitation of liability clause we have negotiated in a software contract, and how they differ.
- By party: what have we done for, or against, this company before? This overlaps with conflict checking and needs care.
- By citation: where have we relied on this authority, and in what kind of argument?
- By task: how did we approach a similar disclosure exercise? This is the hardest to index and often the most useful.
Make the material findable first
AI search sits on top of your documents. If they are poor, the results are poor. A few modest steps help far more than any tool setting.
- 1
Decide what counts as a precedent. A signed contract is a record of one deal. A clean template or a clause the team approves is a precedent.
- 2
Mark the source of each item: which matter, which year, which partner approved it.
- 3
Remove or flag outdated material, such as a clause drafted before a change in the law. Old and wrong documents are the main danger in any archive.
- 4
Keep a short list of preferred positions so that the search shows what you would use now, not only what you once used.
Confidentiality is the hard part
A precedent taken from a live client's matter carries that client's secrets. A search tool that makes every document in the firm available to every user is a confidentiality problem waiting to happen, even if it is also a very good search tool.
Before turning anything on, answer one question clearly: when a lawyer searches, which documents can the search read on their behalf? The right answer is only those the lawyer is already allowed to see. If the system searches everything and filters afterwards, check that the filter holds in every place a result can appear, including summaries, suggestions and counts.
Ethical walls and restricted matters
Some matters are walled off from parts of the firm, because of a conflict, a sensitive transaction or a client's request. Those walls must apply to search as firmly as they apply to the file. Test this directly. Create a restricted test matter with a distinctive phrase in a document. Ask a user outside the wall to search for that phrase in several ways, including by asking a question in plain language. If it appears in any result, snippet or generated answer, the wall has a hole.
Test again whenever the tool is updated. Treat this as a standing check, not a one-off.
Cleaning client material into know-how
Many firms decide that the best route is to create a separate, cleaned layer. A lawyer takes a good clause or note, removes names and identifying details, records where it came from and files it in a knowledge area visible to everyone. The raw matter stays restricted. The cleaned copy becomes the shared precedent.
This takes effort, and it should be rewarded rather than left to goodwill. Some firms allow a small amount of non-billable time per month for it. Others make it part of closing a matter: before the file is archived, the lawyer nominates anything worth keeping.
- Stays restricted
- Carries the client's secrets
- Names and identifying details removed
- Source recorded
- Filed in a knowledge area visible to everyone
Checking what the search returns
Search results are leads, not authority. Suppose a lawyer asks for the firm's usual indemnity wording and the tool returns a clause from a deal where the firm acted for the other side, on a different law, with a different balance of power. The words may look right and be quite wrong. Always open the source, check the context and check the date.
If the tool writes a summary or an answer, make sure each statement links back to a passage you can read. An answer with no visible source is unusable in a professional setting.
- Open the source, check the context and check the date
- Make sure each statement links back to a passage you can read
- Treat a result as authority
- Use an answer with no visible source
Questions to ask a vendor
- Does search respect the same permissions as the underlying files, including restricted matters?
- Can I see which passage each result or answer came from?
- Where is the text processed, and by which providers?
- Can I exclude a client, a matter or a folder from search entirely?
- Is search activity recorded, and can I review it?
- What happens to the index if I delete a document?
Where it does not help
Search will not tell you whether a precedent is good. It will not replace a knowledge lawyer's judgment about which version should be the standard. And it will find only what exists. A firm with no habit of recording its lessons gets little from a better search box. Start with the habit.
Finding past work in ChamberOne
- Clients, parties, documents, clauses and citations are linked, so related work can be found from one record.
- Document analysis lists the type, parties, key dates, clauses and citations, each linked to the paragraph it came from.
- Citations can be found again from other matters.
- A matter or client marked Restricted does not appear in lists, search, reports or the audit log for anyone who is not on its team, an Owner or named.
What it does not do
- The analysis is produced by an AI model and can miss or misread things. Check it against the document.
- A conflict check result is a warning based on names. The decision to act stays with the lawyer.
- A native graph database is on the roadmap. It is not in the product today.
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
How to check AI legal research before you rely on it
A routine for tracing every AI-suggested authority to its source, confirming it is still good law and testing the proposition it is cited for.
5 minute readGeneral chatbots versus legal research tools: what differs and why it matters
How general chatbots and legal research tools differ on sourcing, links, jurisdiction coverage and currency, and when a general model will do.
5 minute readRetrieval-augmented generation explained for lawyers
A plain explanation of how AI answers from a document set, why it reduces but does not remove fabricated content, and what to ask vendors.
5 minute read