Target the repeated admin in every file: onboarding letters and checklists, chasing emails, turning search results and title documents into a draft report for the client, drafting enquiries, status updates and completion checklists. AI drafts and summarises; the conveyancer checks every point against the documents. Keep it away from bank details, identity decisions and undertakings.
Going by the stage timings below, a reasonable target for a typical residential file is 45 to 90 minutes saved, mostly in writing rather than reading. The time you won't save is the legal review: a title report the AI drafted still needs you to read the title, because the one restrictive covenant it glossed over is the one that stops the client building their extension.
Where the admin sits in one transaction
Walk a single purchase file from instruction to registration and the repetitive work is easy to spot. The minutes below are illustrative for a straightforward residential purchase; yours will differ, so time a few files before and after.
| Stage | Admin that repeats on every file | AI's job | Illustrative minutes saved | Stays human |
|---|---|---|---|---|
| Instruction and onboarding | Client care letter, document requests, answering "what happens next?" | Tailor the welcome pack to the transaction; draft replies to common questions | 10-15 | Terms, fees, conflict checks |
| Identity and funds | Requesting documents, explaining why you need them | Draft requests and plain-English explanations | 5 | Every verification decision |
| Contract pack and searches | Reading and summarising long documents | First-pass summary with page references | 15-25 | Reading the documents, judging risk |
| Enquiries | Drafting additional enquiries, chasing replies | Draft enquiries from gaps you identify; draft chasers | 10-15 | Which enquiries matter; whether replies are adequate |
| Report to client | Turning findings into a readable report | Draft from your reviewed notes | 15-20 | Every statement of risk or advice |
| Throughout | Status updates to clients, agents, brokers | Draft updates from the file's milestones | 10-15 | Anything about dates not yet agreed |
| Completion and after | Checklists, completion letters, registration follow-up | Fill checklists and letters from the file | 5-10 | Funds, undertakings, the completion itself |
The pattern is the same throughout: AI saves writing time, not reading or deciding time. Anything that feels like it's saving you reading time deserves suspicion.
Onboarding: a welcome pack that fits the transaction
Most firms send the same client care letter and document list to everyone, and then answer the same questions by phone because the letter didn't fit the client. AI can tailor a standard pack in a couple of minutes. Give it your standard letter and the transaction facts (first-time buyer, mortgage, leasehold flat, gifted deposit) and ask it to reorder and trim the content so the steps that apply to this client come first, in plain English, without changing fees, terms or legal wording.
A before and after of one paragraph:
Before: "Where any part of the purchase monies is being provided by a third
party, we will require evidence of the source of such funds together
with a signed declaration from the donor in our prescribed form."
After: "Because your parents are giving you part of the deposit, we'll need
two things from them: bank statements showing where the money came
from, and a short signed form confirming it's a gift, not a loan.
We'll send them the form directly."The second version prevents three phone calls. Keep the legal meaning identical and check it before sending; the AI only changes how it's said.
Identity requests are where a tailored draft most often overreaches. Asked to "explain what ID we need and why", one illustrative draft listed "a recent utility bill, which can be a PDF printed from your online account" as proof of address, and told the client the firm "must see this within seven days". Neither came from the firm: its policy doesn't accept downloaded bills, and there's no seven-day rule. Give the AI your firm's own list of acceptable documents and its standard explanation, and tell it to rephrase only. The decision about whether a document is good enough stays with the person verifying it.
Searches and title: a first-pass summary you then verify
This is the biggest single saving and the biggest risk. Use AI for a first-pass summary with page references, then read the documents yourself with the summary beside you as a checklist. In an approved business tool, try a prompt like this:
You are assisting a conveyancer. Below are the title register, the title plan
description and the search results for a residential purchase. Produce a
table with columns: Item | What the document says (quote key words) |
Page/entry reference | Why it may matter to a buyer.
Include every covenant, easement, restriction, charge, note and adverse
search entry. Do not state that anything is "not a concern", "standard" or
"can be ignored"; the conveyancer will assess that.
If any page is illegible or missing, list it at the end.An illustrative extract of the output:
Item: Restrictive covenant | "not to erect any building or structure on the
land hatched blue without consent" | Charges/Covenants entry 2 | May limit
extensions or outbuildings in the hatched area.
Item: Easement | "right of way on foot only over the path coloured brown" |
Property entry 3 | Neighbours may have access over part of the garden.
Item: Planning search | Application for a two-storey rear extension at the
adjoining property, approved | Search p.4 | Construction next door; possible
effect on light or access.
Illegible: none.What the conveyancer adds: the AI can't see which area is hatched blue, so it can't say whether the covenant covers the part of the garden where the client wants a garden office; you check the plan. It described the right of way accurately but missed a second easement in a deed referred to in the register, because the deed itself wasn't in the input; the register said "as mentioned in the deed dated...", and you need to obtain and read it. A summary is only as complete as the documents you give it, which is why "list missing pages" and a manual cross-check against the register entries matter. The quickest version of that cross-check is a count: if the register has seven numbered entries across its sections and the table has six rows, find the missing one before you read anything else. On an illustrative file it was a note about a deed of variation, the one entry the tool had folded into the covenant above it.
Once you've reviewed everything, the same tool can draft the client report from your own notes on each item, which is where the 15 to 20 minutes are saved. For the general limits of AI document review, see what AI contract review catches and misses, and before uploading client documents anywhere, whether chat-with-PDF tools are safe for client files.
Enquiries and chasers
Once you've identified what's missing or unclear, AI turns your shorthand into properly worded enquiries. "No completion cert for the loft conversion; boundary fence responsibility unclear; ask about the neighbour's extension" becomes three numbered, precisely worded enquiries in your firm's style in under a minute.
An illustrative draft from that shorthand:
1. Please supply a copy of the completion certificate for the loft
conversion. If none was obtained, please confirm when the work was
carried out and by whom.
2. Please confirm which party is responsible for the boundary fence on the
left-hand side as viewed from the road, and whether the seller has
maintained it.
3. Please confirm whether the seller is aware of any issues with the
property.
4. Please supply details of the extension at the neighbouring property,
including the planning reference.Two edits before it goes. Enquiry 3 appeared from nowhere, and it's the kind of catch-all that earns a reply of "not aware" and wastes a round. Delete it. Enquiry 4 asks the seller for something your planning search already gives you; what you actually need is whether the seller received notice of the works or has any agreement with the neighbour about access or scaffolding. Rewrite it to ask that. Then check that each enquiry maps to one gap on your own list, and nothing more.
AI helps again when replies come back. Paste your enquiries and the replies side by side and ask for a table: enquiry number, what was asked, what the reply says, and whether the reply answers the question, answers part of it, or declines to answer ("buyer to rely on own searches", "not aware"). An illustrative row: "2 | Fence responsibility | Seller believes the fence is theirs and has replaced panels | Answers part: belief, not a document; no reference to deeds." Whether a partial answer is good enough for this client and their lender is your call, but the table stops a weak reply slipping through because it arrived in a long email with nine good ones.
Chasing is where the time leaks. A sequence of chasers for the other side, drafted once and stored as templates, keeps the tone professional as the weeks go by. Here's a second chaser (illustrative):
Subject: [Property address] - our ref [ref] - replies to enquiries
Dear colleagues,
We wrote on [date] and [date] with additional enquiries 1 to 4. Our client
and their lender are waiting on the replies to 2 (completion certificate for
the loft conversion) and 4 (the adjoining extension) before we can report.
Could you let us know by [date] when we can expect replies, or whether any
of these need to go back to your client? Thank you.On sale files the direction reverses, and AI helps you get ahead of the buyer's enquiries. Before the contract pack goes out, paste in the seller's completed property information form alongside your search results and title summary, and ask the tool to list every place where the form and the documents disagree. Take an illustrative sale: the seller answered "no" to alterations, but the planning search shows an approved rear extension from eight years ago, and the form says the boundary fence on the left is theirs while the title plan shows no marking either way. Neither is necessarily a problem, but both would come back as enquiries three weeks later. Raising them with the seller now, and getting the completion certificate in hand, takes a week out of the timeline. The answers on the form stay the seller's own; the AI only spots the inconsistencies.
Keep a person on the send button. A chaser that goes out automatically after the other side has already replied (by phone, say) looks careless, and conveyancing runs on goodwill between firms.
Status updates without the phone calls
"Where are we?" calls from clients, agents and brokers take up a surprising share of a conveyancer's day. If your case management system records milestones (searches received, enquiries raised, mortgage offer received), AI can turn them into a short weekly update. An illustrative one generated from a milestone list:
Update on your purchase, week of [date]
Done: searches received and reviewed; mortgage offer received.
In progress: we're waiting for replies to four questions we asked the
seller's conveyancer on [date]; we chased on [date].
Next: once replies arrive we'll send you our report and the documents to sign.
What we need from you: nothing this week.Rules for these updates: they state milestones, never predict dates that haven't been agreed, and never comment on the other side's conduct. Send them in one batch each week and "where are we?" calls drop noticeably. A quick sum for the team described further down: 20 open files, each producing about two status calls a week at six minutes a call, is four hours a week on the phone. If weekly updates remove half of those calls, that's two hours back every week, before counting the emails to agents and brokers. The approach for litigation and other matters is in how small law firms use AI to answer client status questions.
The date rule gets tested when a client asks directly. An illustrative draft reply to "will we complete by the 30th?" read: "We're on track to complete by the 30th once the replies come in." Nobody had agreed the 30th, and the replies weren't in. The corrected version says only what's known: "We can't fix a completion date until the replies arrive and both sides agree a date. The 30th is still possible if the replies come this week, and we'll confirm as soon as a date is agreed." Clients forward these messages to removal firms and employers, so a hopeful sentence becomes someone else's booking.
What AI must never touch in a transaction
- Bank details and payment instructions. Any change to where money goes is verified by phone to a number you already hold, never from the email itself. Keep automation and AI triage entirely away from these messages.
- Identity and source-of-funds decisions. AI can draft the request; a person decides whether the evidence is adequate.
- Undertakings, exchange and completion. These are professional acts with personal consequences. AI can prepare the checklist; you perform the act. Watch how it fills one: in an illustrative pre-completion checklist, the AI ticked "signed transfer received" because the client had emailed "I'll post the signed transfer tomorrow". A promise became a tick. Instruct it to mark an item done only where the document itself is on the file, with its date of receipt, and to mark everything else "outstanding".
- Advice on risk. Whether a covenant, search result or missing certificate is acceptable for this client and their lender is your call.
The first point matters most. In an illustrative scenario based on a common fraud pattern, a firm set up an AI summary of incoming client email to save reading time. A fraudster who had compromised the client's email account sent "updated bank details for the balance of the deposit refund". The summary read "Client update: new account details for refund", filed alongside routine messages. Only the firm's rule that any payment-detail change triggers a phone call to the client on the number from the onboarding form stopped the money going. AI also makes fraudulent emails more convincing, with correct spelling and a plausible tone, so the checks matter more, not less. See how to spot deepfake voice and video scams for the newer versions of this attack.
Tools for a small conveyancing team
Start with what's already in your systems. Some legal case management systems now include AI that can answer questions about a matter from its documents and correspondence: LEAP's Matter AI, for example, lets users query a matter's documents, emails and key dates and runs text recognition over scanned PDFs and handwritten notes. If your system has something similar, test it on closed files first.
For drafting and summarising, a general assistant on a business plan is enough: Microsoft 365 Copilot (Copilot Business lists at $21 per user per month on annual billing), ChatGPT Business or Claude Team (both $25 per seat monthly or $20 billed annually, minimum two seats). All three keep business content out of model training by default. Whatever you choose, keep each file's material in the file, and stick to tools approved for client data. The wider drafting method for legal letters is in how small law firms use AI to draft letters and routine documents.
One residential team's first three months
The team in this illustration is one conveyancer, one assistant and a part-time administrator, running about 20 purchase and sale files a month. Using the stage table as a guide, they start with three changes: tailored welcome packs, first-pass title and search summaries, and weekly milestone updates.
- Month one: the conveyancer runs the title summary prompt on ten closed files and compares its tables with her own reports. It misses one item across the ten, a note referring to a deed that wasn't uploaded, which becomes the "missing documents" rule.
- Month two: live files. Welcome packs save about 12 minutes each; summaries and report drafting about 30; weekly updates replace roughly two calls per file per week.
- Month three: chaser templates added. The team measures about an hour saved per file on average, around 20 hours a month, spent on taking more instructions without adding staff.
What didn't change: the conveyancer still reads every title, every search and every contract. The hour saved came from writing, formatting and phoning, which is where it should come from.
Conveyancers' questions about AI on live files
Can AI read scanned title documents and plans?
Modern assistants read typed scans reasonably well, and some case management systems run text recognition over scanned PDFs and handwritten notes so their AI can search them. Old deeds, handwriting and faint copies still produce errors, so check any quoted wording against the image. Plans are weaker still: AI can describe a plan but shouldn't be trusted to interpret boundaries, colours or easement routes. Read plans yourself.
Do clients need to be told we use AI?
Include a short statement in your client care terms: that you use software, including AI tools on business plans that don't train on client data, to help prepare documents and updates, and that a conveyancer reviews all legal work. Check whether any lender or panel terms you work under say anything about AI or outsourcing, and follow them.
Is AI useful on leasehold information packs?
Yes, for turning a long pack into a list of the points that matter: service charge history, planned major works, ground rent terms, restrictions on letting or pets, and missing documents. Ask for page references for every point so you can check them quickly. The judgement about whether the terms are acceptable for this client and their lender remains yours.
Further reads
- Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms — Per-seat or pay-as-you-go: pricing models for legal AI tools.
- ChatGPT or a Legal AI Tool: Which Should a Small Firm Use? — Whether a general assistant or a legal tool suits your team.
- AI Implementation Plan for a Small Law Firm: The First 90 Days — A 90-day plan for bringing AI into a small legal practice.
- How to Train Staff to Spot AI-Written Phishing Emails — Training staff to spot convincing fake emails in a transaction.
- AI for Small Law Firms: What to Automate First — Which legal admin to automate first, beyond conveyancing.
- Can Solicitors Use ChatGPT Without Breaching Confidentiality? — The confidentiality question behind uploading client files.
- How to Automate Buyer and Vendor Follow-Up With AI — Map every follow-up moment, capture viewing feedback in a minute, let AI draft buyer nudges and vendor updates, and keep offers and bad news with a negotiator.
- Can AI Keep Mortgage Clients Updated During an Application? — Milestone update templates, the rules that stop AI predicting lender dates, channel choices including WhatsApp's template rule, and a 40-case example.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: LEAP Matter AI product page; Microsoft 365 Copilot Business pricing; OpenAI and Anthropic business plan data terms.