Best AI for Conveyancing: ID Checks, Searches and Client Chasing

For conveyancing, use a specialist ID and source-of-funds service for identity checks, a business-grade assistant such as ChatGPT Business or Microsoft 365 Copilot to summarise search results and draft routine chasing emails, and your case-management system for reminders. A fee earner reviews every output, and no client data goes into a free tool.

The tool should follow the transaction stage, not the other way round. A firm that buys a platform first and finds the job later usually pays for features nobody opens. The sections below take a file from instruction to completion and name the job AI does at each point, with the check you keep.

Follow me on Instagram@sagnikteaches

The jobs, by transaction stage

A residential file has the same shape every time, which is why AI helps. Four stages carry most of the repetitive work.

Connect on LinkedInSagnik Bhattacharya

Notice what is missing: AI does not decide title issues, advise on risk or approve ID. Its job is to cut the reading and typing so the fee earner has more time for judgement.

Subscribe on YouTube@sagnikteaches

ID checks and source of funds

This is the stage where a purpose-built service beats a general assistant by a wide margin. A chat tool can describe a passport. It cannot run liveness checks, validate a document's security features or hold an audit trail your regulator will accept.

Specialist services for property work, such as Thirdfort, run remote checks through an app. Thirdfort's own site says the check typically takes a client a couple of minutes once they start, that the app stops clients submitting poor-quality photos of ID documents (which cuts repeat checks), that there are alternative routes inside the platform for clients who will not use the app, and that source of funds runs on access to banking data plus a questionnaire. The site does not publish prices; it describes a monthly platform fee plus either pay-as-you-go checks or bulk credits, which it says stay valid for 24 months. So ask any vendor for a written quote for your volume, and ask what the quote excludes.

What to put on the vendor checklist

• Which certifications or standards it claims, and a link to the evidence. A claim on a marketing page is a starting point, not proof.
• Where client data is stored, for how long, and how deletion works when a file closes.
• Whether the report feeds your case-management system or needs re-keying.
• What happens when a client fails the check: the fallback route, and who the firm calls.
• How the firm exports its records if you leave the vendor.

Our tutorial on what a fee earner should test in a conveyancing AI demo turns this into a demo script. For the onboarding forms and reminders around the check, see conveyancing client onboarding with AI.

Where an assistant does help at this stage

Once the specialist report exists, an assistant can draft the client-facing wording around it: a message explaining why the check is needed, how to do it, and what happens if a document fails. Write that once, have a partner approve it, and reuse it as a template. An assistant can also turn a source-of-funds explanation into a plain checklist for the client ("have your last three months of statements ready, plus the letter confirming the gift"), which cuts the back-and-forth emails that eat a morning.

Searches: using AI to read, not to rule

Search packs are long and repetitive, which makes them a fair job for a summarising tool, so long as the fee earner reads the original. The risk is a clause the summary flattens. The safe use is "point me to the parts to check", not "tell me whether it is fine".

You are assisting a conveyancing fee earner. Below is the text of a local authority search result, with names and addresses removed.
Produce:
1. A list of every entry that mentions a road, a planning matter, a notice, a charge or a restriction, with the page or section reference.
2. A list of entries that are blank or say "not known", because I will need to raise enquiries.
3. A list of anything that looks inconsistent with the other entries.
Do not state whether the property is acceptable. Do not give legal advice. Quote the original wording for each item.

Illustrative output for a drainage and water report: eleven entries listed with section references, three marked "not known", and one flagged because the report names a public sewer while another section says private drainage. The fee earner opens the report at those sections. The flag turned out to be a generic caveat, but the "not known" entries became enquiries. The tool saved the first scan. It did not make the decision.

Redaction before you upload

If your tool's terms do not clearly cover client data, remove names, addresses, title numbers and anything personal before pasting. A simple rule works: the model only ever sees text with "[client]" and "[address]" in place of those fields. Keep the original as the file record, and keep the redacted copy only as long as you need it.

For the longer-form job of drafting a report on title and the checks you keep, read the tutorial on AI and the report on title.

Client chasing without nagging or leaking

Chasing is where most files lose days. The seller's side is slow with a form, the buyer has not sent a document, a lender wants a certificate. Most of the lost time goes on remembering who owes what, rather than on drafting. So the biggest gain is usually the reminder rule in the case-management system, with an assistant used only for tone.

1. Define the triggers. For example: ID not completed three days after the link is sent; property information form not returned after five days; a deposit not in the account two days before the exchange target.
2. Write three message levels. A friendly first reminder, a firmer second with the consequence, and a phone-call prompt for the fee earner.
3. Keep the messages free of detail. A reminder should say what is outstanding and where to upload it, never repeat the property address, price or bank details.
4. Log it. Each reminder lands on the file so a colleague can see what the client was told.

Write three versions of a reminder to a client who has not completed their ID check. Version 1: friendly, sent three days after the link. Version 2: firmer, sent on day six, explaining that we cannot move the matter forward until the check is done. Version 3: a short script for a fee earner phoning the client.
Keep each under 90 words, avoid jargon, include the placeholders [first name], [your name] and [link]. Do not mention the property or any money figures.

A realistic sample of the second version (illustrative): "Hello [first name], we're still waiting for your ID check to be completed, which means we can't yet send documents to the other side. It takes about five minutes in the app: [link]. If you're having trouble, reply here or call me and we'll sort it together. [your name]". It is clear, it holds no sensitive detail, and a fee earner can send it after reading it once.

A worked month: a four-fee-earner practice

Say a four-fee-earner residential practice opens 40 files a month. Illustrative time figures follow, based on the firm's own estimate rather than a measured study.

Thirty-two hours is about four working days across the practice. Against that, count the costs: a specialist ID service's per-check charge, a business-grade assistant at, for example, $25 a seat monthly or $20 on annual billing for ChatGPT Business, or $21 per user a month on annual billing for Microsoft 365 Copilot Business (Microsoft shows a first-year promotional price of $18 until 31 December 2026), plus the time to write and approve templates. Check the vendor's current page for each price. The firm's own figures, not mine, should decide whether it works. If the "after" column is wrong by half, it is still positive. If it is wrong by three quarters, you have a very expensive reminder system.

Supervision, confidentiality and what regulators have said

Legal regulators have started to spell out what they expect. One solicitors' regulator published a warning notice on the misuse of AI on 17 August 2026, and its main points are worth putting in the firm's policy whoever regulates you:

• Using AI does not reduce or transfer professional responsibility. Solicitors remain accountable for all work and advice, whether or not AI was used.
• Client information should go into AI systems only where appropriate contractual, technical and organisational safeguards are in place. Data should stay in a secure environment, should not be used to train models unless authorised, and should not be kept longer than necessary.
• Putting client correspondence into an open, public AI tool is likely to breach confidentiality, and privilege may be lost for good.
• Supervisors remain accountable for work done through others, and firms need effective systems, controls and governance for AI risk.

Check what your own regulator has published, and ask your compliance officer to confirm what applies to your firm. This tutorial is practical guidance, not legal advice.

A one-page AI use rule for a conveyancing team

1. Only approved tools, listed in the firm's policy, may touch client files.
2. No client name, address or title number goes into any tool that has not been vetted for contractual data terms.
3. Every AI output used on a file is reviewed by the fee earner, who is named on the file.
4. Bank details are never sent, changed or confirmed by AI-drafted email; they are verified by phone to a known number. See a call-back procedure for changed bank details.
5. Any error caused or missed by an AI output is logged and discussed monthly.

Examples of the wording on a live file

Abstract advice about "reviewing outputs" is easy to nod at and forget. Three short examples show what the review looks like on the day.

A search summary, before and after the fee earner's pass

The assistant's first output on a redacted environmental report (illustrative): "No significant flood risk identified. Radon: not in affected area. Contaminated land: no entries." The fee earner opens the report and finds the flood section says "low to medium risk on surface water", with a note about a nearby watercourse. The summary had rounded "low to medium" to "no significant". After the pass, the file note reads: "Surface water risk low to medium per section 4; advise client to check insurance availability; raise enquiry on drainage history." The lesson is to include in the prompt: "Quote the original wording and do not soften or round any risk rating."

A chasing log entry that survives a handover

If a fee earner goes on leave, a colleague should know what the client has been told. A good log entry has the date, the channel and the outstanding item: "4 Nov, email reminder 2 sent: ID check outstanding, link resent. Next step: phone on 6 Nov if still open." An assistant can draft these from the sent message in seconds, which makes a thorough log cheaper than a thin one. Keep the entry free of personal data beyond what the file already holds.

A client update that explains a delay without alarm

Say a seller's solicitor is slow to answer a replies-to-enquiries letter. The assistant drafts: "Hello [first name], a quick update. We sent our questions about the property on [date] and we're waiting for the other side's answers. We've chased once and will again on [date]. There's nothing you need to do at the moment. I'll write again by [date] with news. [your name]". The fee earner changes one thing: the promise to write by a date, which is only kept if the diary entry exists. Never let a template promise a date the firm does not diarise.

Data protection questions to settle before a go-live

Whichever tool you choose, ask your compliance officer or data-protection lead to confirm these in writing before the first live file. Each takes a short conversation, not a project.

• Role of the vendor. Is it a processor acting on your instructions, and does the contract say so? A processor agreement is a normal request that serious vendors expect.
• Training use. Does the vendor use your content to train models? Business plans of the major assistants say they do not by default; consumer plans often have a switch, and the label differs by vendor. Get the business plan's wording from the vendor's own terms.
• Retention and deletion. How long are inputs kept, and can you delete a matter's data when the file closes?
• Location of processing. If you have clients who expect their data to stay in a particular region, check the vendor's storage options. Some vendors offer a region choice only on their enterprise tiers; others document none.
• Access. Who at the vendor can see your content, and under what conditions? Also check how your own staff sign in; a shared login defeats the audit trail.
• Impact assessment. Where a new tool processes client personal data at scale, a short written assessment before go-live is sensible. Your data-protection lead will know the format.

If a vendor cannot answer these in plain language within a week, treat that as information about how they will handle a problem later.

Training the team in an hour

Most failures are habit failures, so short, specific training pays. A one-hour session covers four things. First, the approved tools and the redaction rule, with a live demonstration using a fictitious file. Second, the three prompts the firm uses (search points, reminder wording, client update) pasted into a shared document. Third, a "spot the error" exercise: give the team an AI summary of a mock search with two planted mistakes and see who finds them. Fourth, the escalation route when something looks wrong.

Repeat the spot-the-error exercise each quarter with a fresh example. It keeps the review habit alive after the novelty wears off. What staff should understand about AI goes wider than these four points; what staff need to know about AI covers the broader picture.

Costs that do not appear on the vendor's page

The subscription is rarely the biggest cost. Count these as well, using your own figures:

• Partner time to approve templates and the AI use rule, often a half day once.
• Fee earner time to review outputs. If review takes as long as doing the job, the tool is not helping; track it for the trial.
• Integration work, if the ID report has to flow into the case-management system. Ask the vendor whether the connection is standard or custom, and who pays for custom work.
• Switching costs. Ask how you would export records and how long the vendor keeps them after you leave.
• Client-facing change: some clients will not use an app and need the paper route, which means running two processes for a while.

Add these to the monthly saving table above, and compare the total with the saving. A tool that looks cheap per check can still lose money if each check needs ten minutes of manual correction.

Choosing the tool: a short decision path

Work through these questions in order. Stop at the first "yes".

1. Do you lose more time on ID and source-of-funds paperwork than anything else? Start with a specialist service, and get a written quote.
2. Do you already have case-management software with automated reminders? Switch the reminders on before buying anything new. Many case-management systems already include reminder rules that nobody has set up.
3. Does your team read many long reports each week? Add a business-grade assistant with the redaction rule.
4. Do clients complain about not knowing where their matter is? Write update templates and a weekly status message, and trial an assistant to draft them.

Run a four-week trial on 10 live files and measure three numbers: minutes saved per file, number of AI-output errors caught in review, and client complaints about communication. If the errors caught are rising, tighten the review rule before widening the use. For more admin jobs worth testing, see how conveyancers use AI to cut admin on each transaction.

Mistakes that show up in the first quarter

The summary that dropped a restriction. A fee earner relied on a three-line summary of a lengthy document and missed a restriction buried in a schedule. The fix was procedural: the summary now lists every schedule and the fee earner ticks each one against the original.

The reminder that said too much. An automated message repeated the property address in an email sent to a shared inbox. The template was rewritten to name only the outstanding task and a secure upload link.

The tool nobody opened. A firm bought ID software and kept sending clients paper forms because the partner preferred them. The answer was a rule that the digital route is the default, with paper as the exception that needs a reason on the file.

These are illustrations of common patterns, not accounts of particular firms. The lesson is the same in each: the tool is rarely the problem, the review step and the habits around it are.

One stage, one file type, four weeks

Pick one stage and one file type. For most firms that is ID and onboarding on residential purchases. Write the template messages, ask two vendors for written quotes, and sit a fee earner in front of each demo with a real, redacted file. Agree the review rule before the first live file. If you want help deciding which stage to automate first, that is what my AI implementation consultation covers; see the contact page.

Questions conveyancers ask about AI tools

Can AI replace the ID check in a conveyancing file?

A specialist ID and source-of-funds service can run the check and give you a report, but the firm stays responsible for the decision. A general chat assistant should never be the ID check. Keep a named person who reviews every report and records the outcome on the file.

Is it safe to paste search results into ChatGPT?

Search results contain client and property details. Use only a business-grade tool whose terms state that your content is not used for training, with a data agreement your firm has reviewed. Otherwise strip names and addresses first, and keep the original report as the record.

How much do conveyancing AI tools cost?

Specialist ID services typically charge a platform fee plus per-check or bulk credits and publish few prices, so ask for a written quote. General assistants are priced openly per seat, for example ChatGPT Business at $25 a seat a month, or $20 on annual billing, and Microsoft 365 Copilot Business at $21 per user a month on annual billing.

Who is responsible if the AI misses something in a search?

The fee earner and the supervising solicitor. Regulators have said using AI does not reduce or transfer professional responsibility. Treat AI output as a reading aid that points you to clauses to check, never as the advice itself.

Further reads

• ChatGPT or a Legal AI Tool: Which Should a Small Firm Use? — ChatGPT or a legal AI tool for a small firm.
• Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms — Per-seat or pay-as-you-go pricing for small firms.
• How to Summarise Bundles and Transcripts With AI Safely — Summarising long bundles safely.
• How Small Law Firms Use AI to Draft Letters and Routine Documents — Drafting routine letters in a small law firm.
• AI Readiness Checklist for Accountants, Solicitors, Consultants — A readiness checklist before rolling tools out.
• AI Time Capture for Solicitors: Recover Hours You Never Billed — Recover unbilled hours with AI time capture.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: a solicitors' regulator's warning notice on the misuse of AI (published 17 August 2026), Thirdfort website (ID, source-of-funds and pricing model), OpenAI ChatGPT Business pricing page, Microsoft 365 Copilot Business pricing page.