Yes for the words, no for the numbers. Your client-accounting or property software should keep producing the statement figures; AI can turn those figures, plus the month's maintenance and tenancy notes, into a plain-English cover update for each landlord in a minute or two. A person checks every figure against the ledger before anything is sent.
The split exists because language models are not calculators, and in client money a wrong number is a trust problem and can be a regulatory one. Asked to "summarise the statement", a model will happily round a fee, net off two lines the wrong way, or mention a rent payment that doesn't exist because the previous eleven months had one. So the AI never computes. It reads a ledger export, copies figures exactly, and writes the sentences around them.
What landlords actually want each month
Most landlords open a statement looking for four answers: how much did I receive, is the rent being paid, did anything happen at the property, and do I need to decide anything? A statement answers the first. The cover update answers the rest, which is why landlords phone the office when there isn't one.
- Money: rent received, deductions (fees, repairs), net paid out, and the date it was paid.
- Rent status: paid in full, paid late, or in arrears, stated neutrally.
- What happened: repairs completed, inspections done, anything reported and pending.
- Coming up: tenancy end or renewal dates, certificate expiries, planned works.
- Decisions needed: quotes to approve, renewal terms, rent review.
Put decisions first. A landlord who reads only the top two lines should still know whether they need to reply.
The data each update needs, and where it comes from
| Item | Source | Export as | Checked by |
|---|---|---|---|
| Rent, fees, repair costs, net payout | Client-accounting or property software | Statement CSV, one row per property | Accounts, against the statement PDF |
| Arrears status | Rent ledger | Column in the same CSV | Accounts |
| Repairs completed and pending | Maintenance log or works orders | Short list per property | Property manager |
| Inspections | Inspection reports | Date plus two-line summary | Property manager |
| Certificate expiries, tenancy dates | Compliance tracker | Dates within the next 90 days | Compliance lead |
| Decisions needed | Property manager's notes | One line each, with any quote amount | Property manager |
If those pieces live in different systems, the first month's work is building one monthly export that brings them together per property. A spreadsheet with one row per landlord and one column per item is enough. That export, not the AI, is the heart of the process, and it is also what makes the monthly batch take minutes rather than hours.
Add one more column while you are building the export: how each landlord likes to hear from you. Some want every detail monthly; some want a note only when something needs deciding; a landlord with a large portfolio may want one summary table rather than twenty notes. Record it as a simple code (full, decisions-only, summary) and tell the AI to follow it. An accidental landlord with one flat and a busy professional with twelve don't want the same email, and respecting that preference is the cheapest personalisation you will ever add.
The same data row reads very differently under each code. For a decisions-only landlord, the September row further down becomes two lines: "One decision this month: the tenant would like to renew when the fixed term ends on 31 December. Shall we offer the current rent, or would you like to discuss a review? Your statement is attached." For a summary landlord, the AI's job is laying out a table, one row per property, copied from the export:
| Property | Rent | Net paid | Needs you? |
|---|---|---|---|
| P-417, 2-bed flat | Paid in full | 927.00 | Renewal decision |
| P-418, 1-bed flat | Paid in full | 742.50 | No |
| P-431, 3-bed house | Paid 6 days late | 1,212.00 | No |
| P-440, studio | Void since 3 Sept | 0.00 | Price review call |
In an early test the AI helpfully added a "Total" row. It came to 2,871.00 against a true 2,881.50: it had dropped the pence on one figure and made an addition slip on the rest. Portfolio totals come from the accounts system as their own column, or not at all. Add "never add a total row" to the prompt for summary landlords.
A prompt that writes around the numbers without touching them
You write monthly cover notes to landlords for [agency], to accompany
their statement. Use ONLY the data row provided.
Number rules:
- Copy every figure exactly as given, including pence. Never add,
subtract, round or calculate anything.
- If a figure you would need isn't in the row, write [CHECK: missing].
Content, in this order:
1. Decisions needed (or "No decisions needed this month")
2. Rent: received / late / arrears, one neutral sentence
3. Money: rent received, deductions, net paid and payment date
4. What happened at the property
5. Coming up in the next 90 days
Style: plain, warm, under 170 words, no exclamation marks. Never
speculate about a tenant's circumstances or reasons for late payment.
Sign off as [property manager name].
Data row: [paste]Here is an illustrative data row and the note it produces:
landlord=Ref L-208 | property=2-bed flat, ref P-417 | rent_due=1,150.00
rent_received=1,150.00 | received_date=01/09 | mgmt_fee=138.00
repairs=85.00 (bathroom extractor fan replaced 12/09)
net_paid=927.00 | paid_date=05/09 | arrears=none
pending=kitchen tap dripping, plumber booked 30/09
coming_up=electrical safety certificate expires 14/11; tenancy fixed
term ends 31/12 | decision=approve renewal on current rent? (tenant
wishes to stay)Decision needed: the tenant would like to renew when the fixed term ends on 31 December. Would you like us to offer a renewal at the current rent, or discuss a review? Rent: September's rent was received in full on 1 September. Money: rent received 1,150.00, less management fee 138.00 and repairs 85.00 (bathroom extractor fan replaced on 12 September), net paid 927.00 on 5 September. At the property: a dripping kitchen tap has been reported and a plumber is booked for 30 September. Coming up: the electrical safety certificate expires on 14 November and we will arrange the renewal inspection in October.
What the check changes: the figures all match the row, but the last sentence promises "we will arrange the renewal inspection in October", which nobody decided. The property manager either confirms that is the plan or changes it to "we'll be in touch about the renewal inspection". Watch for promises in the "coming up" section; it is where the AI fills gaps with plans.
A before and after landlords notice
Before: each landlord receives the statement PDF with a one-line email, "Please find attached your statement for September." A landlord whose net payment dropped by $85 sees the drop, not the reason, and phones. After: the cover note explains the $85 in its first lines, and a decision that would have waited for the landlord's next call is asked for directly. In this illustrative agency, the change that matters isn't the time saved writing; it's the calls that no longer happen, and the renewals decided in September rather than in a rush in December.
The awkward months need a person's touch
Most months, most properties are uneventful. The notes that matter are the ones that aren't, and those need a senior person's eye:
- Arrears. State the position and what the agency is doing, following your arrears process. Never speculate about why ("the tenant may be struggling financially"). That is tenant personal data and guesswork.
- A large repair bill. Lead with what happened and why the work was needed, then the cost, then whether it was within the agreed approval limit.
- A void period. Say what marketing is running, viewings so far and any feedback, and offer a call.
- A complaint or dispute. Take it out of the batch entirely. Write it by hand.
Flag these rows in the export (an "exception" column works), so the AI drafts them separately and they go to a senior manager for approval rather than into the routine check.
An arrears month shows why. The illustrative row: rent due 1,150.00, received 575.00 on 4 September, arrears 575.00, reminder sent 6 September, call with tenant booked 10 September. The AI's first draft of the rent sentence:
Unfortunately the tenant seems to be going through a difficult time and was only able to pay half of September's rent, but they have assured us the balance will follow shortly.
Nothing in the row says the tenant is having a difficult time, and nobody has been assured of anything; the call hasn't happened yet. The approved version:
September's rent was part-paid: 575.00 received on 4 September, leaving 575.00 outstanding. We sent a reminder on 6 September and are speaking to the tenant on 10 September. We'll update you after that call, and before the next statement if the position changes.
It is shorter, it gives the landlord everything they can act on, and it doesn't put a guess about someone's circumstances in writing.
Void periods are the other note landlords read closely, and the AI tends to be relentlessly upbeat about them. Give it the viewing count and the actual feedback, and let it report both: "Six viewings since 3 September. Two viewers said the second bedroom felt small for the rent; none has applied. We'd like to talk about the asking rent this week." A landlord who hears "lots of interest" in week three and a price reduction in week four stops trusting the notes.
The two-minute check before each one goes out
- Right landlord, right property. Check the property reference in the note against the recipient. Sending one landlord's figures and tenant details to another is the single most damaging error in this process, and it is a data breach, not just an embarrassment.
- Every figure against the statement. Rent received, each deduction, net paid, dates. The prompt says copy exactly; check it did.
- No [CHECK] markers left. Search the batch for them before sending anything.
- No promises nobody made. Especially dates for works and inspections.
- Nothing about the tenant beyond what the landlord needs. Paid, late or in arrears, yes. Circumstances, no.
A realistic mistake to design against: in one illustrative agency, the AI was given the whole month's export at once and asked to "write a note for each landlord". For a landlord with two flats, it combined both properties' figures into a single net payment line that matched neither statement. The fix was structural: one row, one property, one note, and landlords with several properties receive one note per property, or a note built from each property's rows separately. Setting up human review without slowing down covers how to keep a check like this quick once it becomes routine.
Running the monthly batch
The day after the statement run, export the combined spreadsheet. Then choose one of two routes, depending on volume.
Up to about 60 landlords: run the batch inside a business AI plan. Upload the spreadsheet to a Project with the prompt saved, and ask for notes in groups of ten, each labelled with its reference. Copy each approved note into the email with its statement. ChatGPT Business and Claude Team keep your content out of model training by default; a consumer plan is not the place for landlord and tenant data. If you'd rather keep tenant names out entirely, the methods in anonymising client data before it goes into AI work well here, because the note only needs a property reference.
Above that: automate it. A Make scenario or Zap reads each spreadsheet row, sends it to the AI with the prompt, and saves the result as a draft email in Outlook or Gmail, with the statement PDF attached and the property reference in the subject line. Nothing sends until a person has checked it. At one AI step per row, 150 landlords a month is modest volume for either platform's paid plans. Agencies that already produce client reports this way in other sectors use the same pattern; how marketing agencies automate client reporting shows it in a different setting.
140 landlords on one statement day
Put it together for a hypothetical agency with 140 landlords, one property each for simplicity. On statement day plus one, the accounts assistant exports the combined sheet (20 minutes once the export is set up). The automation drafts 140 notes in a few minutes. Around 110 are uneventful; checking each against the statement takes about a minute, so under two hours for the routine batch. The remaining 30 carry a decision, a pending repair or an exception; the property managers check their own, at three to five minutes each, about two hours between them. The four genuinely awkward notes go to the senior manager.
That is roughly four staff-hours for 140 personalised updates, work that simply wasn't done before. The time is real, so decide who does it, and on which day, before switching it on. It also moves effort from the phone to the page: how much of a property manager's week AI can take over puts this in the context of everything else.
Quarterly and year-end summaries
The same approach works for longer periods. For a year-end summary, export income and expense totals by category from the accounts system, plus the year's significant events per property, and ask the AI for a one-page narrative: occupancy, rent changes, major works, and what is planned for next year. The same number rules apply. Many landlords pass the summary to their accountant, so keep it descriptive: the AI should not comment on what is deductible or how anything will be taxed. That is for the landlord's own adviser.
It will try, though. An illustrative year-end draft for a three-bedroom house read well until its fourth sentence:
The property was let for the full twelve months, with rent rising from 1,250.00 to 1,300.00 at the March renewal. The main expense was the replacement boiler in January at 2,340.00. Two routine inspections found the property well kept. The boiler replacement should be treated as a repair rather than an improvement, which may reduce your tax bill for the year.
Strike that last sentence. Whether a boiler replacement is a repair or an improvement is exactly the kind of judgement that depends on the rules where the landlord is taxed and on facts the agency doesn't hold. The note can list the expense and attach the invoice; the landlord's accountant decides what it means. Add "never comment on tax, deductibility or allowances" to the year-end prompt, and read the final paragraph of every summary with that in mind, because that's where the AI puts its advice.
Further reads
- AI for Letting Agents: Handle Tenant Queries Without Extra Staff — The tenant side of the same communications workload.
- AI Inventory and Check-Out Reports: A Letting Agent's Guide — Move-out findings are often the hardest landlord update to write.
- How to Classify Business Data Before Using AI Tools — Decide which ledger and tenant fields may go into an AI tool.
- How to Check AI Is Doing Good Work, Not Just Fast Work — Keep checking accuracy once the monthly batch feels routine.
- How to Build a Shared Prompt Library for Your Team — Store the cover-note prompt so every manager uses the same one.
- AI Email Triage for Professional Firms: Sort, Summarise, Draft — Sort the landlord replies your updates generate.
- How Property Managers Use AI to Triage Maintenance Requests — Write your urgency tiers down, let AI ask the missing questions and classify each request, keep safety rules outside the model, and test on last quarter first.
- How Property Managers Use AI to Screen Tenant Applications Fairly — Written criteria, one summary format for every applicant, human decisions with recorded reasons, and a monthly check that your rules aren't quietly unfair.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: ChatGPT Business and Claude Team plan pages (data use and pricing); Zapier and Make pricing and help pages; Microsoft 365 Copilot plan pages.