AI helps a landlord or agent with 20 to 60 units by sorting repair reports into urgency bands, drafting replies to routine tenant messages, preparing rent reminders and notices for a person to approve, and summarising each month's requests. It should never judge an emergency alone, send a legal notice unchecked, or receive sensitive tenant details.
The gain shows up as a consistent first reply, a log of what is still open and a few hours back each month; the number of repairs stays the same. That is why every draft below passes through a person before it reaches a tenant, and why the privacy checks come before the tools.
A month of tenant messages in 40 units, sorted
Picture a landlord with 40 units who handles about 90 messages a month. The mix below is illustrative, so count your own for a month before trusting it: 28 repair reports, 20 questions about rent and payments, 14 about access and entry, 10 neighbour or noise complaints, 8 move-in and move-out queries, and 10 others. The minutes are assumptions too. Replace them with a stopwatch reading from your own inbox.
| Type of message | Per month | Minutes each now | Minutes with an AI draft | Minutes saved |
|---|---|---|---|---|
| Repair reports | 28 | 6 | 3 | 84 |
| Rent and payment questions | 20 | 4 | 2 | 40 |
| Access and entry arrangements | 14 | 5 | 2 | 42 |
| Neighbour and noise complaints | 10 | 10 | 10 | 0 |
| Move-in and move-out admin | 8 | 8 | 5 | 24 |
| Everything else | 10 | 4 | 3 | 10 |
| Total | 90 | 200 |
Two hundred minutes is about three hours twenty a month. That is worth having and not a reason to reorganise a business. The larger gain is quality: every repair report gets the same first questions (a photo, two access windows, anything unsafe), so fewer messages bounce back and forth. Noise complaints show no saving on purpose. They go to a person from the first line, because a draft that takes a side can make a dispute worse.
Sorting repair reports into four bands
Four bands are enough for a small portfolio. The triggers matter more than the labels, and they should come from you and your tradespeople, not from the assistant.
| Band | Typical triggers | First response, my starting point |
|---|---|---|
| Emergency | Gas smell, carbon monoxide alarm, fire or smoke, flooding, water near lights or sockets, front door that cannot be secured, anyone at risk | Safety message at once, phone call within 15 minutes |
| Urgent | No heating or hot water, a leak that is contained, fridge-freezer failure, blocked only toilet | Same working day |
| Routine | Dripping tap, loose handle, appliance fault with a workaround, slow drain | Acknowledge within one working day, visit within one to two weeks |
| Cosmetic | Scuffs, marks, minor paint damage | Batch into a quarterly visit |
Those response aims are suggestions. Your tenancy agreements and the rules where the property is may set specific times, and damp and mould reports deserve early attention because of the health risks, so ask your adviser what applies before you copy my table. The tutorial on triaging maintenance requests goes deeper into the bands and into working with contractors.
For the emergency band, do not let the assistant improvise. Write three or four holding messages once (gas smell, water near electrics, fire or smoke, a door that will not lock) and have your gas engineer or electrician read them. The assistant then chooses a message by letter and never composes safety advice of its own. An example for water near electrics, to be reviewed by your electrician before use: "If water is dripping near a light, socket or appliance, do not use it. If it is safe to reach, switch that circuit off at the consumer unit. If you see sparks or smell burning, leave the home and call the emergency services. I am arranging a contractor now and will phone you within 15 minutes."
A leak report, triaged: the prompt, the output and the fix
You triage repair reports for a landlord with 40 units. Use only the
text of the message and never invent facts.
Bands:
- Emergency: gas smell, carbon monoxide alarm, fire or smoke, flooding,
water near lights or sockets, front door cannot be secured, anyone
at risk.
- Urgent: no heating or hot water, contained leak, fridge-freezer failure.
- Routine: can wait a week with a workaround.
- Cosmetic: appearance only.
Rate by what is described, not by the tone of the message. If two bands
could apply, choose the higher one. If the message is too vague to rate,
say so and list the questions to ask.
Return: band, a one-line reason, up to four questions to ask the tenant,
which approved holding message to send (A to D, or none), and what the
landlord should do next.
Message: [paste the message with names and phone numbers removed]A tenant might write: "Hi, water is coming through the bathroom ceiling and dripping near the light. Started last night, a bit worse this morning. I think the flat above is rented too. Can someone come today? Thanks, Unit 4." A good run of the prompt returns something like this.
Band: Emergency
Reason: water is dripping near a light fitting, an electrical risk.
Questions: Is the water touching the light or a switch? Is the ceiling
bulging or sagging? Can you reach the consumer unit? Any burning smell?
Holding message: B (water near electrics)
Next: phone the tenant within 15 minutes, call the plumber, try the
unit above, record the time and ask for photos.Leave out the line "rate by what is described, not by the tone" and an assistant can return Urgent for this message, because the tenant is polite and writes "a bit worse". Adding the tone rule and "choose the higher band" fixes that. The assistant also cannot see the ceiling, so its questions are the useful part: a person reads the answers and decides.
A vague message tests the other behaviour. "Heating's playing up again" should come back as Urgent pending answers, with questions such as which rooms are cold, whether the boiler shows an error, what the weather is doing and whether anyone in the home is elderly, very young or unwell. If the assistant guesses a cause instead of asking, tighten the prompt.
Log each report in a sheet so the month-end summary is easy to ask for. A filled-in row looks like this:
| Date | Unit | Issue | Band | First reply | Contractor | Access window | Status |
|---|---|---|---|---|---|---|---|
| 12 Oct | 4 | Bathroom ceiling leak near light | Emergency | 09:12 | Plumber called 09:20 | Today, 11:00-13:00 | Open, upstairs unit contacted |
Asking for the month-end summary
The same log feeds a monthly summary, which is where an assistant saves real effort. Paste the table (unit codes only) with a prompt like this.
Here is my repair log for October. Summarise it in under 150 words for
the landlord. List any unit with more than one report about the same
item within 60 days, name the three jobs open longest, and count the
reports in each band. Do not guess at costs that are not in the log.
I will calculate response times myself.A reply might read: "28 reports: 3 emergencies, 5 urgent, 17 routine, 3 cosmetic. Repeats: Unit 9 boiler (three reports in 52 days), Unit 14 extractor fan (two). Longest open: Unit 6 window seal, 19 days, waiting on a quote; Unit 2 shower screen, 17 days; Unit 11 cupboard door, 12 days. A quote for replacing the Unit 9 boiler may make more sense than a fourth repair." The last sentence is a suggestion for you to weigh, not a finding. The prompt keeps response times out of the assistant's hands because assistants can get a median wrong across a long column; a spreadsheet formula such as MEDIAN on a column of hours is the dependable way to get that number.
Testing the prompt on 20 old reports before a live one
Before the prompt sees a live message, run it on 20 past repair reports with names removed and compare its band with what you actually did. Mark each result as a match, over-cautious (a band higher than you chose) or under-cautious (lower). Over-cautious results are annoying and cheap. A single under-cautious result on an emergency means the prompt is not ready, however good the other 19 look.
Suppose the test returns 15 matches, 4 over-cautious results and 1 under-cautious one. The under-cautious one is a dripping tap rated Routine that you had treated as Urgent because the tenant had a newborn and the drip was onto a baby's room. The fix is a rule that raises the band when a message mentions a baby, an older person or anyone unwell, and that asks a question instead of assuming. Rerun all 20 after every change so that a fix for one case does not break another. Keep the 20 in a file, because you will want to rerun them whenever the assistant's underlying model changes, which vendors do often.
Reply drafts that promise nothing they cannot keep
The riskiest sentence in an AI-drafted reply is a helpful one. Left alone, assistants write "we'll have this fixed by tonight" or "sorry for the damage", which promises a date and edges towards admitting fault. Give it house rules in a short document, and paste in ten past replies you were happy with so it picks up your voice. The rules I would start with:
- Use the tenant's first name and keep it to one short paragraph.
- Say what happens next and who will contact them, never a repair date unless a contractor has confirmed one.
- Never state a cause or say whose fault it is.
- Ask for a photo and two access windows in the first reply.
- No legal language, no threats, no mention of the tenancy terms.
Before and after, for the leak above. The unconstrained draft read: "Sorry about the damage. We will get a plumber out today and the leak will be fixed by tonight." With the rules in place it became: "Hi [first name], thanks for the details. I'm arranging a contractor now and will phone you by 11:00 to agree a time. Please keep away from the bathroom light. If you see sparks or smell burning, leave the home and call the emergency services. A photo of the ceiling would help. [your name]". It admits nothing, promises only a phone call, and carries the safety line. The tutorial on resident and owner message templates has a longer library to start from.
Angry messages are where drafts help most, as long as the facts come from your log and not from the assistant's memory. A tenant writes: "This is the third time I've written about the boiler. Nobody has called me back and it's freezing." Paste in the message and the timeline from your log: reports on 3 and 9 October, an engineer visit on 6 October, a part on order. A usable draft: "Hi [first name], you are right to chase. I can see your reports on 3 and 9 October. The engineer visited on 6 October and ordered a part that has not arrived. I am phoning the supplier today and will call you by 16:00 with a firm date. If anyone at home is unwell or very young, tell me and I will look at temporary heating today." An unedited first pass might add "sorry for the delay on our side", which reads as an admission, so cut it. It might also offer a free heater, a commitment only you can make, so that line becomes "I will look at temporary heating".
Rent reminders and notices: what to hand over
A reminder needs wording once and a schedule. One pattern, to adapt to your agreements: a friendly email three days before the due date, a short text the day after if nothing has arrived, a phone call from a person after five days, and a decision by the landlord or agent after ten. The assistant writes the templates. It should not calculate arrears or late fees, because your ledger holds those numbers and a wrong figure in a reminder is hard to take back.
Keep text reminders short. A plain message of up to 160 characters fits in one segment, and Twilio's documentation notes that one emoji or a curly apostrophe can push a message into an encoding that fits only 70 characters per segment, so it splits and costs more. Names and amounts vary in length, so keep the template under about 140 characters with its placeholders. Two examples to adapt:
Before due date (129 characters):
Hi [first name], rent of $[amount] for [unit] is due on [date]. Pay as
usual via [method]. Questions? Reply here. [business name]
After due date (135 characters):
Hi [first name], we haven't seen this month's rent ($[amount], [unit])
yet. If it's on its way, thanks. If not, reply or call [number].For email, Gmail's schedule send holds up to 100 scheduled emails, which covers a 40-unit month of reminders with room for notices, though Google warns that a scheduled email may go a few minutes after its time.
Notices are different. A notice of entry, a rent increase or an arrears letter can have legal effect, and the required wording and notice periods depend on where the property is. The safe division of labour is below.
| Document | The assistant may | The assistant must not | Who signs it off |
|---|---|---|---|
| Notice of entry for a repair or inspection | Fill in unit, reason, date and time window from your log; flag missing fields | Choose the notice period | You, against the period in your agreement and local rules |
| Rent increase | Draft the plain-language covering letter | Write the formal notice or work out the new rent | A solicitor's or agent's master template |
| Rent arrears | Draft a soft first reminder | Calculate arrears, interest or fees; write a formal warning | A qualified adviser for anything beyond the first reminder |
| End of tenancy | Produce a checklist from your own process | Decide deposit deductions | You, with photographic evidence |
The tutorials on arrears letters and rent increase letters cover the rules each one must respect.
Privacy checks before tenant details touch an AI tool
Tenant messages hold more than people expect: names, phone numbers, who lives where, when the flat is empty, sometimes health or money worries. If you hold tenants' personal data, data-protection law such as the GDPR may apply. In practice that means having a reason for each item you keep, telling tenants how you use it, keeping it no longer than you need, and asking your adviser whether sending it to an AI provider needs a written contract. The checks below turn that into habits.
| Item in a message or file | Paste it into an assistant? | Do this instead |
|---|---|---|
| Repair description | Yes | Paste as written, minus names |
| Tenant's name and phone number | No | Refer to "Unit 4"; add contact details yourself afterwards |
| Bank details, ID or immigration documents | Never | Keep in your secure records only |
| Health, disability, age or family details | Never | Write "tenant has a mobility need" if the repair depends on it |
| Photos | Crop first | Remove faces, post and documents; keep the fault |
| Whole lease with names | No | Paste only the clause you are asking about |
Redaction is quick once it is a habit. Before: "Hi, I'm [name] in 12B, my number is [number], the lift has been out for a week and I use a wheelchair." After: "Tenant in Unit 12B reports the lift has been out for a week; the tenant has a mobility need." The repair is clear, and the name, number and health detail stay in your records. Contractors need a small set of details too: unit, fault, access windows, and the tenant's first name and phone number. Send those from your own email or work-order system rather than pushing them through an assistant, and tell the contractor not to share them further.
Account choices matter as much as wording. Business plans such as ChatGPT Business do not train on your content by default, according to OpenAI, while personal plans have a setting you should check. Every person who uses the assistant needs their own login, because the terms of both OpenAI and Anthropic forbid sharing credentials. Delete chats you no longer need, and tell callers when a call is recorded. The tutorial on keeping customer data private covers account settings in more detail. Nothing in this routine should decide who gets a tenancy. Applicant screening needs a separate process with its own care, and it is the one place where I would involve an adviser before any AI touches the decision.
A spreadsheet and an inbox first, software when the volume earns it
A do-it-yourself setup costs very little: the inbox you already use, a spreadsheet log, an AI assistant at about $20 a month on a personal plan (a business plan needs at least two seats), and a form for repair reports. On Jotform's pricing page, the free Starter plan allows 5 forms, 100 submissions a month, 500 stored submissions and 100 MB of space. Phone photos at a few megabytes each will fill 100 MB quickly, so either ask tenants to email photos or move to the Bronze plan, listed at $39 a month ($34 billed annually) with 1 GB.
Property platforms now sell AI as an extra layer. DoorLoop's pricing page lists three plans: One at $2.50 per unit a month, Connect at $4.00, which adds a receptionist agent answering tenant texts, calls and portal messages, and Multiply at $5.00, which adds a maintenance agent and a lease renewal agent. Each has a monthly minimum of $199, $299 and $399. At 20 to 60 units the minimum, not the per-unit rate, sets what you pay. The page also says agents start out needing your approval for everything, and that annual billing costs less than monthly, so check which price the toggle shows.
A quick sum with the figures above. The step from One to Connect costs $100 more a month at your size. If the receptionist agent saves the three hours twenty estimated earlier and your time is worth $25 an hour, that is about $83 of time, which does not cover $100. The case for it would be out-of-hours cover and consistency rather than time saved, and the platform also does rent collection, accounting and owner statements that you may be paying for elsewhere, so compare it with your whole stack. For the call side alone, the tutorial on AI or human answering services for property managers compares the options.
Three numbers that show it worked after a month
- Time from report to first reply. Read it from your log. If the median falls and no emergency waited longer than your own aim, the triage is working.
- Complete reports on first contact. Count how many repair reports arrived with a photo and two access windows after the first exchange. Rising numbers mean the questions in your replies are doing their job.
- Corrections and misses. Keep a tally of drafts you had to correct and of anything the assistant rated lower than you would have. One missed emergency means the prompt and holding messages need work before you use it again.
If you would rather map this with someone, that is what my AI implementation consultation covers; the contact page has the details.
Questions landlords ask before using AI with tenants
Can an AI chatbot answer tenants directly instead of me?
It can answer routine questions from your leases and policies, and some property platforms sell this as a receptionist agent that handles texts, calls and portal messages. I would start with drafts that a person sends. If you go further, limit it to routine topics, route anything unsafe straight to a person, and test it on 20 real past messages first.
Do I have to tell tenants an AI is involved?
Say in your privacy notice how tenant data is used, and say plainly when a tenant is talking to an automated assistant. If you have tenants in the EU, the AI Act's duty to make people aware they are dealing with an AI system has applied since 2 August 2026 and falls mainly on the provider. Ask your adviser how it applies to you.
Which repairs should never be left to AI triage?
Anything involving gas, carbon monoxide, fire or smoke, flooding near electrics, structural damage, or a tenant who says they are at risk. The assistant's only job there is to spot the words and alert a person at once. It should never decide that such a report can wait.
Is it safe to paste a tenant's photo into ChatGPT?
Photos often carry more than the fault: faces, post with names, documents on a table. Crop to the damage and remove anything personal first. ChatGPT accepts PNG, JPEG and non-animated GIF files up to 20 MB each. The picture can help describe what is visible, but a tradesperson has to assess the fault.
Further reads
- AI for Letting Agents: Handle Tenant Queries Without Extra Staff — The agent's view: handling tenant queries without extra staff.
- Property Maintenance Software: Jobber, ServiceM8 and Fergus Compared — Work-order software compared, for when a spreadsheet stops being enough.
- How Property Managers Use AI to Screen Tenant Applications Fairly — Why applicant screening needs more care than repair triage.
- How Much of a Property Manager's Week Can AI Take Over? — How much of a manager's whole week AI can realistically take.
- Can ChatGPT Draft a House Rental Lease? What Landlords Must Check — Prompts that make ChatGPT flag local-law terms, a check for invented numbers, and the pre-signing list for a house let.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: DoorLoop and Jotform pricing pages, OpenAI's help pages on file uploads and ChatGPT Business data use, Google's Gmail help page on scheduled sending, Twilio's SMS length documentation, checked 29 October 2026. Volumes, minutes and sample messages are illustrative.