Free AI for Property Management: What a Small Portfolio Can Do

A manager with up to a few dozen units can run three jobs on free plans: sorting repair requests into a log, drafting listings and replies to enquiries, and pulling key dates out of leases and inspection notes. Pay when you hit the daily upload cap, need tenant data kept out of training by default, or share the work with a colleague.

The limit isn't the assistant's quality. It is what you feed it. A repair message contains a name, an address and sometimes a photo of the inside of someone's home, and a lease holds rent, deposit and signature details. Free consumer plans were designed for individuals, so the working rule for a landlord is to paste the problem, not the person.

Follow me on Instagram@sagnikteaches

What free plans cap when you manage a dozen units

The figures below are the vendors' own published limits, read on 1 November 2026. Where a vendor publishes no number, the table says "not stated" instead of guessing. Landlord jobs are document-light but photo-heavy and privacy-heavy, so the rows follow that.

Connect on LinkedInSagnik Bhattacharya

Do the portfolio sum before you pick a tool. A manager abstracting 12 leases in a month needs 12 uploads. ChatGPT Free allows 3 a day, so that is four separate days, while Claude Free takes all 12 in one chat (the 20-file limit) though a long lease will draw heavily on the five-hour usage allowance. Photos behave differently: a leak photo is one image, but a tenant who sends eight photos of a damp wall uses most of a Gemini prompt.

Subscribe on YouTube@sagnikteaches

The fourth row shapes everything that follows. All three consumer services keep or review some chats, and tenant messages are personal data about people who never agreed to be used as training material. That doesn't make free tools unusable. It means the workflows below are built so that the pasted text contains no tenant at all.

Workflow one: a repairs inbox that sorts itself

Start with a spreadsheet, because the log is the asset and the AI is only the sorter. Columns: date, unit code, issue, urgency (1 to 3), trade, first reply sent, contractor, status and cost. Unit codes such as U4B stand in for names and addresses, and the key that links codes to tenants stays in a separate sheet that never goes near a chat.

Paste messages in batches of five to ten, with the unit code in place of any name, and give the assistant rules it can follow.

You triage repair requests for a small residential portfolio. For each
message, return a table with: Unit, Issue in 8 words, Urgency, Trade
needed, Questions to ask the tenant, and a two-sentence holding reply.
Urgency: 1 = risk to safety or severe damage today; 2 = loss of
heating, hot water, security or a toilet within 24 hours; 3 = routine.
If a message mentions gas smell, fire, flooding, exposed wiring or
no heating in freezing weather, set Urgency 1 and write CALL TENANT
NOW in the Issue column. Do not guess causes. Do not promise times
or costs in the holding reply.

U4B: Bathroom tap has dripped for a week, now a brown patch on the
ceiling below it.
U7A: No hot water since last night, boiler shows an error light.
U2C: Smell of gas in the hallway, not sure where from.

A reply in the right shape, shown as an illustration, puts U2C first: "CALL TENANT NOW: possible gas smell in hallway", urgency 1, a gas engineer or emergency line, a question about whether anyone is indoors and windows are open, and a holding reply telling the tenant to leave the property and phone the gas emergency number. Next comes U7A (no hot water, boiler error light, urgency 2, heating engineer, ask for the error code and whether heating works), then U4B (dripping tap with a ceiling stain, urgency 2, plumber, ask whether the stain is growing and whether the unit below is affected).

What you fix matters more than what it writes. First, the urgency scale is your rule: change the thresholds to suit your units and your insurer. Second, a gas smell or a fire report is never answered by an AI draft. A person phones the tenant first, and the log entry follows. Third, read the holding replies for promises. The prompt forbids times and costs, and a tool can still offer "someone will be there tomorrow", so delete that line. Fourth, a stain on a ceiling is the kind of message a tool can under-rate, because water damage grows quietly. If the tenant says it is spreading, move it to urgency 1 yourself.

Now the sum. Suppose a 12-unit portfolio logs about 18 repair requests a month. By hand, reading, logging and writing a first reply takes around 6 minutes each: 108 minutes. With batch triage it takes around 2 minutes each to check the table and paste it into the log: 36 minutes. That saves 72 minutes a month, or 1.2 hours. It is a modest gain, and the larger benefit is consistency: every request gets the same questions asked, so the contractor's first visit is less often wasted. For the fuller approach to routing and follow-up, see how property managers use AI to triage maintenance requests.

Workflow two: listings, viewings and enquiry replies

A listing is the easiest job to hand to a free tool, because it needs no tenant data, only facts about the unit. The risk is the opposite one: the tool adds facts you never gave it. The fix is a fact sheet and a hard instruction to use nothing else.

Write a rental listing from the fact sheet below. Use only these
facts. If something is not on the sheet, leave it out. Headline of
under 70 characters, a 90-word description and five bullet points.
Describe the property, not the type of person who should rent it.

Fact sheet: two-bedroom flat, first floor, no lift. Gas central
heating, boiler fitted 2024. 62 square metres. Unfurnished. One
allocated parking space. No smoking. Rent $1,450 a month. Available
from 1 December.

A good result might headline the listing "Two-bedroom first-floor flat with parking, available 1 December" (illustrative), run to about 90 words of description and give five bullets covering heating, space, parking, unfurnished and no smoking. The check is the point. Compare every claim with the sheet. Watch for inventions such as "close to local shops", "quiet residential street" and "ideal for young professionals". The first two are unverifiable claims you could be held to; the third describes the tenant rather than the home, which invites a discrimination complaint. Delete all three.

The same fact sheet supports enquiry replies. Ask for three templates: available, no longer available, and viewing slots. Keep the questions to what you genuinely need before a viewing, such as the preferred move-in date and when the person can view. Use [first name] and [your name] as placeholders and add the real names yourself. Never paste an enquirer's full message with their phone number or address into a free tool. Paste the question they asked, or retype it.

Here is an available-reply template once the tool has drafted it and you have corrected it. It is illustrative, with placeholders where your details go.

Hi [first name], thanks for your enquiry about the two-bedroom
flat. It is available from 1 December at $1,450 a month. Viewings
are on [day] at [time] or [day] at [time]. Could you tell me your
preferred move-in date and which slot suits you?
Kind regards, [your name]

Note what is missing: no promise of approval, no hint of how you will choose between applicants, and no screening questions at all. Those belong to a later, documented application step that you control.

Two guardrails belong here. Selecting applicants is a decision about people, and decisions about who gets a home attract rules on equal treatment in most places. Keep AI out of the shortlist, as explored in how property managers use AI to screen tenant applications fairly. And keep reply wording consistent: ready-made templates in ChatGPT message templates for residents and owners make that easier than writing each reply fresh.

Workflow three: lease abstracts and inspection comparisons

A lease abstract turns a 30-page agreement into one spreadsheet row. It is the job where free tools save the most time per document and where a wrong answer costs the most, so it comes with a verification step that isn't optional.

Redact first: remove the names of the parties, contact details, signatures and any bank details, or paste only the clauses you need. Then ask for a table, one row per item, with the clause number and a short quote for each answer.

From the lease text below, return a table with columns: Item, Value,
Clause number, Quote (maximum 15 words). Items: start date, end date,
rent amount and payment day, rent review date, notice period for the
tenant, notice period for the landlord, break clause, deposit amount,
who pays which utilities, rules on pets and smoking, notice required
before access for repairs. If an item is not in the text, write
NOT FOUND. Do not infer.

The quote column is what makes this checkable. Open the lease, search for the quote and confirm the value beside it. Do that for every date and every amount, not a sample, because a notice period that is wrong by a month can sink a legitimate step. Where the tool writes NOT FOUND for something you remember being there, search the document yourself: the answer may sit in an addendum you didn't paste.

A realistic miss shows why the quote column exists. In an illustrative abstract the tool reports a 30-day tenant notice period and quotes clause 6 accurately. Your check against the full file finds a later addendum that changed the period to 60 days. The tool read clause 6 correctly and simply never saw the addendum. The fix is procedural: paste the lease and every addendum into the same chat, and keep the instruction to write NOT FOUND rather than guess.

Here is the sum. Abstracting 12 leases by hand at about 20 minutes each takes 4 hours. With the tool and a full check at about 8 minutes each it takes 96 minutes. The saving is 144 minutes, or 2.4 hours, as a one-off, and the spreadsheet then drives your calendar of reviews and renewals. The abstract is a finding aid and never replaces the lease for a legal step.

Inspection comparisons are the second use. Type your check-in and check-out notes, room by room, with unit codes and no names, and ask for differences by room, each labelled damage, wear, cleaning or unclear. The tool can spot that "scuffed skirting, bedroom 2" appears only at check-out. It can't decide what counts as fair wear, and it can't see photos you haven't shared. Any deduction is your decision, backed by the photographs and the inventory. The fuller routine, with sample reports, is in AI inventory and check-out reports: a letting agent's guide.

If you want to ask questions across all your leases at once, Gemini Notebook (formerly NotebookLM) is the free-tier tool built for it. Google's help says a Standard account holds 100 notebooks and 50 sources per notebook, so one notebook can hold a dozen redacted leases. Read the data note on that plans page before uploading anything. Google says your data isn't used to train Gemini Notebook unless you send feedback, and that feedback lets it review the whole interaction, queries and uploads included. Only Workspace accounts get the promise of no human review even when feedback is given, so a personal account should never send a thumbs-up or thumbs-down on a lease notebook.

What tenant data does on a free plan: a sorting list

Tenants have a stake in how their information is handled even when no regulator is watching, and a landlord who loses their trust loses far more than an hour a month. Sort every item before it goes into a prompt.

Photos need their own habit. An image shows more than the leak: a letter on a table, a child's drawing, a reflection in a window. Crop to the fault, and never send a photo of an identity document. Because phone images can carry details beyond what you see, send a screenshot of the cropped area rather than the original file.

The unit-code key is what holds the system together. Keep it in a separate sheet, offline or in a locked drive, and rejoin names to outputs only inside your own documents. There is a simple honesty test too. If a tenant asks what you do with their data, you should be able to say: "I use an AI assistant to sort repair requests. Messages go in under a unit code, without names, addresses or photos of your home beyond the fault." You can only say that truthfully if you follow the table.

Four jobs to keep for yourself, whatever the plan

Some property work shouldn't pass through a chat window on any tier, because the cost of a plausible-sounding mistake is higher than the minutes saved.

• Emergencies. A person phones the tenant. The log comes second.
• Anything with legal effect. Notices, rent increases and deposit deductions follow rules on format, timing and evidence that differ by place and change over time. Use official forms or an adviser, and let the tool draft the courteous covering note. Rent increase letters with AI shows where the rules bite.
• Sums. Arrears, pro-rata rent, deposits and service charges go through a spreadsheet. A language model can produce a plausible total without adding the figures.
• Choosing people. Deciding who gets a tenancy, and settling disputes between tenants, stay with you, because both are decisions about people that you must be able to explain.

Each of these keeps a smaller free-tool role: drafting the courtesy, finding the clause, formatting the table. The decision and the arithmetic stay on your side of the screen.

The point at which to pay: a sum for a 40-unit portfolio

Volume isn't what pushes a landlord off the free tier. Take a 40-unit portfolio with about 60 repair requests and 25 enquiries a month. At the saving rate from the repairs workflow above, 4 minutes per request, plus an assumed 3 minutes per enquiry, that is 240 plus 75 minutes, or 315 minutes. Add 8 new leases a year at 12 minutes saved each, about 8 minutes a month, and the total is around 323 minutes: roughly 5.4 hours. Valued at $30 an hour, that is about $160 a month of manager time against a seat that costs around $20.

The maths favours AI at almost any size, so what changes with scale is which plan fits. Four triggers say pay, and each can be checked.

1. Tenant data keeps ending up in prompts. If anonymising slows you more than the tool saves, move to a plan with no training on your content by default. ChatGPT Business Standard is $20 per user a month billed annually ($25 monthly) for 2 to 200 users. Claude Team Standard is $20 annual or $25 monthly for 2 to 150 seats. Two seats is $40 a month at the annual price. If you already have a Microsoft 365 business plan, check whether Copilot Chat is switched on: Microsoft says it is included at no extra cost with eligible plans, with enterprise data protection for work accounts.
2. You need more than three documents a day. A lease-renewal month on ChatGPT Free stalls at the third upload.
3. A second person shares the work. Shared Projects and a single bill belong on a business plan. Two people on one login is a different problem and breaks the terms of some services.
4. You want it to act, not just draft. Chat plans don't read your inbox, update the log or send the reply. That job belongs to property-management software with AI built in, or an automation, and it is a separate buying decision. A resident FAQ bot built from your own leases is one example of that next step.

My rough thresholds are judgement rather than rules. Under about 15 units, free tools with the anonymising habit are enough. Between roughly 15 and 50 units, two business seats or a work-account Copilot Chat is worth it. Beyond that an audit trail starts to matter, and software that records who did what earns its price.

A 30-minute setup and a monthly check

1. Build the unit-code sheet and keep it offline (10 minutes).
2. Create the repairs log with the columns above (5 minutes).
3. Save the three prompts in one document, with a note on what may and may not be pasted (5 minutes).
4. Switch off model improvement on the account and use Temporary or Incognito chats for anything near the border (5 minutes).
5. Run five old messages through the triage prompt and compare the urgency with what you decided at the time (5 minutes).

Once a month, spend ten minutes checking five AI outputs against their originals: two urgencies, two lease values and one listing line. If more than one in ten is wrong, tighten the prompt or take that job back by hand. A failed check costs ten minutes, and an unnoticed wrong notice period can cost a tenancy dispute. Arrears wording deserves the strictest check of all; can AI draft rent arrears letters that stay within the rules sets out where the line sits. If you'd rather map your own repairs, enquiries and paperwork with someone, that is what my AI implementation consultation covers.

Free AI for landlords: common questions

Can free AI decide which applicant gets a tenancy?

It shouldn't. Rules on equal treatment apply to anything that influences who is offered a home, including a chatbot-assisted shortlist, and an assistant can't explain its reasons in a way you could defend. Applications also hold sensitive personal data that doesn't belong in a consumer chat. Write consistent criteria, apply them yourself, and keep applicant files out of AI tools.

Is it safe to upload a lease to a free chatbot?

Not as it stands. A lease carries names, addresses, rent, deposit terms and signatures. Redact party names, contact details and bank information first, or paste only the clauses you are asking about. The limits table earlier shows what each free plan does with uploaded content, and none of them is a place for a full signed document.

Can free AI write rent arrears or eviction notices?

It can draft a courteous first reminder. Formal notices must follow exact rules on wording, timing and delivery that vary by place and change over time, and a wrong date or clause can make a notice invalid. Use the official form or an adviser for anything with legal effect, and keep AI to the polite first step.

Do I need to tell tenants I use AI tools?

If tenants' personal data goes into a tool, your privacy notice should describe who processes it and why, which a free consumer plan makes awkward to justify. If you paste only anonymised text, the question mostly disappears. Ask a qualified adviser what your notice should say for the places where you operate.

Further reads

• AI Landlord Updates and Owner Statements in Minutes — Monthly owner statements and landlord updates in minutes.
• How Much of a Property Manager's Week Can AI Take Over? — How much of a manager's week AI can realistically take.
• AI or Human Answering Service: Which Suits a Property Manager? — AI or a human answering service for a property manager.
• Free AI Tools vs Paid Plans: What a Small Business Should Pay For — The general free-versus-paid decision for any small firm.
• Best Free AI Tools for Small Business Owners and Their Catches — Free tools for owners, and the catch with each.
• How Can AI Help in Property Management? Tenant Messages and Repairs — Tenant messages and repairs with AI, from the manager's side.
• Shadow AI: How to Stop Staff Pasting Client Data Into Free Tools — Stopping staff pasting tenant or client data into free tools.
• AI Assistant for Property Management: A Resident FAQ Bot From Leases — A resident FAQ bot built from your own leases.
• RealPage AI Leasing Agent vs a Standalone Chatbot for Small Portfolios — Knock's AI leasing agent versus a standalone chatbot for a small apartment portfolio: PMS fit, voice, price transparency and exit terms.
• Property Investment AI Tools: A Stack for a Five-Property Portfolio — A five-property investor can start with free deal and bookkeeping tools, then pay only for the workflow that repeatedly saves review time.
• AI Property Measurement Software: Is It Accurate Enough to Use? — A ten-flat test you run yourself to check photo, LiDAR and floor-plan tools against a laser measure, and when to hire a surveyor.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI help articles on data controls, file uploads and ads, Anthropic pricing, privacy and upload help pages, Google Gemini Apps help and Privacy Hub, Google Gemini Notebook plan limits, Microsoft 365 Copilot pricing page, read on 1 November 2026 and checked again on 4 November 2026.