AI for Realtors in the Philippines (2026): 12 Prompts, 6 Workflows and the Guardrails That Keep You Licensed, Honest and Faster

AI can draft your listings, captions and follow-ups faster. It cannot know your facts, your clients or the law. Here are six checked workflows, twelve prompts and the guardrails for Philippine realtors.

The guide at a glance

+14%average issues resolved per hour in a Stanford/MIT study of support agents (+34% for novices) ^[A6]

25%faster in the BCG/Harvard study, inside AI's frontier (worse outside it) ^[A9]

46%of US agents reported no noticeable AI impact in the 2025 NAR survey (self-reported) ^[A1]

2007the year of the "reply in 5 minutes" study behind the famous stat ^[A4]

0independent studies we found backing "10x faster" for realtor work ^[A6]

6workflows from the infographic, fact-checked

12prompts, 4 in full and all 12 in the builder

18data types sorted: paste, anonymise, avoid or never

1. What AI really changes (and what it does not)
2. The six workflows from the infographic, fact-checked
3. Your starter stack
4. Guardrails: privacy, ads, AML, scams, hallucinations
5. Twelve prompts that work
6. Listings that sell: fact sheet, draft, audit
7. Photos, staging, tours and drone: what is OK
8. Follow-up that does not feel like spam
9. Market research and pricing talk
10. Contracts, notes and translations
11. Eight agent types, eight playbooks
12. Where platforms help your workflow
13. A 30-day pilot and the metrics
14. Twelve mistakes
15. Our recommendations
16. FAQ
17. Sources
18. What we could not verify

1. What AI really changes (and what it does not)

Picture a fast, eager junior assistant. It has read a huge amount, writes a clean first draft in seconds and never gets tired. It has also never seen your listing, never met your client and sometimes states a guess as a fact.

That is the right mental model. AI is very good at drafting, tidying and summarising what you give it. It is unreliable at knowing things you did not give it: today’s price, a tax rate, the status of a title.

A trending infographic promises that AI will find your leads, write everything “10x faster” and close your deals. We checked what the research really supports. The short version: AI helps, mostly on drafting, mostly by a modest amount, and it helps new workers most.

The evidence, with its quality label on every row

We split the evidence into four honest grades. Independent studies are the strongest. Surveys of agents are self-reported: people describe their own results. Vendor claims are marketing. One famous statistic is simply old.

What to take from the table

• The independent studies agree on a range, not a miracle. Roughly 10 to 30 percent on drafting tasks, with newer workers gaining the most ^[A6][A9]. That is a reasonable reading of the research, not a Philippine measurement.
• Time saved is not the same as better work. A UK government trial found time savings without a measurable productivity gain ^[A8].
• Outside AI’s strengths it gets worse. In the Harvard and BCG study, users were less likely to be correct on tasks outside the frontier ^[A9]. For realtors, that frontier ends where facts you cannot supply begin.
• Nearly half of US agents saw no noticeable impact in the 2025 NAR survey, and about one in three did not use AI at all ^[A1]. Self-reported, US only.

The two famous numbers, put in their place

“Reply within 5 minutes.” It comes from a 2007 observational study of web leads, later echoed by a 2011 Harvard Business Review study of 2,241 firms. It measured how often a lead was contacted and qualified, not how often it closed ^[A4][A5]. Fast replies are good manners and good business. But never say “21 times more likely to close”.

“10x faster.” We found no independent evidence for it in realtor work ^[A6][A9]. If someone sells you a 10x promise, ask for the study. Then measure your own tasks (section 13).

2. The six workflows from the infographic, fact-checked and upgraded

The infographic is titled “AI for Realtors: More leads. More listings. More success.” It has six blocks: lead generation, smart follow-ups, content creation, market analysis, property marketing visuals and deal closing. A mock chat in the corner announces that an “AI Realty Assistant found 15 new leads”.

We kept the six blocks, because they are the right six jobs. We checked each claim and rewrote each block as a workflow you can actually run.

Workflow 1

Lead sorting and priorities

The infographic says

“Lead generation: AI finds and scores leads”

Reality check

AI can sort leads only from data you already hold. It does not find new buyers by itself, and no site-level lead scoring is evidenced. Better data in, better list out.

Our upgraded workflow

1. Keep every inquiry in one sheet: lead ID, source, date, budget band, timeline, next step.
2. Paste redacted rows (IDs, not names) into your assistant.
3. Ask for leads with no next step or stalled more than 7 days (prompt 12).
4. Ask for a top-5 list with a one-line reason each, then decide the order yourself.
5. Log what you changed and why, so you can check the list later.

Human check: You confirm every stage, date and priority; the assistant can silently change them.

Tool categories: General AI assistants, Starter pipeline: sheet, notes, calendar, CRM with AI features

Workflow 2

Follow-ups you actually send

The infographic says

“Smart follow-ups”

Reality check

Speed helps, but the 5-minute stat is a 2007 observational study of web leads (contact, not closings). Automation without consent becomes spam. Draft with AI, send as yourself.

Our upgraded workflow

1. Write your rules once: who gets what, when, and the stop option.
2. Have AI draft a reply bank for the common questions (price, viewing, documents).
3. Personalise name and property by hand before each send.
4. Use a 7-touch cadence per lead type and stop when asked.
5. Log opt-outs the same day.

Human check: A human reads every message; never auto-send to people who have not opted in.

Tool categories: General AI assistants, Messaging and chatbots, Starter pipeline: sheet, notes, calendar

Workflow 3

Listing and social content

The infographic says

“Content creation 10x faster”

Reality check

No independent evidence for 10x in realtor work. Studies point to roughly 10-30% on drafting tasks, more for novices ^[A6][A7][A9]. The facts still need your check.

Our upgraded workflow

1. Write a fact sheet first: area, price, status, dues, turnover, documents.
2. Ask the assistant to draft from the fact sheet only (prompts 1, 2, 11).
3. Run the draft through the claims checker (section 6).
4. Edit tone and Taglish by hand.
5. Publish with the broker name your firm requires.

Human check: Every number, distance and amenity is matched to a source before posting.

Tool categories: General AI assistants, Design and layout

Workflow 4

Market and comps explanations

The infographic says

“Market analysis”

Reality check

AI can summarise data you supply. From memory it may invent prices, tax rates or zoning. Use it to explain your comps, not to create them or value a property; an appraisal needs a licensed appraiser.

Our upgraded workflow

1. Build your own comps table with date, sqm, price, condition and source.
2. Ask the assistant to explain your range in plain words, with price per sqm (prompt 8).
3. Recompute the maths yourself.
4. Check flood and hazard info against official maps and the LGU.
5. State clearly that it is a broker opinion of value, not an appraisal.

Human check: Recalculate every figure; separate asking prices from closed prices.

Tool categories: General AI assistants, Starter pipeline: sheet, notes, calendar

Workflow 5

Photos, staging and video

The infographic says

“Property marketing: visuals, virtual tours and flyers”

Reality check

Fine for lighting, crop, mood art and scripts. Adding or removing fixtures or views, or AI renders beyond approved materials, can mislead. No PH rule on AI-image labels found; we recommend a label.

Our upgraded workflow

1. Shoot real photos with a shot list; keep the originals.
2. Limit edits to exposure, straightening and crop; label anything else.
3. Use AI for flyer layouts and reel scripts, not for the unit itself.
4. For pre-selling projects, show only the developer's approved materials.
5. Add a label (see section 6) and put original next to staged where you stage.

Human check: Ask: would a buyer feel misled on the viewing day? If yes, do not post it.

OUR CAUTION: AI renders beyond approved materials for a registered project are UNVERIFIED as a rule; it is our caution based on PD 957 section 19 principles.

Tool categories: Design and layout, Image generators, Virtual staging and photo clean-up, Floor plan to 3D, tours, drone

Workflow 6

Practice, notes and deal support

The infographic says

“Deal closing: nurture leads and handle objections”

Reality check

AI can rehearse objections and tidy your notes. It does not close deals or know the law. Keep price, tax and legal answers with primary sources, and a human on every offer.

Our upgraded workflow

1. Role-play the five hardest objections (prompt 5) and note better answers.
2. Dictate post-viewing notes and have them structured; proofread Taglish.
3. Draft the follow-up and the next-step list.
4. Get tax, transfer and legal figures from BIR, the Registry of Deeds, the developer or a lawyer.
5. Give the client the source, not the AI.

Human check: A licensed human handles negotiation, signing and every price or tax statement.

Tool categories: General AI assistants, Transcription and meeting notes

Matching demand with supply

Workflow 1 only works if demand data is organised. One model worth studying is a public board of anonymous requirements, where owners and agents answer privately.

3. Your starter stack: ten categories, honest prices

Do not shop for tools by hype. Shop by job. The cards below group the tools into 10 categories, with what each is for, example names, a price band in US dollars, Philippine notes and the privacy catch.

Two details in the cards deserve a second look. First, almost every CRM with AI features is built for the US market, where leads come from MLS and IDX feeds ^[A21]. A spreadsheet is a perfectly serious start. Second, your phone is your camera, recorder and editing desk. If you are choosing one, our comparison of the best phone for real estate brokers in 2026 looks at it from a broker’s side.

The big US brokerages are building AI into their own platforms. One launched an in-house assistant in July 2026 and reported about 15,000 agents and more than 97,000 conversations within weeks ^[A24]. That is usage inside one company, not proof of results, and it is not available to Philippine agents.

4. Guardrails: privacy, honest ads, AML, scams and made-up facts

Guardrails are what let you use AI every day without losing sleep. There are seven, and each one is short enough to remember.

Guardrail 1: Protect personal data

The Data Privacy Act applies to AI use. The NPC’s Advisory 2024-04 says the law covers AI training, testing and deployment. The advisory’s key ideas are accountability, transparency, legitimate purpose, proportionality, accuracy, fairness and data minimisation ^[A10]. In plain words: you stay responsible for the personal data you share, you should share only what the task needs, and you should not pretend a person wrote what a machine did (the advisory calls that “AI washing”).

Two practical points follow from our reading. Public AI tools process data outside the Philippines, so pasting client data is a cross-border share. And a breach of sensitive data may trigger a notice duty to the NPC; the 72-hour rule is the standard DPA one, but we did not verify how it applies to a small agent (UNVERIFIED).

The paste test: 18 data types, four verdicts

Before you paste anything, find it in the list. The verdict is OK, anonymise first, avoid or never. 6 of the 18 are a flat never, 4 are avoid, 4 need anonymising and 4 are fine. Several rows are our own judgement and say so in the last column.

Try it: pick a data type and get a verdict card with a safer alternative. It runs in your browser and nothing you choose is sent anywhere.

Based on the Data Privacy Act and NPC Advisory 2024-04 as summarised in research. Several rows are our own judgement, flagged in the basis field. Whether registration or breach notice applies to a small agent is UNVERIFIED. Not legal advice; ask your broker or the NPC.

How to anonymise in eight steps

1. Replace names with roles or IDs (Buyer A, Lead L-014, Seller 3).
2. Remove phone numbers, emails, social handles and exact addresses.
3. Remove ID numbers, TIN, account numbers, title and tax declaration numbers.
4. Use ranges: a budget band, a general area, a size range.
5. Crop faces, plates, mail and documents out of photos.
6. Keep a private key (your own sheet) linking IDs to people. Never paste it.
7. Ask permission before recording or sharing anything a client told you in confidence.
8. After the task, delete the chat if the tool allows it.

Guardrail 2: Your broker answers for your ads

Under the Real Estate Service Act (RA 9646), practising means a PRC licence: broker, salesperson, appraiser or consultant. Salespersons work under a broker’s direct supervision, and the broker is accountable for a salesperson’s misrepresentation ^[A12]. For a team, that has one clear meaning: “the AI wrote it” is not a defence, and the broker needs to see the ads.

Whether PRC rules require a licence number or broker name in advertisements is not confirmed in our research (UNVERIFIED). Including both is good practice. You can check a practitioner’s licence on the PRC’s online verification.

Guardrail 3: Do not let AI embellish a registered project

For projects, PD 957 requires a License to Sell before an owner or dealer offers or advertises units, and section 19 says ads must reflect real facts and not mislead ^[A13]. Commentary suggests brochures and ads can become enforceable promises, and that “artist’s perspective” disclaimers do not automatically override that ^[A13].

Guardrail 4: Know if AML applies to you

The amended anti-money-laundering law (RA 11521) made real estate developers and brokers covered persons, with duties such as reporting covered transactions of PHP 7.5 million or more and registering with the AMLC ^[A11]. Whether and how this applies to an individual salesperson, and what the client-checking steps are, is UNVERIFIED in our research. Check whether it applies to you.

The AI angle is simple. Client-checking files are sensitive, so they stay out of public AI tools. AI may help you draft a blank checklist. It should not see client files or judge what looks suspicious.

Guardrail 5: Truth in advertising, and who owns AI images

The Consumer Act (RA 7394, article 110) bans false or misleading advertising for consumer products and services ^[A16]. We have not confirmed whether it applies to property listings or whether PD 957 and RESA govern instead (UNVERIFIED). Treat it as the general principle.

On copyright, the Philippine IP Code protects works by people, and AI-made works are not explicitly addressed; the US Copyright Office view is that purely AI-made output is not protected ^[A17]. Your AI images may therefore be free for others to copy. Check each tool’s commercial-use terms, do not imitate living artists or other firms’ branding, and never feed competitors’ photos into a tool.

Guardrail 6: Scams and deepfakes

Real estate is a favourite target. The CICC and SEC issued deepfake warnings in 2025, and typical property cases include fake listings, fake agents, fake payment instructions and impersonated owners ^[A18]. Voices and faces can now be cloned, so a familiar face on a call is not proof.

Guardrail 7: Hallucinations. Verify what AI tells you

AI invents plausible links, places and numbers. It will state a tax rate, a zoning rule or a title status with total confidence and be wrong. The fix is a habit: AI drafts, you cite primary sources such as the BIR, the LGU zoning office, the Registry of Deeds, DHSUD or the developer.

A good test is a question with a known answer. Ask your assistant what a tax declaration proves about ownership, then compare it with a source-backed explainer on tax declaration vs title. The gap you find is the reason this guardrail exists.

You also owe readers honesty about the tool. If a chatbot replies for you, it must say it is automated and hand off to a human ^[A10]. Where AI shaped the copy, say so when it matters; we recommend it, even though no Philippine rule we found requires it.

5. Twelve prompts that work (four in full)

A prompt is just a brief to your assistant. The twelve below follow one pattern, and each carries the same two safety lines: “Use only the facts I give; mark anything uncertain as [VERIFY]” and “If a fact is not provided, write NOT PROVIDED instead of guessing”.

Five tricks that make any prompt better

These are our own working habits, not research results.

1. Give facts, not adjectives. Area, price, status and dues beat “nice and spacious”.
2. Say who it is for. A first-time buyer, an OFW and an investor need different words.
3. Ask for what is missing. “List the facts you needed but I did not give” turns gaps into a to-do list.
4. Forbid the usual inventions. No amenities, distances, prices or legal claims you did not supply.
5. Run a second pass. Paste the draft back and ask the assistant to find every number and claim that needs a source. Then check them yourself.

Here are four prompts in full, each with its fill-in fields, a check list and a pitfall. They are the ones you will use every week: the listing draft, the follow-up cadence, the comps explanation and the Monday pipeline review.

You are a Philippine real estate copywriter. Write a [LENGTH]-word [KIND] listing in [LANG] (English, Filipino or Taglish) using ONLY these facts: [FACTS]. Tone: [TONE]. End with a call to action to message [CONTACT_NAME]. Use only the facts I give; mark anything uncertain as [VERIFY]. If a fact is not provided, write NOT PROVIDED instead of guessing. Do not invent amenities, distances, prices or legal claims, and avoid words like "prime" or "guaranteed". After the text, list the facts you needed but I did not give.

Draft a 5-message Viber/Messenger follow-up for a lead who [SITUATION] about [PROPERTY] ([DATES]). Write in [LANG]: short, polite, warm, each message with one clear question, spaced Day 0, 1, 3, 7 and 14. Include a stop option in the first and last message. Use only the facts I give; mark anything uncertain as [VERIFY]. If a fact is not provided, write NOT PROVIDED instead of guessing.

Here are my comparable listings and sales: [COMPS]. Explain in plain language for a seller how I arrived at the range [RANGE]. Show price per sqm, adjustments and uncertainty. Do not add comps, and do not value the property yourself. Use only the facts I give; mark anything uncertain as [VERIFY]. If a fact is not provided, write NOT PROVIDED instead of guessing.

Here is my redacted pipeline using lead IDs only: [PIPELINE]. Find leads with no next step or stalled more than 7 days, then list my top 5 priorities this week with a message draft for each. Output as a table. Use only the facts I give; mark anything uncertain as [VERIFY]. If a fact is not provided, write NOT PROVIDED instead of guessing.

The other eight

The remaining prompts cover the Taglish caption, the buyer brief interview, objection practice, the viewing checklist, the neighbourhood guide, the seller pricing memo, the lease summary and the reel script. The builder below has all twelve.

Try it: choose one of the twelve prompts, fill the fields and copy the finished prompt with its check list. It runs in your browser and nothing you type is sent anywhere.

A one-page brief to ask for

One useful output to ask an assistant for is a one-page brief on a project or listing, built only from the developer’s own materials. A good brief has a fixed shape.

6. Listings that sell: fact sheet, draft, audit

Most bad AI listings fail in the same way. The agent types a few adjectives, the assistant fills the gaps with plausible amenities, and a wrong number goes live. The cure is a three-step habit.

1. Fact sheet. Write down only what you can prove, with a source for each line.
2. Draft. Give the fact sheet to the assistant with prompt 1 and ask for the facts it still needs.
3. Audit. Run the draft through the claims checker, fix every flag, then match each number to its source.

Step 1: the fact sheet

Step 2: the draft

Paste the fact sheet into prompt 1 (section 5). Choose the language: English, Filipino or Taglish. Then read the answer as an editor. Look for anything you did not give the assistant: a pool, a view, a school nearby. Delete it, or prove it.

If you want to see what a clean one-property post looks like, our how-to guides on how to post your property in minutes and how to make a premium property post show the format step by step.

Step 3: the audit

The checker looks for 19 kinds of problem and 4 missing basics. Each flag means “check this”, not “this is illegal”. It is a first pass and misses most real problems, so your own read comes after it.

All checks run in the browser. A flag means "check this", not "this is illegal". The regexes are a first pass and miss most real problems. Price per sqm is computed only when both a price and an area are found, and it is a rough figure: it mixes rent with sale prices and floor area with lot area.

Here is the checker on two sample texts. The first is full of puffery and guarantees. The second makes legal claims about the title, the project, taxes and floods.

On the clean sample (“2-bedroom condo unit in Quezon City, 54 sqm per developer plan, asking PHP 6,500,000”) the checker finds no warnings. It computes a rough PHP 120,370 per sqm. That figure is only a guide: it mixes rent with sale prices and floor area with lot area, and the checker only computes it when it finds both a price and an area.

Try it: paste a listing and see the flags, a list of numbers to verify, the missing basics and a label generator for altered images. Everything runs in your browser; nothing you paste is sent anywhere.

7. Photos, staging, tours and drone: what is OK and what misleads

Pictures sell, and pictures can mislead. The test that works everywhere is simple: would a buyer feel misled on the viewing day? If yes, do not post it.

Start with real photos

A shot list beats any generator. Cover the exterior and street, the gate and security, the lobby and real amenities, each room from two corners at eye level, bathrooms, the kitchen, the balcony view in daylight, the meter area, parking and the floor plan. Add the neighbourhood and any flood-prone points honestly. Shoot vertical clips for reels.

The line between editing and misleading

No Philippine rule on labelling AI-edited listing photos turned up in our research (UNVERIFIED). The US examples are a useful benchmark, not Philippine law: California AB 723 (from 1 January 2026) requires disclosure of altered listing images and access to the original; Wisconsin Act 69 follows from 2027; and one Wisconsin MLS requires originals, a remark and carries a daily sanction ^[A14][A15]. PD 957 section 19 and the Consumer Act principles still apply ^[A13][A16].

Our rule: label every altered or AI image, keep the originals, and show the original next to the staged version. Suggested labels:

• Virtually staged: the furniture in this image was added digitally. The original photo is shown beside it. Use when: Furniture or decor was added to a real photo.
• AI-generated illustration, not a photo of the actual property. Use when: Mood art, covers or concept images.
• Digitally enhanced for lighting and colour only. Use when: Exposure, white balance, straighten and crop edits. Optional.
• Developer's artist's perspective. Final units may differ; see the approved plans and ask for the developer's License to Sell and Certificate of Registration numbers. Use when: Pre-selling renders. Our caution: a disclaimer may not override PD 957 section 19 (commentary source; confirm).
• Text drafted with AI assistance and checked by [your name]. Use when: Optional. No Philippine rule found requiring it; we recommend it where AI shaped the copy.

Virtual staging: cheap, useful, and easy to overdo

Price comparisons put AI staging at roughly US$0.23 to US$15 per image, against US$16 to US$50 for human-edited staging and US$1,500 to US$6,000 for physical staging of a home ^[A19]. These are vendor and comparison-blog figures (VENDOR), so verify them. The saving is real. The risk is also real: staged furniture that does not fit the room, or a window view that never existed.

Floor plans, 360 tours and drone

A 360 tour shows the real space, which is why buyers trust it. A floor plan turned into a 3D walkthrough is a visual aid, not a photo; label it that way. We cover the method in Floor Plan to 3D: AI Video for Philippine Property Listings.

Drone shots of the real site can help, but they record neighbours and passers-by. Blur or trim people and plates. The CAAP rules for drones were not verified in our research (UNVERIFIED), so check them and the consent question before you fly and before you post.

Reels: script with AI, film for real

Prompt 11 writes a 30-second script with a hook, four shots, on-screen Taglish text and a caption. Film the real rooms. Make sure the music licence is cleared, the owner or tenant agreed to filming, and no faces or plate numbers show. Never use AI-generated footage of the property itself.

8. Follow-up that does not feel like spam

Most deals are lost in the quiet gap after the first reply. A good follow-up is short, warm, useful and easy to refuse. A bad one is a barrage.

Speed matters, but keep the famous “5 minutes” in proportion: it comes from a 2007 study of web leads that measured contact and qualification, not closings ^[A4][A5]. Our suggestion is a clear first-reply target in business hours, such as 5 to 15 minutes. That is our own suggestion, not a research result.

Rules before cadence

• Message only people who contacted you or opted in, and keep a simple record of that. Do not scrape numbers from groups. research + our judgement
• Always offer a way to stop, and honour it the same day. The DPA recognises a right to object to direct marketing; the specific NPC direct-marketing rules were not confirmed. UNVERIFIED
• The SIM Registration Act aims to trace text scams; it is not a marketing-consent rule. research
• NTC rules on commercial SMS and sender IDs were not found. UNVERIFIED
• Bulk SMS with links or "@" and ".com" is reportedly filtered by Philippine carriers. This comes from our own server experience, not an outside source, so keep SMS short and link-free. UNVERIFIED
• A human reads and sends every message. Do not set AI to auto-send to leads. research + our judgement
• If a chatbot replies for you, it must say it is automated and hand off to a human. research

Seven touches, six lead types

For each lead type there are 7 touches, on suggested days. Day offsets and send times are our suggestions. Notice how the cold-lead and past-client rhythms stretch out, and how both end with an offer to stop.

Day offsets and send times are our suggestions, not research results.

An example week

Placeholders in curly braces are filled by hand: name, agent, property and the viewing slot. Touches 5 to 7 follow on days 14, 21, 30.

Draft the whole bank with prompt 3, then personalise the name and property by hand before each send. Log every opt-out the same day. If a lead replies “no” or “stop”, the cadence ends there, whatever the sheet says.

Try it: pick a lead type, a channel and a start date to get the seven-touch schedule with editable skeletons, a calendar file and a print view. It runs in your browser and nothing you type is sent anywhere.

9. Market research and pricing talk

This is the section where AI is most tempting and most dangerous. Ask it “what is the price per sqm in this area?” and you will get a confident number. It may be invented.

The rule that keeps you safe is a division of labour. You build the data. AI explains it. The Harvard and BCG study is the warning: outside the tasks AI handles well, users were less likely to be correct ^[A9].

Where each kind of fact actually lives

A neighbourhood guide you can trust, in five steps

1. Run prompt 7 and require a source for every claim, or the word UNSOURCED.
2. Open every link. Delete any that do not exist, because AI invents plausible links.
3. Measure every distance on a map, walking or driving, and note the hour.
4. Take flood and hazard information only from official maps and the LGU.
5. Walk the area at the hours your client would. Listen for the noise.

Pricing talk with a seller

Build the comps table first. Then use prompt 8 to turn it into plain language for the seller: price per sqm, adjustments and uncertainty. Prompt 9 drafts a one-page memo, with the days-on-market scenarios clearly labelled as your estimates.

Three lines you should not cross. Do not let the assistant add comps. Do not call the result a valuation: AI cannot value a property, and an appraisal needs a licensed appraiser. Do not promise a sale time.

For investor conversations, use AI to draft the report outline and recompute every yield yourself. A useful cross-check is our Metro Manila investor dashboard 2026, which shows how yield, occupancy and rent can be laid side by side.

Search is where AI helps your buyers

Share one more idea with your clients: let them describe what they want in plain words and have the portal turn it into filters.

If your buyers are new to portals, send them to how to search for a property.

10. Contracts, notes and translations

Lease and contract summaries

A summary table helps you read a lease faster: parties, term, rent, deposit and advance, escalation, repairs, renewal and penalties. Prompt 10 produces one. Two rules apply, and neither is optional.

1. Redact first. Remove names, ID numbers and signatures before anything is pasted.
2. The summary is not the contract. Compare each line to the original, because skipped clauses and invented terms are common. A lawyer reviews risky clauses.

Voice notes and Taglish

Dictating notes right after a viewing is one of the best habits you can build. But Taglish is where transcription tools stumble. A 2023 test found OpenAI Whisper’s Tagalog output often needed revision, while a 2026 vendor comparison claims far lower error rates ^[A22]. Neither Otter nor Fireflies is confirmed to handle Taglish well.

So test it on your own voice, proofread every transcript, and ask permission before you record anybody. A recording holds names, voices and sometimes money details. When in doubt, write three lines of your own notes and use those.

Translations for OFW and foreign buyers

Assistants are good at polite, clear drafts in Filipino, Chinese, Korean or Japanese. They are not proven on real estate legal terms (UNVERIFIED). Have a native speaker check any message that includes prices, dates or terms, and never paste a whole signed contract. Redact it first.

Keep clients informed with sources

The best use of AI answers is a starting point that you pair with a source. When a client asks about loans, titles or taxes, give the AI-assisted summary and a real article to confirm it.

For the loan question, our guide to the bank vs Pag-IBIG home loan and DTI guide is a good example: it shows the debt-to-income arithmetic so that the client can check your answer for themselves.

11. Eight agent types, eight playbooks

An agent’s week is different depending on what they sell. Below are 8 playbooks with the first five automations, the guardrails that matter most and the pitfalls we expect.

Playbook 1

Solo agent

You handle listings, chats and viewings alone, mostly on your phone.

First five automations

• Listing copy from a fact sheet
• Taglish captions
• Viber reply bank
• Voice-note summaries (proofread)
• Weekly pipeline review

Guardrails

• No client IDs or numbers in chats
• Facts only from you
• Human sends every message

Estimated time saved: 3 to 6 hours a week ESTIMATE (ours, not measured)

Our estimator, typical level: about 3.0 hours a week for this type of work.

Pitfalls: Privacy slips, tone that is not yours, slower replies if you over-automate.

Playbook 2

Small team or brokerage

A team lead wants everyone to use AI the same, safe way.

First five automations

• Shared prompt library
• Lead triage rules
• Weekly pipeline review
• A one-page AI policy
• Claims checker before posting

Guardrails

• Broker reviews ads (the broker is accountable for salesperson misrepresentation)
• No shared personal accounts
• Log which tool saw which data

Estimated time saved: 2 to 5 hours a week ESTIMATE (ours, not measured)

Our estimator, typical level: about 3.8 hours a week for this type of work.

Pitfalls: Each agent using personal accounts with client data; no one owns the policy.

Source: ^[A12]

Playbook 3

Pre-selling condo sales

You sell project units and answer the same questions all day.

First five automations

• FAQ replies from developer-approved material
• Payment-scheme explainers (checked against the developer)
• Follow-up cadences
• Viewing and site-visit checklists
• Short project briefs

Guardrails

• Stay inside approved materials (PD 957 section 19)
• Ask for License to Sell and registration numbers
• No extra amenity claims

Estimated time saved: 4 to 8 hours a week ESTIMATE (ours, not measured)

Our estimator, typical level: about 4.0 hours a week for this type of work.

Pitfalls: AI embellishing renders or amenities; mixing up payment schemes.

Source: ^[A13]

Playbook 4

Rental agent

You list, screen and schedule many small leases.

First five automations

• Listing variants per platform
• Tenant screening Q&A scripts
• Viewing scheduler messages
• Lease abstracts (redacted)
• Owner update messages

Guardrails

• Tenant data stays out of public AI
• Lease summaries are not legal advice
• No discriminatory wording

Estimated time saved: 4 to 8 hours a week ESTIMATE (ours, not measured)

Our estimator, typical level: about 4.2 hours a week for this type of work.

Pitfalls: Over-trusting lease summaries; tenant data in chats; vague "near" claims.

Playbook 5

Commercial / warehouse broker

You prepare spec sheets and market notes for business clients.

First five automations

• Spec-sheet drafts
• Market summaries from your own data
• Letter-of-intent drafts for a lawyer
• Investor report outlines
• Meeting notes

Guardrails

• Recompute every number
• Client confidentiality
• Have a lawyer review LOIs

Estimated time saved: 2 to 5 hours a week ESTIMATE (ours, not measured)

Our estimator, typical level: about 2.0 hours a week for this type of work.

Pitfalls: Invented market data; confidential details pasted into public AI.

Playbook 6

Property manager

You run maintenance requests, tenant notices and owner reports.

First five automations

• Maintenance request triage
• Tenant notice drafts
• Monthly owner report outlines
• Schedule reminders
• Translation drafts

Guardrails

• Tenant privacy
• No automatic eviction language
• Human approves notices

Estimated time saved: 4 to 8 hours a week ESTIMATE (ours, not measured)

Our estimator, typical level: about 4.0 hours a week for this type of work.

Pitfalls: A tone that is too harsh; sending a notice with the wrong dates.

Playbook 7

New agent or trainee

You are learning scripts, the market and the paperwork.

First five automations

• Objection role-play
• A glossary of terms
• A daily study plan
• Script practice for calls
• Post-viewing notes

Guardrails

• Check every fact with your broker
• Do not copy AI statistics into messages
• Ask before recording

Estimated time saved: 3 to 5 hours a week ESTIMATE (ours, not measured)

Our estimator, typical level: about 3.0 hours a week for this type of work.

Pitfalls: Best supported by evidence (novices gained most in the Stanford/MIT study) but easiest to mislead by wrong facts.

Source: ^[A6]

Playbook 8

OFW and foreign-buyer desk

You serve buyers abroad across time zones, in several languages.

First five automations

• Translation drafts
• Time-zone-aware update messages
• Video-tour scripts
• Document checklists
• FAQ replies

Guardrails

• Verify every payment instruction out of band
• Foreign ownership limits need a law check (UNVERIFIED here)
• Native speaker checks contracts and prices

Estimated time saved: 3 to 6 hours a week ESTIMATE (ours, not measured)

Pitfalls: Remote payment fraud, deepfake calls, mistranslated terms.

Source: ^[A18]

Notice a pattern. The playbooks with the biggest estimated savings (pre-selling, rental and property management, each 4 to 8 hours) are the ones with the most repeated questions. The ones that need the most care (commercial, OFW and property management) involve money, confidentiality or tenants, so the guardrails matter more than the hours.

12. Where platforms help your workflow

Good tools sit around your own workflow, not instead of it. When you choose or use a listing platform, a portal or a CRM, there are eight things worth checking. Each one links back to the section where we explained it.

1. Does it turn pasted text into fields you review? Less retyping, no skipped checks. Section 6.
2. Does it handle photo compression and consistent ratios on upload? Then your originals can stay at home. Section 7.
3. Can you see what buyers want without exposing anyone? Demand data that feeds matching. Section 2.
4. Can first replies go through a neutral contact? Then AI drafts need not publish your number. Section 4.
5. Does search understand plain language? It is where buyers feel AI the most. Section 9.
6. Do project pages give a one-page brief with a verify step? A good template for AI briefs. Section 5.
7. Does it back answers with source-based articles? Section 10.
8. Is a human in the loop for reminders? One acknowledgement and one follow-up, not a barrage. Section 8.

One gap to be honest about: we could not confirm whether Lamudi, Dot Property or OLX offer AI listing tools, so we do not describe them (UNVERIFIED). Ask each platform what it actually does, and what it does with your data.

If you want to see how a one-page project summary looks, browse the premium pages at manilaopenhouse.com/premium.

13. A 30-day pilot and the metrics that matter

You do not need to believe us, or the infographic. Run a small experiment on your own work for 30 days and let your numbers decide. It costs almost nothing, and it is the only honest answer to “does AI help me?”.

The 8 numbers to track

Pick the ones that match your pilot tasks, not all of them. The first week gives you a baseline; every later number is compared with it. Report your sample size next to each figure, because a month rarely shows closings. Judge by leading indicators such as reply time, viewings booked and corrections.

Estimating the hours before you start

The estimator below models 14 typical agent tasks. For each task we set a baseline in minutes, and we assume a reduction of 10, 20 or 35 percent on tasks AI is suited to (conservative, typical or optimistic). On tasks outside AI’s strengths, such as market summaries, document review and investor reports, we count zero saving. That is our planning choice, based on the BCG finding that AI users do worse outside the frontier ^[A9].

ALL baseline minutes and reductions are ESTIMATES (ours). No Philippine data exists. Independent studies suggest roughly 10-30% on drafting tasks, more for novices (Stanford/MIT +14% average, +34% novices; BCG +25% speed inside the frontier). Our 10/20/35% levels are assumptions, not findings. BCG/Harvard found that outside the AI frontier users were less likely to be correct, so we count 0 saving on out-of-scope tasks (our planning choice) and show them separately.

For example, with the typical assumption, a solo agent’s 18.2 baseline hours a week fall by about 3.0 hours (about 13.0 hours a month, or 16.5 percent of the total). The saving is real, and it is much smaller than any 10x headline.

Try it: choose your agent type, edit the weekly task counts and pick an assumption level to see the hours saved per week and per month, with the out-of-scope tasks shown separately. It runs in your browser and nothing you type is sent anywhere.

14. Twelve mistakes to avoid

These are the errors we saw again and again in the research. The first three can cost you a client, a complaint or worse.

1. Pasting client IDs, TIN or bank documents into a public AI tool.
2. Publishing AI listing text without checking numbers and amenities.
3. Letting AI state prices, taxes, zoning or the status of a title.
4. Using AI-made or altered images of the unit without a label.
5. Posting AI renders for a registered project that go beyond the approved materials.
6. Auto-sending follow-ups to people who never opted in.
7. Trusting "10x faster" claims without measuring your own tasks.
8. Quoting the 5-minute response stat as proof of more closings.
9. Mixing US survey numbers with Philippine claims, or different surveys together.
10. Believing AI-made links and sources without opening them.
11. Replacing the broker's review of ads with "the AI checked it".
12. Accepting changed payment details by chat; ignoring deepfake and fake-owner scams.

15. Our recommendations

We promised a decisive guide. Here are our 8 recommendations, in order. If you only do the first four, you will avoid most of the harm and keep most of the benefit.

Frequently asked questions

Sources

Compiled 2026-10-15. The 26 sources match the bracketed citations in the text. Several pages could not be opened (HTTP 403 or blocked), so some numbers rest on search summaries; those are marked. Tiers: independent (research studies), secondary (law commentary or news), self-reported (surveys of agents) and vendor (a company selling something). Treat vendor sources as supporting evidence only.

1. [A1] NAR 2025 technology survey summary (GAAR / Florida Realtors), Sept 2025 (self-reported)
   https://www.gaar.com/blog/article/nar-realtors-embrace-ai-digital-tools-to-enchance-client-service
2. [A2] NAR technology report coverage, 2026-09-22 (Inman, RISMedia). Pages not opened (HTTP 403); numbers from search summaries (self-reported)
   https://www.inman.com/2026/09/22/nar-technology-report-ai-adoption/
3. [A3] Inman, Jan 2026, "everyone in real estate is using AI" (search snippet only) (self-reported)
   https://www.inman.com/2026/01/29/so-pretty-much-everyone-in-real-estate-is-using-ai-now/
4. [A4] LeadSigma, caveats on the five-minute lead-response statistic (vendor)
   https://www.leadsigma.com/resources/blog/five-minute-lead-response-stat.md
5. [A5] Lead response studies overview (InsideSales/MIT 2007; HBR 2011) (vendor)
   https://ainora.lt/blog/lead-response-time-5-minutes-study-2026
6. [A6] Brynjolfsson, Li, Raymond, "Generative AI at Work" (Stanford/MIT) (independent)
   https://papers.ssrn.com/abstract=4426942
7. [A7] Early impacts of Microsoft 365 Copilot (Microsoft Research) (independent)
   https://www.microsoft.com/en-us/research/?p=1136635
8. [A8] UK Department for Business and Trade Copilot trial (Civil Service World) (independent)
   https://www.civilserviceworld.com/news/article/dbt-trial-finds-copilot-saves-staff-time-but-doesnt-improve-productivity
9. [A9] Harvard/BCG "jagged frontier" study of 758 consultants (Harvard Crimson coverage) (independent)
   https://www.thecrimson.com/article/2023/10/13/jagged-edge-ai-bcg/
10. [A10] NPC Advisory 2024-04 on AI and the Data Privacy Act (Baker McKenzie; regulations.ai) (secondary)
    https://insightplus.bakermckenzie.com/bm/data-technology/philippines-npc-releases-new-guidelines-on-artificial-intelligence-and-child-oriented-transparency
11. [A11] AMLC covered persons, real estate (BusinessWorld; fetch blocked, summary only) (secondary)
    https://www.bworldonline.com/?p=374012
12. [A12] RESA, RA 9646 (batasnatin.com) (secondary)
    https://batasnatin.com/laws/ra-9646
13. [A13] PD 957 and License to Sell (batasnatin; respicio.ph and ren.ph commentary) (secondary)
    https://batasnatin.com/laws/pd-957
14. [A14] The Close, MLS rules on AI listing photos (2026) (secondary)
    https://theclose.com/news-ai-listing-photo-rules/
15. [A15] The Close, AI photo disclosure risk; Northstar MLS guidelines (secondary)
    https://theclose.com/news-ai-listing-photo-disclosure/
16. [A16] Consumer Act (RA 7394) art. 110 and the Ad Standards Council (ndvlaw; BusinessWorld) (secondary)
    https://ndvlaw.com/how-does-philippine-law-punish-false-advertising-by-influencers/
17. [A17] AI-generated work and copyright in the Philippines (Inquirer; Philstar) (secondary)
    https://business.inquirer.net/417098/copyright-ai-generated-work
18. [A18] Deepfake and real estate scam warnings (Philstar; DZRH on CICC; listd.ph) (secondary)
    https://www.philstar.com/business/2025/03/29/2431791/deepfakes-and-deep-trouble
19. [A19] Virtual staging cost comparisons (price-comparison blogs, vendor-ish) (vendor)
    https://aihomedesign.com/blog/services/ai-virtual-staging-guide/
20. [A20] WhatsApp Business API pricing and PH notes (ChatDaddy reseller blog; Meta docs) (vendor)
    https://chatdaddy.tech/blog/whatsapp-business-api-philippines
21. [A21] CRM pricing comparisons (Luxury Presence; Superdupr), US-centred (vendor)
    https://www.luxurypresence.com/blogs/lofty-pricing/
22. [A22] Transcription tools and Tagalog tests (CostBench; American Archive 2023; Soniox, a vendor) (vendor)
    https://blog.americanarchive.org/2023/08/24/assessment-of-whisper-ai-as-a-tagalog-language-transcription-tool/
23. [A23] AI subscription prices (eWeek; glbgpt PH guide) (vendor)
    https://www.eweek.com/news/ai-pricing-cheat-sheet-2026/
24. [A24] Compass AI Assistant (RISMedia, July 2026) (vendor)
    https://www.rismedia.com/2026/07/21/compass-expands-ai-capabilities-ai-assistant-home-platform/
25. [A25] SIM Registration Act and text scams (Inquirer) (secondary)
    https://newsinfo.inquirer.net/1952416/sim-registration-not-silver-bullet-vs-text-scams-ntc
26. [A26] Luxury Presence AI real estate statistics (vendor) (vendor)
    https://www.luxurypresence.com/blogs/ai-real-estate-statistics/

What we could not verify

These are the gaps. We list them so that you can weigh our advice honestly and so that the next edit knows what to check.

• No Philippine data exists on realtor AI adoption or time saved; every time figure in this post is our estimate.
• The 2026 NAR numbers (Inman/RISMedia) come from search summaries; the NAR report itself was not read, and the 2025 and 2026 surveys differ.
• The Inman January 2026 figures (97%, 82%) come from a search snippet; the underlying survey is unidentified.
• Whether PRC rules require a licence number or broker name in advertisements was not confirmed.
• PD 957 section 19 text was read through commentary sites, not the statute; whether a render disclaimer overrides it is our caution, not a verified rule.
• The scope of the AMLC covered-person rules for individual salespersons, and the KYC steps, were not confirmed (summary only).
• NPC Advisory 2024-04 details for small agents (registration, cross-border view of consumer chatbots) and the 72-hour breach notice rule were not verified.
• Whether RA 7394 article 110 applies to real estate listings, versus PD 957 and RESA, is untested.
• No Philippine rule on AI-image disclosure was found; the US examples (NAR ethics wording, California AB 723, Wisconsin, Metro MLS) are benchmarks only.
• All tool prices come from vendor or price-comparison blogs; peso prices and GCash payment are unverified.
• Taglish transcription quality for Otter and Fireflies, and the vendor error-rate figures for Whisper, are unconfirmed.
• Whether Lamudi, Dot Property or OLX offer AI features, Viber Business pricing, and NTC rules on commercial SMS were not found.
• Carrier filtering of SMS with links or ".com" comes from our own server experience, not an outside source.
• CAAP drone rules, foreign-ownership limits, PH copyright treatment of AI works, and current hazard-map URLs were not verified.