Skip to content
GuidesReal EstateMarketingFamilyLive EventseSports
All free classes

Visual AI for real estate: How can real estate agents use AI with photos instead of typing?

Take a photo or screenshot, crop out passwords and client details, and give it to the AI with one line about what you need instead of typing out what's in it. Then check every detail it gives back, because you, not the AI, answer for what reaches a client.

The slides, as a contact sheet.

All 18 frames from the class deck. Pick one to put it under the glass, or use the arrow keys.

Show, Don't Type Most of us are still typing descriptions of things AI can already see.

Slide 01 / 18

INNOVATION LAB · NORTH · AUGUST 11, 2026

Show, Don't Type

Read the slides as text

  1. 01 Show, Don't Type

    INNOVATION LAB · NORTH · AUGUST 11, 2026 · Most of us are still typing descriptions of things AI can already see. · Presented by Build Things That Build Things

  2. 02 Four ways to stop typing what Gemini can see.

    TODAY'S PLAN · Screens. Documents. Photos. Voice. Phone out, Gemini open. · 1 · SCREENS · the confusing screen becomes steps · 2 · DOCUMENTS · quotes in, questions out · 3 · PHOTOS · visible versus verify · 4 · VOICE · 60 seconds becomes a deal brief · PHONE OUT · GEMINI OPEN

  3. 03 The agents who love the speed still do not trust the output.

    WHY THIS CLASS EXISTS · 71% · NAME TIME SAVINGS AS AI'S TOP VALUE · Saving time is the number-one value. Accuracy is the number-one worry. Today teaches both at once. · WHY AGENTS USE IT · SAVES TIME · 71% · WHY AGENTS HESITATE · ACCURACY WORRIES · 63% · → · SPEED + JUDGMENT · Source: NAR, RPR survey of 225 members, Feb 12, 2026 · accuracy 63% · compliance 49%.

  4. 04 Observe. Orient. Decide. Act. You run every stage.

    KEEP CONTROL · AI is a tool inside the stages, never a stage of its own. AI generates. You validate. Repeat. · YOU RUN · EVERY STAGE · AI ASSISTS INSIDE. YOU DECIDE. · OBSERVE · SHOW IT THE ARTIFACT · ORIENT · SORT WHAT IS TRUE · DECIDE · PICK THE MOVE. YOURS. · ACT · SHIP, THEN LOOP

  5. 05 One sentence governs everything today.

    "AI can tell you what it sees. It cannot tell you what’s legally, factually, or professionally safe to conclude." · The Show, Don't Type rule · The output law: conclusions and sends stay human. No legal, inspection, valuation, or Fair Housing conclusions. Nothing sent, published, or entered without your review.

  6. 06 Three things never go into Gemini.

    BEFORE ANYTHING VISUAL GOES IN · Passwords and login codes. Wire instructions. Unredacted client files. Everything else you show it today is fair game. · THE GATE · SCREENSHOT ✓ · REDACTED DOC ✓ · LISTING PHOTOS ✓ · ✕ PASSWORDS · ✕ WIRE INSTRUCTIONS · ✕ CLIENT PII

  7. 07 Screenshot to solved: the confusing screen becomes steps.

    WORKFLOW 1 · SCREENS · MLS settings, error messages, dashboards, forms. It must also say what it cannot confirm. · THE CONFUSING · SCREEN · UPLOADED AS-IS · 1 · WHAT THIS SCREEN IS · 2 · WHAT IT IS TRYING TO DO · 3 · STEPS TO RESOLVE · 4 · WHAT IT CANNOT CONFIRM · 5 · WHEN TO CALL SUPPORT

  8. 08 Screenshot the screen that annoyed you this week.

    RUN IT NOW · Attach it and ask exactly this. Read the cannot-confirm answer first. · TYPE OR SAY EXACTLY THIS · “What is this screen and what is it trying to do?” · “What can you not confirm from this image?” · “What do I do next?”

  9. 09 Document to questions: quotes in, conclusions out.

    WORKFLOW 2 · DOCUMENTS · The sample: a fictional inspection excerpt. The double-tap at breaker 14 on page 2 becomes a question for the electrician. It does not recommend. · THE DOCUMENT · p. 4 · p. 11 · p. 23 · QUOTED, NOT PARAPHRASED · QUESTIONS FOR · The right professional · INSPECTOR · ATTORNEY · LENDER · HOA

  10. 10 Run the inspection excerpt. Check one quote yourself.

    RUN IT NOW · Use the sample from the QR page or any redacted document you have. · TYPE OR SAY EXACTLY THIS · “Quote the sections that matter, with page numbers.” · “What is missing from this document?” · “What should I ask the inspector or attorney?”

  11. 11 Listing photos in. Two lists out: visible, and needs verification.

    WORKFLOW 3 · PROPERTY PHOTOS · Visible features, a verification list, seller and buyer questions, three honest content angles. An image supplies observations, not verified property facts. · VISIBLE · ✓ GRANITE-LOOK COUNTERS · ✓ DECK OFF KITCHEN · ✓ PANEL: 200A LABEL · OBSERVATIONS ONLY · VERIFY FIRST · ? ROOF AGE · ? PERMITS · ? SQ FT · NOT PROPERTY FACTS YET

  12. 12 Point the camera. Ask a real question. Get real value.

    FOUR CAMERA MOVES · Fixture era. Inspector list. Comp-sheet figures. The money shot. · "WHAT ERA IS THIS FIXTURE?" · age + style, diplomatically worded · "WHAT MUST THE INSPECTOR CONFIRM?" · observations routed, never verdicts · PHOTO OF A COMP SHEET · every figure + what verifies it · THE MONEY SHOT · caption + room described for a buyer

  13. 13 Run three photos from your camera roll right now.

    RUN IT NOW · Client-safe photos only, or the demo set from the QR page. Name each photo as you attach it. · TYPE OR SAY EXACTLY THIS · “List everything visible in these photos.” · “List what still needs verification before I repeat it.” · “What will a buyer ask me about this house?”

  14. 14 A 60-second voice note becomes a complete deal brief.

    WORKFLOW 4 · VOICE · Objective notes, client preferences, open questions, tasks, a client-message draft, a CRM-ready note. Nothing is sent or saved automatically. · 60-SECOND VOICE NOTE · OBJECTIVE NOTES · CLIENT PREFERENCES · OPEN QUESTIONS · FOLLOW-UP TASKS · DRAFT · NOT SENT · FACTS TO CONFIRM

  15. 15 Record 60 seconds about your last showing.

    RUN IT NOW · Property, reactions, exact client quotes, promises made, next steps. Upload the audio and ask exactly this. · TYPE OR SAY EXACTLY THIS · “Turn this into objective notes, client preferences, and open questions.” · “Draft the follow-up message and a CRM note. Do not send anything.” · “What facts do I still need to confirm?”

  16. 16 One loop for every artifact you will ever show AI.

    THE BUILD ROOM · Outcome, artifact, context, surfaced unknowns, verification, one action. · 1 · OUTCOME · 2 · ARTIFACT · 3 · CONTEXT · 4 · SURFACE UNKNOWNS · 5 · VERIFY · 6 · ONE ACTION · REPEAT ON THE NEXT REAL PROBLEM

  17. 17 Pick one thing already on your phone. Run the whole loop.

    PICK ONE REAL PROBLEM · A screenshot, listing photos, or a redacted document. Or record 60 seconds. Right now.

  18. 18 Stop describing one problem. Show it instead.

    THIS WEEK · Make AI name what it does not know, then take the next verified action.

Roll the reel.

Reel 144:39Recorded live at Innovation Lab North

Watch on YouTube (opens in a new tab)

Read and practice on this page. Everything on this page is free: the recording, the slides, the prompts, and the workbook. No account needed. The free portal account is for downloading the files.

  • VisibleDeck off the kitchen. It's in the frame.
  • UncertainGranite, or granite-look? The photo can't settle it.
  • VerifySquare footage and permits. No photo proves them.
Illustration, not a listing photo. Magnified isn't verified.

The one ruleAI can tell you what it sees. It can't tell you what's legally, factually, or professionally safe to conclude.

You leave with four ways to show instead of type (screens, documents, photos, voice) and one habit: mark every answer the way this print is marked.

Taught live by Ryan Wanner Innovation Lab North, Compass Lakeside, Hendersonville August 11, 2026

Each class was taught live by Ryan Wanner to working real estate agents.

The method · 8 steps

Show it, then check it.

The order the class runs it, from the photo on your phone to the answer you're willing to put your name on.

  1. Step 1: Reach for a picture before you type

    If you're about to type out what a room, a screen or a page looks like, take a picture instead. Ask yourself one thing: will this image tell the AI more than my words would?

  2. Step 2: Check permission and the whole image

    Use a picture you are allowed to share in an approved account. Remove passwords, login codes, wire instructions and unnecessary identifying details, including background text and metadata. Cropping alone may not prevent identification; use a fictional or public example when unsure.

  3. Step 3: Say the one job you need done

    Attach the image and name the job: what is this error and how do I fix it, or what will a buyer ask about this kitchen. If talking is faster, say it out loud instead of typing.

  4. Step 4: Ask what it can't confirm

    Have the AI list what it can see and, apart from that, what the picture can't prove: roof age, permits, square footage, what's behind a wall. That second list is what you check with the seller or inspector.

  5. Step 5: Fix what you know before it goes out

    If it says the counters look like granite and you know they are, tell it. Correct the square footage if it's off. Nothing goes to a client until you've read it, because you answer for what the picture really shows.

  6. Step 6: Turn a permitted page image into questions

    Use a page you are allowed to share. Ask for relevant passages, page references and a plain-English reading, then compare every quoted passage and factual claim with the original. Escalate legal, repair and value decisions to the appropriate qualified professional.

  7. Step 7: Stage for a client's taste and label it

    With a Context Card that notes a client's furniture style, AI can stage an empty room their way. Keep the card to taste, not who they are. Label every staged image as virtually staged.

  8. Step 8: Tell it to use your photos exactly

    When you turn listing photos into a video or a tour, tell it: use exactly these photos, don't edit, don't add. AI video likes to add a plant or a door. Check every frame against the real house.

Use the class workbook · Get the supporting files

Read and work through the lesson here. Downloadable files use your free student account.

Frames from the recording.

Ryan's own words from the class. Each time opens the recording at that moment.

  1. Watch from 0:49

    “it is so much easier to just send an image into the AI than to describe what's going on within the image.”

    Ryan's starting point: when a picture can say it, sending the picture beats writing it all out.

  2. Watch from 15:52

    “it'll say like the countertops look like they're granite.” You know for a fact they're granite. Go in and make that change.”

    AI drafts a listing description from photos and guesses at finishes, so you fix what you know for a fact.

  3. Watch from 17:09

    “It can tell you what it thinks about a picture, but you are the person who's ultimately responsible for the reality of that picture.”

    AI can describe a picture, but you answer for what's really there before anything reaches a client.

  4. Watch from 19:34

    “if it's something that you don't want in the AI, crop the picture and just give it a little bit of extra written context.”

    If a photo holds something that shouldn't go into AI, crop it out and type the bit of context the AI still needs.

  5. Watch from 20:07

    “push that into my AI and say, hey, my computer is running into this error. What is it? How do I fix it?”

    A screenshot of an error and two plain questions get you an explanation of what's wrong, even when the fix isn't yours to make.

  6. Watch from 28:45

    “You upload the image and you say, “Stage this room for this client.” It'll stage it with furniture that matches their style.”

    With a client's Context Card that notes their furniture style, AI can stage an empty room for that one client instead of for anybody.

  7. Watch from 32:38

    “So you can say like quote the sections that matter, give me page numbers.”

    With photos of a long document, ask for the passages that matter and their page numbers, so you can check each one yourself.

  8. Watch from 33:35

    “Hey, what are all the things that my buyer would want to negotiate about this?”

    A listing photo can become the list of things a buyer may want to negotiate, before they bring it up.

  9. Watch from 34:43

    “you can actually go around the room and be like, “Hey, what's this? Hey, what's this? Tell me about this.””

    On the Gemini phone app, Live lets you walk a room with the camera and ask about what it sees as you go.

  10. Watch from 37:51

    “can I add this image and it's going to give the AI more than I could if I was just writing text to it?”

    Ryan's test for any problem: will an image give the AI more than your typed words would?

  11. Watch from 43:00

    “use exactly the photos. Don't edit, don't add. You are just making a video walkthrough from my photos.”

    When AI turns listing photos into a video, tell it plainly not to change or add anything, because AI video likes to add things that aren't there.

Where showing goes wrong.

Typing out what a picture could show
Describing a room, a screen or a page in a long message takes longer and leaves things out. If a picture would settle it, send the picture.
Trusting what the AI says it sees
Visible color or layout can be described as an observation. Material type and square footage are inferences or unknowns until checked against a reliable source. Record that distinction and verify before use.
Uploading the whole screen
A screenshot can carry a password, a client's name or a loan number along with the error you care about. Crop to the part you need first.
Letting AI video add to the house
AI video likes to add a plant or an extra door. If you don't tell it to use your photos exactly, the tour can show a house that doesn't exist.
Posting staged photos without a label
A virtually staged room is an idea, not the room's real condition. Label it as virtually staged every time it goes out.
Adding schools or neighborhoods to a staging card
To stage a room, the AI needs the client's furniture style and taste. Leave out schools, safety and who lives nearby. Describe the property, never the people.

Workbook · Show, Don't Type: AI That Can See What You See

The workbook, on the light table.

What you're building: a repeatable visual AI workflow: show Gemini a screen, document, property photo set, or voice note and get verified, role-safe next actions.

Set up Gemini.

The class runs on Gemini: the browser for desk problems, the phone for field problems.

Setup 1 of 2

Gemini (app + gemini.google.com)

One AI, two doors: gemini.google.com on desktop for screenshots and file uploads, the mobile app for camera, photos, and voice.

Best for Every workflow in this class. Desk problems go through the desktop door, field problems go through the phone door.

Before anything client-adjacent Check the Keep Activity setting before anything client-adjacent: gemini.google.com > Settings & help > Activity. Even when off, chats are kept up to 72 hours and Google says human reviewers review some data.

Set it up · 4 steps, then your first five minutes
  1. Install the Gemini app on your phonegemini.google.com/app (opens in a new tab)App Store on iPhone, Google Play on Android. Search for Google Gemini.
  2. Sign in with the Google account you use for workNot a shared family account. The chats you create here are work material.
  3. Open gemini.google.com on your desktop and confirm the same accountgemini.google.com/app (opens in a new tab)
  4. Find the plus/attach affordance next to the prompt boxThat is where screenshots, photos, documents, and audio files enter.

Your first five minutes

  • Take a screenshot of any non-client screen and attach it to a test prompt on desktop.
  • On the phone, attach one photo from your library and ask what is visible in it.
  • Copy one prompt from the class prompt pack and run it end to end.
  • Keep every output in draft. A draft is not a send.

Setup 2 of 2

Gemini Live (camera + screen share)

Point your camera at something, or share your phone screen, and talk about it in real time.

Best for Field questions: a fixture you cannot name, a material you cannot identify, a confusing app screen while you are standing somewhere.

Set it up · 4 steps, then your first five minutes
  1. Open the Gemini mobile appGemini Live is mobile-app only (Android and iOS). It is not available in the Gemini web app.
  2. Tap LiveOn Android you can also swipe left to reach Live. Steps verified against Google's July 22, 2026 walkthrough at blog.google.
  3. Tap Turn on video streaming to share your camera, or Turn on screen sharing to share your screen
  4. Talk about what the camera or screen is showingNever point Live at client documents, wire instructions, or anything with unredacted client information.

Your first five minutes

  • Start a Live session and point the camera at any object on your desk. Ask what it is.
  • Share your screen and open a settings page you find confusing. Ask what each option does.
  • End the session and review what was discussed.
  • Decide your personal rule for what the camera never sees: documents, screens with client data, other people.

The gate.

The privacy gate runs before anything is attached. The verification pass runs before you act on an answer.

The privacy gate

The verification pass

Fair Housing: describe the property, never the people. AI doesn't decide who a home or a neighborhood is for. Anything that touches protected classes, steering, or neighborhood suitability goes to your broker, not the chatbot.

Paste this first.

A compact card you paste ahead of any artifact so every visual analysis stays bounded, honest, and reviewable.

quick-context-sheet.md, the visual workflow context card

Fill in the brackets once. Then paste it ahead of any screenshot, photo, document, or voice note.

quick-context-sheet.md
# Visual Workflow Context

## Who I am
I am [name], a licensed real estate agent in [state], working primarily in [market/city], affiliated with [brokerage].

## Your role
You are a visual analyst. You describe, organize, and question. You do not advise, conclude, or decide. You never state legal, inspection, valuation, or Fair Housing conclusions; those become questions for a named professional.

## Privacy gate
Never request or repeat passwords or login codes, unredacted client information, or wire instructions. If the artifact contains any of those, say so and stop.

## Output discipline
Separate everything you produce into OBSERVED / INFERRED / CANNOT CONFIRM. Never promote an inference to a fact. Flag ambiguity instead of resolving it.

## Draft state
Everything you produce is a draft for my review. Nothing gets sent, published, or entered into a system without a human reading it first. A draft is not a send.

Four ways to show, and a bonus.

Each recipe names what AI can't confirm from that kind of artifact. Copy the prompt, open Gemini, paste, then attach a permitted copy of the artifact. Remove private details first; preserve the relevant visual evidence.

On the table for every recipe quick-context-sheet.md + the artifact itself

  1. 1A Screenshot to solved

    The confusing screen becomes steps.

    How to run it Privacy gate first: crop out client data before capturing. Copy the prompt, open gemini.google.com/app, paste, attach the cropped screenshot. One screen per prompt.

    Can't confirm Account state, settings on other screens, permissions, the data behind the interface.

    Good output looks like · 4 checks
    • WHAT THIS SCREEN IS / WHAT IT IS TRYING TO DO / STEPS TO RESOLVE / WHAT I CANNOT CONFIRM FROM THIS IMAGE / WHEN TO CALL SUPPORT.
    • Steps reference things actually visible in the screenshot, not generic platform advice.
    • The CANNOT CONFIRM block is specific. If it is empty, the AI is overclaiming.
    • Support handoff names the vendor and includes a paste-ready problem description.

    The prompt: Prompt 2 in the pack

  2. 2A Document to questions for the right professional

    Quotes in, questions out.

    How to run it Fictional or fully redacted documents only. Never a live client contract. If Gemini rejects a big PDF, split it and attach the parts. Copy-then-open flow: prompt first, then the file.

    Can't confirm Exhibits, dates, amounts and attachments the document mentions but doesn't contain.

    Good output looks like · 4 checks
    • QUOTED LANGUAGE with page references / PLAIN-ENGLISH SUMMARY / MISSING INFORMATION / QUESTIONS FOR THE PROFESSIONAL.
    • Every quote is verifiable against the page in seconds.
    • Questions are addressed to a named role and forwardable as written.
    • No recommendations, no advice, no conclusions anywhere in the output.

    The prompt: Prompt 3 in the pack

  3. 3A Property images to intelligence

    Visible versus verify.

    How to run it Name each photo before uploading: kitchen, panel, roofline. Privacy gate: no people, no documents, no client data visible in frame.

    Can't confirm Age of systems, what's behind the walls, permits, square footage, whether anything works.

    Good output looks like · 4 checks
    • VISIBLE FEATURES / NEEDS VERIFICATION / QUESTIONS FOR THE SELLER OR INSPECTOR / QUESTIONS A BUYER WILL ASK / THREE HONEST CONTENT ANGLES / ONE FACTUAL FOLLOW-UP DRAFT.
    • The NEEDS VERIFICATION list is the product; it becomes the seller questionnaire and the inspector conversation.
    • Content angles are supported entirely by what is visible.
    • No comments on who a property or neighborhood is suited for.

    The prompt: Prompt 4 in the pack

  4. 4A Voice note to deal brief

    60 seconds becomes a deal brief.

    How to run it Record 60 seconds in your phone recorder: property, reactions, exact client quotes, promises made, next steps. Upload the audio file or dictate in the app. Privacy gate before upload.

    Can't confirm Anything you said from memory. It stays a draft until you check it.

    Good output looks like · 4 checks
    • OBJECTIVE NOTES / CLIENT PREFERENCES / OPEN QUESTIONS / FOLLOW-UP TASKS / CLIENT MESSAGE DRAFT / CRM-READY NOTE / FACTS NEEDING CONFIRMATION.
    • Client quotes come from the recording, not from invention.
    • Every task has an owner and a date, or is flagged for the agent to set one.
    • The message draft stays a draft until the confirmation list is checked.

    The prompt: Prompt 5 in the pack

  5. 5A Bonus: confusing email screenshot triage

    A confusing email becomes one safe next step.

    How to run it Privacy gate first: crop client data before capturing. Copy the prompt, open gemini.google.com/app, paste, attach the cropped screenshot. AI never decides that a suspicious email is safe.

    Can't confirm Sender authenticity, link destinations, account state, anything outside the visible message.

    Good output looks like · 4 checks
    • WHAT THIS EMAIL IS / RED FLAGS / WHAT IT MEANS FOR ME / WHAT I CANNOT CONFIRM FROM THIS IMAGE / NEXT ACTION.
    • Anything touching money movement routes to phone verification on a known number.
    • Legitimacy uncertainty is stated, never smoothed over.
    • One safe next action, not a reply draft to a possibly fraudulent sender.
    Recipe 5A · Bonus: confusing email screenshot triage
    I am a real estate agent. The attached screenshot is an email I do not fully understand: a vendor notice, a platform alert, a lender update, or something that might be spam or a phishing attempt. I have cropped out client information. Do not guess at anything the image does not show.

Respond in exactly these blocks:

WHAT THIS EMAIL IS
Who it appears to be from, what kind of message it is, and what it is asking me to do. If the sender or legitimacy is unclear, say so plainly.

RED FLAGS
Anything visible that suggests phishing or fraud: mismatched sender, urgency pressure, requests for credentials or payment details, odd links. If wire instructions or money movement are involved in any way, tell me to verify by phone using a known number and to never confirm through the email itself.

WHAT IT MEANS FOR ME
A plain-language reading of the content, labeled as a reading, not advice.

WHAT I CANNOT CONFIRM FROM THIS IMAGE
Sender authenticity, link destinations, account state, and anything outside the visible message. Mark each item CANNOT CONFIRM.

NEXT ACTION
One safe next step. If the email is conclusion-shaped (legal, money, contract terms), the next step is a question for the responsible professional, not a reply.

Your seven days.

One frame at a time. Each frame is one real artifact, run end to end.

  1. TODAY

    Run the screenshot workflow on one real confusing screen

    Pick the screen that has actually been slowing you down. Crop it, run the Screenshot Troubleshooter, and act on the verified steps.

  2. TOMORROW

    Turn one document into questions for the right professional

    Fictional or fully redacted only. Send at least one of the generated questions to the actual inspector, attorney, lender, HOA, or appraiser it names.

  3. THIS WEEK

    Turn one listing photo set into a verification list

    Name the photos, run the Property-Image Analyst, and convert the NEEDS VERIFICATION list into your seller questionnaire.

  4. DAY 5

    Turn one post-showing voice note into a deal brief

    60 structured seconds: property, reactions, exact client quotes, promises, next steps. Review the FACTS NEEDING CONFIRMATION list before the client message leaves draft.

  5. ONGOING

    Before every paste, run the privacy check and keep drafts as drafts

    Name, address, financial info, personal info: sanitize first. Nothing gets sent, published, or entered anywhere without human review.

The prompts, on a proof sheet.

All seven prompts from the class, as written. Long prompts start folded. Copy always copies the whole prompt.

Show, Don't Type: AI That Can See What You See

  • Class: Show, Don't Type: AI That Can See What You See
  • Date: August 11, 2026
  • Location: Compass RE Lakeside, Innovation Lab North

How to use this pack

To use the prompts below: copy the prompt below, open gemini.google.com/app on desktop (or the Gemini app on your phone), paste the prompt, then attach the artifact. Screenshot, photo, document, or audio file. The prompt goes in first, the artifact goes with it, and the answer comes back structured.

Use one prompt at a time. Remove private details and irrelevant material first. Preserve the visual evidence needed for the question; if you cannot share it safely, describe a fictional example instead.

The privacy gate

Before anything gets attached or pasted, run the gate:

  1. Never share passwords or login codes.
  2. Never share unredacted client information.
  3. Never share wire instructions.
  4. Never accept legal, inspection, valuation, or Fair Housing conclusions from AI.
  5. Nothing gets sent, published, or entered into a system without human review.

If an artifact fails the gate, redact it or do not use it. A screenshot with a client's full name and loan number visible is not class material and not workflow material.

Fair Housing

Visual analysis never produces conclusions about protected classes, steering, or neighborhood suitability. AI does not decide who a home or a neighborhood is for, and it does not interpret what it sees through demographic assumptions. Any question that touches Fair Housing, legal exposure, property condition conclusions, or valuation goes to the responsible professional: your broker, an attorney, an inspector, a lender, or an appraiser.

The core rule for the whole class: AI can tell you what it sees. It cannot tell you what's legally, factually, or professionally safe to conclude.

And the output rule: everything these prompts produce is a draft. A draft is not a send.

  1. Prompt 1 The show-it handoff (any artifact)

    Prompt 1
    I am a licensed real estate agent in [state], working in [market]. I am attaching an artifact: [what it is, e.g. a screenshot of my MLS listing input screen / a photo set of a property / a fictional inspection summary]. This artifact is [what it is not, e.g. not a legal document, not a verified property record, not client-approved material]. Here is what I need from it: [one specific job, e.g. "tell me what this screen is asking for and what is missing before I can submit"].

You describe, organize, and question. You do not advise, conclude, or decide. Your output is for me only.

Rules:
- Never ask me for, and never repeat back, passwords, login codes, wire instructions, or unredacted client information.
- Separate everything you produce into OBSERVED / INFERRED / CANNOT CONFIRM. Never promote an inference to a fact.
- Never state a legal, valuation, inspection, or Fair Housing conclusion. If the artifact raises one of those questions, phrase it as a question for the responsible professional and name that professional.

Respond in exactly these blocks:

1. OBSERVED
What the artifact literally shows or says. No interpretation in this block.

2. INFERRED
What the artifact reasonably suggests, each item clearly labeled as an inference, not a fact.

3. CANNOT CONFIRM
What the artifact cannot prove, however plausible it seems. If this block is empty, you are overclaiming.

4. QUESTIONS FOR THE RESPONSIBLE PROFESSIONAL
Every conclusion-shaped item, converted into a question and addressed to a named role: broker, attorney, inspector, lender, appraiser, or HOA.

5. ONE USABLE NEXT ACTION
Exactly one concrete step I can take in the next ten minutes.

Do not send, publish, or enter anything anywhere. The human reviews first.

    When to use it · what good looks like

    When to use: As the reusable frame for any visual workflow. Fill in the brackets once, save it, and paste it ahead of any screenshot, photo, document, or voice note when the specialized prompts below do not fit.

    What good output looks like: Clearly separated blocks with OBSERVED, INFERRED, and CANNOT CONFIRM kept honest, no advice dressed up as fact, and a single concrete next action. If the output blends observation and conclusion into one paragraph, re-paste the prompt and tighten the job line.

  2. Prompt 2 Screenshot Troubleshooter

    Used in Recipe 1A, Screenshot to solved

    Prompt 2
    I am a real estate agent and I am stuck on the screen in the attached screenshot. I have removed client information before capturing it. Do not guess at anything the image does not show.

Analyze the screenshot and respond in exactly these blocks:

WHAT THIS SCREEN IS
Identify the software or platform if the image supports it. If you are not certain, say so.

WHAT IT IS TRYING TO DO
Explain what this screen wants from me in plain language: what it is asking for, what state it is in, and what any visible error or warning means.

STEPS TO RESOLVE
Numbered steps, based only on what is visible. Start from what is on screen right now.

WHAT I CANNOT CONFIRM FROM THIS IMAGE
List everything relevant that the screenshot does not show: account state, settings on other screens, permissions, data behind the interface. Label this block CANNOT CONFIRM material and do not fill the gaps with guesses.

WHEN TO CALL SUPPORT
If the fix likely lives outside this screen, tell me which vendor to contact and give me two sentences I can say or paste that describe the problem precisely.

    When to use it · what good looks like

    When to use: Any confusing screen: MLS input pages, transaction platform errors, e-sign envelopes that will not send, CRM settings, lockbox apps, marketing dashboards. Crop the screenshot to the confusing part, one screen per prompt.

    What good output looks like: Steps that reference things actually visible in your screenshot, not generic advice for the platform in the abstract. The CANNOT CONFIRM block should be honest and specific. If it is empty, the AI is overclaiming.

  3. Prompt 3 Document Questioner

    Used in Recipe 2A, Document to questions for the right professional

    Prompt 3
    I am a real estate agent. The attached document is fictional or fully redacted. I want to understand it well enough to ask better questions. I am not asking you to interpret it as a professional would.

Respond in exactly these blocks:

QUOTED LANGUAGE
Pull the exact passages that carry the most weight, quoted word for word, each with its page reference. Do not paraphrase inside this block.

PLAIN-ENGLISH SUMMARY
What the document appears to say, in plain language, clearly framed as a reading of the text and not a professional interpretation.

MISSING INFORMATION
What the document references but does not contain: exhibits, dates, amounts, defined terms, attachments. Mark each item CANNOT CONFIRM from this document alone.

QUESTIONS FOR THE PROFESSIONAL
Specific questions this document should generate, each addressed to a named role: inspector, attorney, lender, HOA, or appraiser. Write them so I can ask them verbatim.

Do not recommend, advise, or conclude. Your job ends at quotes, summary, gaps, and questions.

    When to use it · what good looks like

    When to use: Turning a dense document into questions for the right professional. Use fictional or fully redacted documents only: sample inspection summaries, blank HOA covenant templates, redacted addendum language. Never a live client contract with names, addresses, or terms visible. If Gemini rejects a large PDF, split it into smaller files and attach the parts.

    What good output looks like: Real quotes with page numbers you can verify in seconds, and questions sharp enough to forward to an attorney or inspector as written. If any block drifts into "you should," the fence failed. Restate the last line and rerun.

  4. Prompt 4 Property-Image Analyst

    Used in Recipe 3A, Property images to intelligence

    Prompt 4
    I am a real estate agent. Attached are photos I took at a property. Analyze what is visible and respond in exactly these blocks:

VISIBLE FEATURES
What the photos actually show: materials, finishes, apparent condition of visible surfaces, layout clues, notable items. Reference photos by name.

NEEDS VERIFICATION
Everything a reasonable person might assume from these photos that the photos cannot actually prove: age of systems, condition behind surfaces, permits, square footage, whether anything works. Mark each item CANNOT CONFIRM.

QUESTIONS FOR THE SELLER OR INSPECTOR
Specific questions the photos raise, phrased for the person who can actually answer them.

QUESTIONS A BUYER WILL ASK
The questions a serious buyer or their agent will ask after seeing these photos, so I can prepare answers before they come.

THREE HONEST CONTENT ANGLES
Three marketing angles supported entirely by what is visible. No superlatives the photos do not earn.

ONE FACTUAL FOLLOW-UP DRAFT
One short message to the seller listing what I still need from them, based only on the gaps above.

Observations only. An image is not a verified property fact. Never comment on who a property or neighborhood is suited for.

    When to use it · what good looks like

    When to use: A photo set from a listing appointment, a walkthrough, or a pre-listing visit. Name each photo before uploading so the AI can reference them: "kitchen", "panel", "roofline". Up to 10 files per prompt.

    What good output looks like: A NEEDS VERIFICATION list longer than you expected. That list is the product: it becomes your seller questionnaire and your inspector conversation. The content angles should read like a careful agent wrote them, not a brochure.

  5. Prompt 5 Voice-Note Deal Brief

    Used in Recipe 4A, Voice note to deal brief

    Prompt 5
    I am a real estate agent. The attached audio is my post-showing voice note. Turn it into a structured deal brief. Use only what I actually said. Where my note is ambiguous, flag it rather than resolving it.

Respond in exactly these blocks:

OBJECTIVE NOTES
What happened: property, time, who attended, what was seen. Facts only.

CLIENT PREFERENCES
What the client liked, disliked, and said. Quote my wording where I quoted them.

OPEN QUESTIONS
Everything the note leaves unresolved, including anything I said uncertainly.

FOLLOW-UP TASKS
Every commitment or implied task, each with an owner and a suggested date. If I did not give a date, mark it for me to set.

CLIENT MESSAGE DRAFT
One short message to the client that proves I heard them and moves us one step forward. No filler, no fake urgency.

CRM-READY NOTE
A compact note I can paste into my CRM: property, outcome, preferences, next action, next date.

FACTS NEEDING CONFIRMATION
Anything I stated from memory that should be verified before it goes to the client or into a record. Mark each item CANNOT CONFIRM until checked.

A draft is not a send. Nothing leaves draft without my review.

    When to use it · what good looks like

    When to use: Right after a showing, while it is fresh. Record a 60-second voice note in your phone's recorder app covering the property, your client's reactions, their exact words where you remember them, anything you promised, and next steps. Then upload the audio file to Gemini with this prompt, or open the Gemini app and dictate the same content. Audio uploads work on the free tier up to 10 minutes total.

    What good output looks like: A brief that sounds like your showing, not a generic template. The client quotes should be your actual words from the recording. The message draft stays a draft until you have read it, checked the FACTS NEEDING CONFIRMATION list, and decided to send.

  6. Prompt 6 Staged or real? (photo pair check)

    Prompt 6
    I am a real estate agent. Attached are two or more photos of the same room, or a single listing photo that does not look right to me. Compare what is actually in the images. Do not guess beyond what the pixels show.

Rules:
- Never ask me for, and never repeat back, passwords, login codes, wire instructions, or unredacted client information.
- Separate everything you produce into OBSERVED / INFERRED / CANNOT CONFIRM.
- Never state a legal, valuation, inspection, or Fair Housing conclusion. If one comes up, phrase it as a question for the responsible professional and name that professional.

Respond in exactly these blocks:

1. WHAT DIFFERS BETWEEN THESE PHOTOS
Item by item: furniture, decor, fixtures, wall color, flooring, window views, anything present in one image and absent in another. Reference photos by name. If only one photo is attached, describe what looks inconsistent within it.

2. SIGNS OF VIRTUAL STAGING OR EDITING
Look at contact shadows under furniture, lighting direction across objects, reflections in windows and mirrors, and perspective consistency. Report what you observe about each, and say plainly when an element does not match the physics of the room.

3. VERDICT WITH CONFIDENCE
Your read on whether the photo appears virtually staged or edited, stated with a confidence level and the observations that support it. This is an opinion about images, not a fact about the property.

4. CANNOT CONFIRM
Everything the images cannot prove: what the room actually contains today, when each photo was taken, what tool made any edit, and anything outside the frame.

Reminder: virtually staged marketing photos must be labeled "Virtually Staged." Unlabeled staging is misrepresentation.

Do not send, publish, or enter anything anywhere. The human reviews first.

    When to use it · what good looks like

    When to use: Two or more photos of the same room, or any listing photo that smells off. Buyers ask this question constantly now, and you should be able to answer it before they do. Attach the photos together in one prompt and name them: "listing photo", "my photo from the showing".

    What good output looks like: Differences you can point to in the actual images, physics observations you can check yourself in ten seconds, and a verdict that stays an opinion with a confidence level. If the answer matters for a transaction, the next step is a direct question to the listing agent, not a screenshot of the AI's verdict.

  7. Prompt 7 Photo of a paper → figure extraction

    Prompt 7
    I am a real estate agent. Attached is a photo of a paper document: a comp sheet, a repair estimate, or a settlement page. I have redacted client information before photographing it. Extract the figures. Do not judge whether any amount is fair, high, low, or correct. Extraction only.

Rules:
- Never ask me for, and never repeat back, passwords, login codes, wire instructions, or unredacted client information.
- Separate everything you produce into OBSERVED / INFERRED / CANNOT CONFIRM.
- Never state a legal, valuation, inspection, or Fair Housing conclusion. Numbers get extracted; conclusions go to the responsible professional.

Respond in exactly these blocks:

1. EVERY FIGURE ON THIS PAGE
A list: each number, its label as printed, and where it sits on the page. If a figure is partially obscured or blurry, say so instead of completing it.

2. WHAT DOCUMENT OR SOURCE VERIFIES EACH FIGURE
For each figure, name the record that would confirm it: MLS record, invoice, payoff statement, tax record, lender disclosure. You are naming where verification lives, not verifying.

3. FIGURES THAT LOOK INCONSISTENT OR INCOMPLETE
Totals that do not sum, dates out of order, missing line items, figures referenced but not shown. Report the arithmetic, not a judgment about intent.

4. CANNOT CONFIRM
Everything a photo of a page cannot prove: whether this is the current version, whether pages are missing, whether the figures were entered correctly at the source.

5. QUESTIONS FOR THE PREPARER
Specific questions this page raises, phrased so I can ask the person who produced the document verbatim.

A photo of a document is not the document. Confirm every figure against the original before acting on it.

Do not send, publish, or enter anything anywhere. The human reviews first.

    When to use it · what good looks like

    When to use: A comp sheet on the table, a repair estimate on a clipboard, a settlement page in a folder. Photograph the page, run this prompt, and get every number out of it and into a checkable list. Redact client names and account numbers before you photograph.

    What good output looks like: A figure list you can check against the page in under a minute, verification sources named for every number, and inconsistencies reported as arithmetic rather than accusations. If a settlement figure or payoff amount is involved, the original document and the preparer are the only sources that count.

The files, in their sleeve.

Create a free student portal account to download this class's files.

The three PDFs are fictional practice documents made for the class. Run the recipes on them before you run anything real.

  • Show, Don't Type Prompt Pack

    Seven paste-ready visual prompts: the show-it handoff, screenshot troubleshooting, document questioning, property-photo analysis, voice-note deal briefs, staged-photo checks, and paper-to-figures extraction.

    Markdown16 KBFree account

    Get the file: Show, Don't Type Prompt Pack

  • Gemini Visual Setup Guide

    Gemini app and web setup, Gemini Live camera and screen share, upload rules, and the Keep Activity privacy check.

    Markdown5 KBFree account

    Get the file: Gemini Visual Setup Guide

  • Visual AI Privacy and Verification Checklist

    The one-page gate: five things that never go into a visual prompt, and the verification pass that runs before anything acts on an answer.

    Markdown2 KBFree account

    Get the file: Visual AI Privacy and Verification Checklist

  • Demo: inspection report excerpt (fictional)Fictional demo

    Four-page fictional inspection excerpt for Workflow 2. Page-referenced findings, a safety item, and gaps AI must flag as CANNOT CONFIRM.

    PDF7 KBFree account

    Get the file: Demo: inspection report excerpt (fictional)

  • Demo: HOA covenant excerpt (fictional)Fictional demo

    Three-page fictional CC&R excerpt: fencing, a 15% leasing cap with a hardship exception, and architectural review deadlines to quote and question.

    PDF5 KBFree account

    Get the file: Demo: HOA covenant excerpt (fictional)

  • Demo: repair estimate (fictional)Fictional demo

    Two-page fictional contractor estimate with an unpriced allowance and exclusions. Run the figure-extraction move: every number and what verifies it.

    PDF4 KBFree account

    Get the file: Demo: repair estimate (fictional)

  • Class skill card

    The class method on one page, written so you can hand it to your AI as instructions.

    Markdown15 KBFree account

    Get the file: Class skill card

Asked across the table.

Real questions from the people in the class, with Ryan's answer. Each time opens the recording where the question starts.

  • How do I get Gemini to look at a long video?

    Asked at 5:21: “so how would I do that? You're saying I just download it to my YouTube channel.”

    Ryan said to upload the video to your YouTube channel, copy its link and give that link to Gemini. Google owns both YouTube and Gemini, so they work well together, and Gemini can work from the video's transcript and pull snippets from it.

  • Should my videos be vertical or horizontal?

    Asked at 15:00: “Do vertical videos get more views than the horizontal ones?”

    Ryan said it depends on where you post. Instagram, TikTok and YouTube Shorts want vertical video for the format. Long-form YouTube is fine horizontal.

  • Can I give Gemini my brand Context Card so the videos it makes match my look?

    Asked at 21:09: “so that way Gemini knows like my look and feel, my brand, and then they can create videos based on that.”

    Yes. Ryan added that he likes to give the AI source media too, like a clip of you talking outside a listing, because AI works well with real video of people and has a harder time making something from scratch. A picture or a clip gives it a starting point.

Read the edited class transcript.

Automatic captions from the recording, lightly corrected. Stretches with personal details, unsourced numbers, dated news or advice the class has since changed are left out. That's 27 stretches in all. Where the recording and the steps above differ, the steps are current.

Read the transcript, 12 chapters

0:00 Show AI the picture instead of typing

All right. So, what we're going to go over today is [clears throat] we are going to kind of change gears a little bit and we are going to talk about different mediums that you can use to interact with AI. So, when I say the word prompt, most people think I have either typed or talked into the AI and given it verbal instructions on what to do, right? That is not always the case as sometimes it is so much easier to just send an image into the AI than to describe what's going on within the image. Right?

So like one of the things that we use it for with our mobile home parks is uh tenant violations. So, we'll have our lease as kind of one piece of context and then you take a picture and you say to the AI, identify everything that would be a violation based on this lease in this picture and it'll go through and you can even say circle it, highlight it, do whatever, right? And so it goes it a much simpler task to just upload the images and your context and say, "Hey, give me all the violations." And then you look at the image, you say, "Hey, you missed X, Y, and Z. add that to your context and give me this again, right? And so it simplifies it from being like, hey, I'm going to give you

this picture that shows this lot and this and blah blah blah blah blah and writing, you know, a 40-line prompt. It condenses it to analyze this image and then give me the required feedback back. on that video. The longer the video, the more difficult it is to actually get useful inputs out of it. But the other nice thing with Google is they own YouTube. So, if you ever have a long video that you want to get analysis on, upload it to YouTube and then take it to Gemini because Google owns YouTube and Gemini. They talk to each other very nicely.

5:21 Long videos and video formats

Um,

so how would I do that? You're saying I just download it to my YouTube channel.

Yeah. So you'd go to your YouTube, you'd push it up to there, and then you'd go grab that link and you'd take that to and then it uses I think like a package called ffmpeg and it goes and it has transcripts. It can actually pull snippets from the data. Um like what credit.

Do vertical videos get more views than the horizontal ones?

Depends on the format. So, if you're posting on like Insta or TikTok or YouTube Shorts, then you want a vertical uh for the formatting. If you're posting on like long form YouTube, then uh horizontal is fine. Yep.

15:22 Listing photos to a draft you fix

And yeah, so you can really speed up time by understanding when you should be using images versus when you should be using text, right? And so if there's a situation where you're like, I'm about to spend 5 minutes describing this situation, if you can get a picture of it, you can put that in so much more easily. And that's like super useful on like listing descriptions and stuff as well. You can grab pictures of the listing and throw it in there and say, "Hey, write a sample listing description for this property and it might write things and it'll say like the countertops look like they're granite."

You know for a fact they're granite. Go in and make that change. Say, "Hey, the the countertops are granite." So like after you run that initial image, it's going to come back with a listing description that may or may not be accurate. Once you read through it, you make all those changes like I know there's granite countertops. uh the square footage is actually 1,200 square feet versus 1,000. There's and you just make all the little corrections that you know that it doesn't send that back and then you get a full-on listing from scratch, right? And so using images and knowing when to use images is going to be one of those it's a skill that allows you to move a lot faster if you know when to use it.

Some people try to use it all the time and it slows them down. They're like, "This sucks. I don't want to do this." Which I understand. But if you can understand like, "Hey, I'm about to spend a bunch of words describing something that a picture would much more easily resolve." I tend to always go for the picture. Um,

17:04 You answer for what's in the picture

and so like this is kind of what we were talking about where AI can tell you what's going on in a picture. It can tell you the details. It can tell you what it thinks about a picture, but you are the person who's ultimately responsible for the reality of that picture. So, if it comes back with some stuff that's way off base, it's not the AI's fault if that goes out to a client, right? Like, it's still your responsibility to manage that and make sure that it's accurate before it goes out to a client. Um, and so it's super duper important that you're quality controlling the AI outputs. It's it's always what I tell people is that there should be a human in the loop before anything irreversible happens. So if I'm about to send an email to a

client, for me personally, and not everyone's like this, for me personally, I want to read that email before it goes out because once that email's out, I either have to do damage control or some other type of resolution. when a five minute read of that email could have resolved that whole problem, right? So, I like to have my emails drafted by AI but not sent. Not yet anyway. I'm sure

19:06 Crop out passwords and client details

so there are some things that you want to make sure cuz again with images it's super easy to snap a picture, send the image. The things you don't want to include or you want to make sure are not in the picture are like if you have any passwords, if you have um client personally identifying information, like all that kind of stuff. You want to make sure cuz it's super easy to send that with the image in bulk and have it just as a part of the image and now it's in the AI. if it's something that you don't want in the AI, crop the picture and just give it a little bit of extra written context. Um, so that is something that we want to make sure that we're not doing.

19:51 Screenshots for tech problems

One thing that I really love to use the image method for is troubleshooting any of my tech problems. So, if I have a warning screen that pops up and I'm like, I have no idea what a 507 error even is, I will take a screenshot of it, push that into my AI and say, hey, my computer is running into this error. What is it? How do I fix it? And it'll actually tell me like, oh, that's a server side error on Netlify's account. There's nothing you can do about it. You just have to wait till service gets restored.

Because it's like, hey, the the thing that is the thing that you would never suspect to have a problem is having a problem. That's why your your app's not building. Okay, great. So, I just need to wait for that thing to not have a problem or I need to try to figure out their solution on my own. One of the two, right? And so like that's one of that I couldn't actually fix. And so there's a lot of value in that in being like, okay, I understand that I there's nothing really I can do. I'm going to move on to my next task and then come back to this when it's actually functioning again. Um and so like

21:09 Give AI something real to start from

you know so you know how we were creating um context cards for brand, right?

Yep.

So if I take that and put it into Gemini because obviously let's I'm working with Claude, right? I have that and then I put that in there so that way Gemini knows like my look and feel, my brand, and then they can create videos based on that. Recommend that.

Yeah. Yep.

In that way.

Yeah. And so like I would also I like giving the AI source media especially if you're going to create a video um because the AI does a really good job manipulating video around people. And so like if you have a video where you're doing like talking or like hey I'm in outside of X Y and Z address you know da da da we've got uh x amount of show blah blah blah whatever you're doing you can then take that put it into Gemini and say hey add some AI effects do x y and z and I want to end this video by flying away on a dragon right and so it'll be like the whole thing and it'll look normal da da da da and then all of a sudden you swoop out with a dragon right

and so there's really cool things that you can do right now with it if it's being used as a supplement to something you already have. It's harder to create from scratch with the AI because there it has nothing to base off of. But if you give it a starting point, and that's one of the reasons that pictures are really useful, you can say like, "Hey, this is my starting point." Um, like as an example, one of the things I did for an old client was we had their house getting crushed by SpaghettiOs or something and the hook for the video was like, will insurance cover this, right? Like find out the if your insurance will cover a SpaghettiOs like overflow or whatever it is, right?

And so you can have like real things that then get overtaken by like crazy things. And that's where you get videos where they're not necessarily considered AI slop. They're considered a more useful tool because it's like, oh, that's real. Oh, there's the AI. That's more like a cinematic thing than I created this specifically to try to sell something.

But yeah, that's 100% like that's one of the best use cases is like putting yourself in things that look like they have value so that you can throw them on socials or throw them to like another great thing to do with that kind of stuff is to then send that social post to people who it would be impactful for. Like if you're talking about like interest rates and you have a bunch of lender relationships, send that video to the lenders and be like here's how you know this video can help blah blah blah blah blah. If you want to share it and try to get organic shares of your stuff

27:00 Repair questions and staging for one client

And and yeah, so it's the the the ability to use pictures to like get a jump start in projects is also super huge. Um, like if you're writing a contract for a house and you're like, I don't even know what's wrong with this house. And you just go take a bunch of pictures. You can say like, hey, I'm doing this for a buyer. This house has a bunch of stuff wrong with it. Give me any ideas based on these pictures of repairs that we might have to negotiate for, all that kind of stuff.

I mean, you didn't do that part of it, but you just did a video. Oh yeah.

Yep.

Yeah.

Yeah. And that's um like we kind of talked about with the context cards, like you can have it uh stage the room in their specific style, right? So a lot of people are doing AI staging right now, but they're staging it just in general general staging stuff, right? Like just like you would hire a company to do. They're taking the picture and it's general staging. But if you know that one client likes antique furniture and one client likes modern white furniture, whatever, you can take the picture for the client with antique furniture load if you have a context card for them or basically so context card is a description of things that are that. You upload the image and you say, "Stage this room for this client." It'll stage it with furniture that matches their style. And so if you trade out

client cards and you do the same thing, it'll stage it again with furniture that matches their style this time, right? And so you're connecting their vision to your asset, not just saying, "Hey, this is what it would look like if if some random person put furniture in here." You're able to say, "Hey, this is what would happen if you put furniture in here." Right? And so I kind of go back to like AI does one of two things. And I don't think there's an in between for real estate is it either absolutely kills human connection, which for some people isn't the worst thing in the world, or it can absolutely bolster human connection, right? And so the way you bolster it is by using it to make people feel like they're everyone's special little guy, right? Like, hey,

trying to get people to connect with kind of the intangibles of what you're doing because that's like the information isn't the hard thing anymore, right? Like anyone can go to ChatGPT and whether they get good information or bad information, they're going to get information one way or the other, right? And so one having them having people understand when it is bad information and two understanding that that's going to happen um and just trying to get ahead of it as much as you can. But so yeah, so screenshots into AI for solving tech problems is one of my favorite things of all time. Super duper simple. Um, and it turns everybody into their own IT person because the AI can explain it to you at whatever level you need it to be explained at. Um, and

massive. Um, and so yeah, so like if you're having a problem like on your computer and you're like this is so annoying, take a screenshot of it and put it into AI. that that's like the fir anytime like I do this now I'm like okay that's a sign that I should be shoving something into AI right because like that if you know how to get to that problem solving loop with AI this is obsolete because you just skip that step and go right to solving the actual issue

31:27 Photos of documents you don't understand

Um, document questions are a really good way to like this is a really good way to get things into AI if you have questions on like a large document. So, if you don't have a digital version of the document, like if you don't have that 100-page PDF that you can shove in here, it's super easy to grab a couple pictures of the things that you have questions about. And ideally, you'd take the pictures of the whole document, feed it in, and whatever. But if it's like, "Oh man, I've got this long document. I don't really understand this uh paragraph that's been written in legalese. Like, I just need to get through this." You can snap a picture of it, upload it, and you're kind of off to the races as far as where you need to go

with understanding, right? Because again, it can break down that complex contract in terms that a fifth grader could understand and you can go back and then start engaging with it at a much deeper level of understanding. Um, and that's just from the picture. So, super duper simple as a way to kind of shortcut typing or pasting a bunch of text. Um, so this is some of the stuff you can do. So you can say like quote the sections that matter, give me page numbers. So if you're like, what are the things that I need to talk to my client about with this document?

I'm like not really sure. I don't want to go to an expert straight away. I'm going to go to the expert, but you can ask it, give me excerpts of the most important things I need to do and tell me exactly where they are in the document or where they are in the So, this can um again impact listing photos. So, if you put a listing photo in, you can say, "Hey, what are all the things that my buyer would want to negotiate about this?" Or you can put it in as a pre uh check and say, "Hey, I'm working on this listing for a seller.

What are the top 10 things that a buyer is going to come at us for?" Whatever it is. Or, like I said, you can use it to stage for personally. You can use it to um you can actually have it add labels. So, if you're like, "Oh, the buyer probably didn't see this. I want to make sure it's very clear." You can say, you know, circle this and put a label beside this crack in the frame or whatever it is, right? And so, images, like I said, can get a lot of unique use cases that would take you a bunch of time to type, and you may be less accurate than just throwing the image in there.

Uh, so that's virtual staging.

34:35 Point Gemini Live around the room

One of the cool things with Gemini is you can actually do live chats with it on the mobile app. So you can open the mobile app, you can uh press the start button and you can actually go around the room and be like, "Hey, what's this? Hey, what's this? Tell me about this." And it will just work like in live fashion. So, it's pretty neat that you can have like a lit literal visual video assistant that's seeing what you see and you can just kind of talk to it about what you're seeing.

One of the easiest ways I recommend to test this is to just like put some listing photos into the AI and just work with it until you get a really cool graphic out of it. Um, something that you're happy with. And when I say cool, I mean that's so dependent on each person in this room, what they think is cool, like what the um criteria is for that. And that's one of the things that is useful in the age of AI is maintaining your personal identity of what is cool and not just uploading the pictures and saying, "Hey, make this."

Because hey, make this, you'll end up with things that look like everybody else if you don't have a context card. if you're not like aggressively prompting.

35:55 Talk through what you want

And then one of my favorite uh hacks for AI is after you get the picture in there, using voice to talk about everything you want. So the thing that I've been doing recently with AI is I haven't really been prompting it. I've been pressing record and just letting it brain dump. So, I will just press the button and then I will just start talking about all the things in my brain that I need to either change or adjust or whatever about this. And so, I I've had ones that are like, "Oh, that's too long. I'm going to need to like delete that." Or like sometimes it creates really really cool things. And so again, I highly recommend that once the image is in there, using voice because your brain doesn't catch up to your uh voice

like it has to to your fingers often times. Um, and so super easy exercise is to put things like put a couple listing pictures in there, talk about an asset that you would like it to create. So maybe you're like, "Hey, I want an Instagram carousel of these uh images." Start with a blank empty room and then stage one with antique furniture, stage one with modern furniture, stage one it is, mid-century modern, uh whatever buzzwords are happy at the time, right? And then you can create, okay, here's what your room would look like like this, like this, like this, like this, like this. And you can post that on your Instagram as, you know, hey, look at how we can reimagine design or whatever it is.

Um, and then the basic thing with the images is add it to your problem solving loop. So if you're like, okay, I have X, Y, and Z problems. Can images being inserted into the AI chat add value to what you're doing or is it superfluous? Like, can I add this image and it's going to give the AI more than I could if I was just writing text to it? And that's always kind of the the filter. There's actually text. One of my favorite things to do off of like YouTube videos and stuff is to like screenshot slides that I'm like, "Ooh, I want the information off of that slide." There's no way to get the information off of that slide. How do I do that? Take a screenshot, push it into AI, and say, "Transcribe this." And now

I have it as text. That is much more workable in the AI. Like video is difficult to get things out of or it can be um text is much easier. So if you want to manipulate the data, try to get it into text and then you much more easy to do something with it. So, the easiest thing to do with this is just to start on your phone, like try out a couple uploading a couple pictures and just kind of talking through seeing what outputs you get. Um, and trying it in different scenarios, like trying And that's it.

But yeah, images as a medium are underutilized because most people don't even think to throw images in as part of their prompt, right? And it it's very useful context especially in real estate because so much of what you guys are interacting with is what the buyer sees, right? And so being able to create visuals that make them like, oh wow, this is exactly what I want. If you can create a visual that gets them to be like, that's exactly what I want, it goes a long way in my experience as far as building that human connection and them being like, wow, you know me so well. Even if it was made with AI, it's because you know them so well. You executed so well that they feel like, "Oh, wow. This is like personal to me.

This isn't for anybody else." That it says, "Hey, Jim." And it's got my furniture styling in the the room. Like that's like a big deal to people, especially people who aren't super AI savvy because they're like, "Wow, this took you so long." Right? And so just I'm like I said, I'm big into like having context cards built out and then being able to use those to create as many assets as you possibly can and making people feel like, "Oh, wow. This is just for me." Um, but yeah.

So, I have a list that I don't have video on.

Mhm.

I'm probably going to upload all the photos that I have with it. Yep.

And then I have to create a video.

Yeah. 100%.

42:08 Keep listing video true to the photos

Yeah. And you can um there there's some really cool things that you can do with images where there's a 3D world builder where if you upload the images and a blueprint of the home, it'll create a landing page that you can literally scroll down and it'll be like a walking tour through the home. And so like you can do a lot of really that's one of my favorite use cases is to upload listing photos and then have it create like a a micro virtual tour of the house and it'll like walk through and um the so the videos with the home make sure that you tell it to use exactly the images and not to add stuff cuz that's one of the things that the AI video gen does like to do is like oh

let's throw in a potted plant here. We'll put an extra door there. We'll do X, Y, and Z. So, just with that, you'll want to rein it in to be like, use exactly the photos. Don't edit, don't add. You are just making a video walkthrough from my photos. And it might take you a time or two to really like dial it in. Um, because that's that's probably what AI video struggles with the most is like 100% like accurate um copy from things. And so just like really try to guard rail it in by telling it like over and over, you know, this is what I want. Do what I want, not what you think is best. Um, but yeah, like that's one of my favorite use cases for it is to turn pictures into video,

especially for like socials. Um, if you've got like a web page that it it's just it's almost always more engaging to people.

Next on the table.

The questions that tend to come up once you've run one picture through the loop.

All free classes

Where the facts come from.

The 7 sources behind this page

Beyond the screenshot

Show AI the right problem, not just the right picture.

This class teaches AI to separate what it sees from what it can't confirm. The Agent Constraint Accelerator does that for your business: one named constraint, a written plan built around it, and weekly action items for you and for Ryan Wanner.