Choose whether to allow Google Analytics and Meta advertising cookies. Reject optional keeps these trackers off. Website hosting and features you use may still process operational information. See our privacy policy for details. Privacy policy
Find the missing fact. Ask the better question. Give buyers clearer decisions without outsourcing your judgment. A practical reference for evidence review, client decision support and the work between the showing and the closing.
John Kirkwood · Course resources · Edition 3.1 · Updated November 16, 2026
A working evidence room, not another blank chat
Start with a question a client actually needs answered. Gather authorized sources, identify what conflicts or remains unknown, and turn the findings into questions for the right professional. Keep the original evidence beside every summary. Use this course resource to revisit the workflows, practice with fictional examples and check your work before using it with a client.
This resource uses fictional examples. It is general education, not legal advice. Your broker, MLS and applicable law govern real use.
Your learning path
Revisit these four areas as you apply the course to your own practice. Start with fictional data and an approved tool.
Grounded drafting: assemble an authorized source packet and revise an AI listing draft.
Client communication: turn showing notes into a useful follow-up without inventing promises.
Risk and review: identify privacy exposures, misleading imagery and fair-housing concerns.
A repeatable practice: build a review checklist and evaluate one small workflow.
Suggested activities: compare a good and bad property description, repair a misleading answer, review a fictional client-data scenario, and explain the human approval step. Use the prompts and checklists below to repeat the exercises with a new task.
Choose a job, then a tool
A small, useful toolkit
My recommended starting point is one approved drafting assistant plus a source-grounded research tool. Add a specialist tool only when it solves a recurring problem. The examples below are recommendations, not endorsements or guarantees. Check current access, plan terms, data controls and brokerage approval before uploading anything.
Market communication
RPR Market Trends AI ScriptWriter
Use it for: a local-market explanation grounded in the selected RPR market data, then adapt the reviewed draft for a short video or seller update.
First task: select the correct geography and period. Generate a script. Match every number to the underlying report.
Access depends on RPR eligibility. Market commentary is not a valuation, appraisal or promise about a specific listing.
Use it for: comparing approved buyer guides and public documents with citations you can open. Google’s current product page calls it Gemini Notebook, formerly NotebookLM.
First task: load two non-sensitive, authorized guides. Ask what each says about the same question and where they differ.
A citation proves where an answer points, not that its interpretation is right. Check document dates, completeness and sharing before reuse.
Use it for: a consistent drafting workspace with approved tone guidance, checklists and fictional examples rather than starting every prompt from scratch.
First task: add your public writing samples and an “unknown facts stay unknown” rule. Ask for a draft and a fact-check table.
Project context is not an access-control or confidentiality guarantee. Review plan settings, memory, connected sources and permissions.
Capabilities checked November 16, 2026. No prices, unlimited-use promises or assumptions about your subscription. An approved equivalent tool is fine.
Beyond a blank chat box
Five workflows worth practicing
The advantage comes from the process: better inputs, visible evidence and a clear approval step. These are original workflow designs, not claims that a product automatically does everything listed.
01 · A transaction evidence room that exposes missing facts
Build the packet: use authorized, de-identified documents with title, date, version and source. Do not upload private transaction files to an unapproved account.
Compare: request a table of question, exact source passage, source date, conflicting statement and next verification step.
Challenge: ask what evidence would change the conclusion. Do not let the assistant decide which legal document controls.
Hand off: route the unresolved issue to the broker, inspector, HOA or other appropriate professional, with the original passages attached for human review.
Fictional filled example: Seller sheet says “roof replaced in 2018.” Inspection note says “age not verified.” Output: “Replacement year remains unverified. Request the permit or invoice and ask the inspector what the inspection established.” This is an uncertainty to resolve, not proof of fraud or a defect.
Decision rehearsal: ask for questions a buyer should resolve before proceeding, using only the packet. Review any contingency or deadline with the broker rather than allowing AI to calculate contractual rights.
02 · A listing launch kit from one verified fact sheet
Gather: authorized facts, approved photos, source dates and a list of unknowns.
Draft: ask for an 80-word listing description, a 30-second video script and five open-house FAQ answers. Every output must use the same facts.
Audit: request a table of claim, supporting source and uncertainty. Personally verify it.
Approve: review advertising language, brokerage attribution, MLS rules and photo disclosures before using anything publicly.
Useful edge: a single correction to a roof age or feature can flow through the whole draft kit, avoiding contradictory marketing. Keep a version and date on the source sheet.
03 · A buyer question pack with an uncertainty register
Use approved buyer education documents and permitted public sources. Keep private client documents out of the demo.
Ask for “what the source says,” “what it does not establish” and “who can answer.”
Create a short question list for the inspector, lender or broker instead of allowing AI to resolve specialist issues.
Verify citations and hand the buyer the reviewed questions alongside the original sources.
Example: “The report recommends further evaluation. Ask the inspector what access was limited and which specialist can evaluate it.” Never convert that into “the roof is fine” or a repair-cost guarantee.
04 · A local-market briefing that shows its working
Choose one geography, property type and reporting period in an authorized data source.
Draft a plain-language explanation with the exact metric definitions and dates.
Recalculate comparisons and check sample size, missing data and whether periods are comparable.
Turn the reviewed explanation into an email draft or short video, with the source and date visible.
Useful edge: answer the seller’s actual question, such as what changed in days on market, rather than producing generic “hot market” commentary. Never let a generated forecast become a promise.
05 · An internal follow-up queue with human approval
With required consent and brokerage-approved systems, capture only the notes needed for the client’s next step.
Extract questions, requested actions and proposed owners into a draft task list.
Have the responsible agent verify deadlines, consent, ownership and the exact message.
Only then send through the approved system. Keep a sent record and reconcile it before retrying.
Useful edge: fewer forgotten questions without an autonomous assistant making commitments. Begin with draft-only operation and test duplicates, missing phone numbers and cancelled requests using fictional contacts.
Trust is part of the workflow
Security and ethics that change what you do
Before you connect an account
Inspect what the integration can read and change. Prefer the narrowest access. Do not connect your whole mailbox to save five minutes on a listing draft. Establish who can revoke access and who reviews external actions.
Before you upload a document
Confirm you have the right to use it. Remove unnecessary identifiers, private financial details, signatures and access codes. Check retention, sharing, training settings and deletion options for the actual plan. A “business” label alone is not your policy.
Before you trust an instruction
Documents and emails can contain malicious directions to an AI. Treat their contents as data, never as authority to reveal information, send a message or change a record. Keep approval separate from document analysis.
Wire changes require independent verification
A familiar writing style, voice or face does not establish identity. A request to change payment details is a stop-and-verify event.
Use a trusted number already on file to confirm with the appropriate party. Do not use the phone number supplied inside the suspicious message. If money moved, contact the bank immediately and follow your incident process.
Use unique passwords, an approved password manager and MFA. Prefer phishing-resistant methods where supported. Prompt changes cannot repair a compromised email account.
Fair access: no protected-class preference, discriminatory ad targeting, demographic steering or automated customer selection based on sensitive traits.
Truthful representation: no hidden defects, invented views, fabricated amenities, fake reviews or testimonials.
Professional boundaries: no AI decisions about legal rights, financing eligibility or technical property condition.
Respect for people: obtain required permission for recording and reuse. Do not clone a client’s voice or impersonate a colleague.
Review the fair-housing and image sections below with your broker. Specific legal requirements can depend on the transaction, location and medium.
Useful jobs for an AI assistant
Listing copy from a verified fact sheet
Showing follow-ups from neutral notes
Buyer FAQs from approved resources
A content outline from your own expertise
Choose one recurring drafting task. AI can produce confident but invented facts. You retain review and approval of every draft.
The source packet
Facts you can support
The source and its date
What remains unknown
What the draft must leave out
A prompt cannot verify its own inputs. Gather authorized listing facts, brokerage rules and intended format. Check MLS data permissions before uploading. Use fictional material for practice.
Fictional listing exercise
Mesa home · 3 bedrooms · 2 bathrooms
1,650 sq. ft. per supplied source
Covered patio · two-car garage
Roof age, school assignment and HOA unknown
These invented training facts describe no actual listing. Omit roof age, upgrades, schools, HOA terms, travel times and investment returns unless you have verified sources. List unanswered questions separately.
A prompt that sets boundaries
Use only the facts provided.
Draft 80 words about property features.
Do not infer buyer identity or add amenities.
List missing facts separately.
Try the full prompt below. After generating a draft, point every claim back to the fact sheet. An assistant’s confidence does not make a draft ready for the MLS.
The human review
Every property claim has a source
No invented amenities or guarantees
No preferred buyer or demographic cues
Brokerage and MLS rules checked
Practice reviewing this flawed draft: “Ideal for young families in a safe neighborhood with top schools and a new roof.” The source packet supports none of these claims. Revise around the known property features and remove preferred-buyer language.
Fair housing in AI-assisted work
Describe the property and its features.
Avoid protected-class preferences.
Use buyer-stated, objective search criteria.
Review ad targeting as well as ad copy.
Federal fair housing protections include race, color, religion, sex, national origin, familial status and disability. State and local protections may add requirements. Avoid discriminatory preferences, demographic steering and protected-trait customer ranking. Ask your broker or counsel about actual practices.
A follow-up that helps the buyer
Recap the features they discussed
Separate questions from known facts
Offer one clear next step
Let the agent approve the message
Fictional follow-up: “Thanks for touring. You mentioned the covered patio. I will verify the roof age and request the HOA documents before we discuss next steps.” Review commitments before sending and never invent a deadline.
Client information needs boundaries
Use fictional or de-identified examples.
Choose brokerage-approved tools.
Check retention, training and sharing settings.
Keep access limited to the task.
Removing a name may leave a person identifiable through an address or transaction details. Keep IDs, financial records, access codes and private negotiations out of unapproved assistants. Check data use, retention, access, consent and brokerage policy.
Images and market claims
Virtual staging needs clear disclosure.
Property condition must stay truthful.
Market claims need dated source data.
A generated citation still needs opening.
Keep original photos. Check current state and MLS rules before using virtual staging. Disclosure does not excuse an image that misrepresents defects, views or permanent features. Verify market figures against dated source data.
Buyer education without overpromising
Turn approved FAQs into plain language.
Keep exclusions and uncertainty visible.
Send contract questions to your broker.
Send financing questions to the lender.
Use AI to prepare questions for an expert. Contracts, financing, valuations and inspection conclusions need the appropriate qualified professional. Keep limitations in summaries and never turn them into guarantees.
A useful answer on your website
One real buyer question per page
A clear answer from your own expertise
Sources, service area and review date
An honest path to speak with you
Write useful answers to real buyer questions with supporting evidence. Avoid invented reviews, keyword stuffing and promises of AI visibility. No one can guarantee an AI system will recommend your agency.
A one-week pilot
Choose one drafting task.
Record time before and after review.
Count factual edits and privacy concerns.
Expand only when the results justify it.
Try five fictional or approved examples. Measure total time including verification, count factual corrections and record privacy concerns. Stop or redesign if errors outweigh the benefit.
Your next draft
What facts will you supply?
What must the AI leave out?
Who approves the final version?
What would make you stop?
Write one draft prompt and its approval checklist. State what the AI must omit, which sources you will use, who approves the output and what would make you stop.
Prompt library
Prompts constrain a draft. They cannot guarantee accuracy or legal compliance. Read the output against the source packet.
Evidence room: contradictions and missing facts
Start with the fictional packet. Open every cited passage yourself before relying on the result.
Listing draft
Showing follow-up
FAQ with evidence
Before anything leaves your desk
Each factual claim matches an authorized, dated source.
Unknowns remain unknown. No invented features, promises or deadlines.
Copy focuses on the property, without discriminatory preferences or steering.
Client information, document permissions and tool settings pass brokerage policy.
Images accurately represent the property, with required disclosure and originals.
A responsible person approves the final communication and its recipients.
Practice answer
Fictional example: “This Mesa home offers three bedrooms, two bathrooms and 1,650 square feet, according to the supplied property information. A covered patio provides outdoor space, and a two-car garage adds parking.”
Missing facts: roof age, school assignment and HOA details. Ask for verification rather than guessing.
One-week scorecard
Task: ____ · Original time: ____ · AI plus review time: ____ · Factual corrections: ____ · Privacy concerns: ____ · Reviewer: ____ · Continue, revise or stop: ____.
Sources and updates
Official sources reviewed November 16, 2026. Legal requirements and platform capabilities change. Verify current rules before using this in a transaction or presenting it again.
The workflow examples and exercises are original instructional recommendations. They do not promise results or imply endorsement by these organizations.
Maintenance record
v3.1 · November 16, 2026: Organized five workflow playbooks, reusable prompts and security guidance as a course resource. Added navigation back to Stirling training. Earlier resource versions are preserved. Review before the next event and whenever brokerage rules or tool privacy controls change.
Owner: John Kirkwood / Stirling AI. Keep source links, review dates and changes together. Replace fictional examples only with authorized, de-identified material.
Your next working session
Choose one playbook. Use fictional data first. Complete this checklist before moving to a real client file. These boxes are session-only and do not store data.
Return before your next project to recheck source dates and tool settings. Check the maintenance record for published revisions; updates are reviewed before publication.