Multifamily NEXT · AI Tuesdays
Week 18 · Persona
Self-guided · ~25 min · Phase 3
Warmup
The Workshop
Power Preview
Resources & Tools
Take Action
Multifamily NEXT · AI Tuesdays
tami@mprodigitaledge.com
Password: 1852
tami@mprodigitaledge.com
Password: 1852
Week 18 of 52 · Self-Guided · ~25 min
0 starred
0 notes
Pre-Show · Warmup
Welcome back. Let us warm up first.
Three quick recall questions from last Tuesday's workshop on Guardrails. Tap each to expand, tap again to close. Just for you, no scoring.
Answer: Be specific (name the behavior you want stopped). Pair positive with negative (do not X. Say Y instead). Keep the list short (three per prompt is the ceiling). Layer over time (start with one or two, add when real bad output shows the gap). All four together separate a working guardrail from a weak one.
Example of use
Renewal outreach prompt with three guardrails: (1) do not include statistics you cannot cite, say "let me confirm" instead; (2) do not use hedge words, state what you know exactly; (3) do not offer any concession, say "let me confirm what the offer includes" instead. Three specific, positive-paired, short, layered.
Answer: They do not sanitize what you paste in. Guardrails shape AI's output. If you paste an unbracketed resident name, apartment number, or balance, the guardrail does nothing about that. Real resident data still stays out. Bracket first, then guardrail. Two different jobs.
Example of use
Right: "[Resident] in [Apartment] is [X days] delinquent on $[Balance]. Guardrail: do not use language that suggests eviction." Wrong: "Sarah Martinez in 214 is 60 days past due on $2,847. Guardrail: do not use language that suggests eviction." The guardrail is identical. The second one leaked real data anyway.
Answer: Three is the ceiling per prompt. Past three, AI starts hedging everything to avoid tripping any one guardrail, and the output goes to mush. Pick the three that matter most for this task. Save the rest for next time. Layered over months beats stacked in one prompt.
Example of use
Renewal prompt with three tight guardrails (no invented stats, no hedge words, no unauthorized concessions) reads warm and specific. The same prompt with nine guardrails (add: no imperatives, no first-person, no exclamation points, no adjectives about the property, no market comparisons, no timeline commitments) reads like AI is walking on eggshells and forgets the actual message.
The Workshop · Opening
Ten prompts. One voice. Every time.
Persona is the fifth prompt structure of Phase 3. Not a Role in one line. A sustained character AI holds across every message in a session. Same experienced dispatcher answering ten vendor emails. Same warm leasing pro replying to ten resident follow-ups. Same compliance-first regional manager reviewing ten renewal drafts. Not ten different helpers with different voices, guessing at your tone each time.
This is the structure you reach for when the work is not one prompt and done. It is a session. The persona is how you keep the voice consistent across the whole thing.
Where Persona sits
Role (Week 14) = one line, one task. Persona (Week 18) = 4 to 8 lines, whole session. Same idea, longer arc. If a Role is the hat AI puts on for one answer, a Persona is the person AI becomes for the whole conversation.
Twenty-two slides. Twenty-five minutes if you read straight through. No timer runs.
The Workshop · Agenda
Four parts, twenty-five minutes.
1
Persona vs Role
Slides 4 and 5. Why Role in one line is not the same as sustained character across a session.
~5 min
2
The four things a Persona specifies
Slide 6. Voice, vocabulary, decision-making style, refusals. What goes in and what stays out.
~4 min
3
When to reach, when to skip, drift, and the bright line
Slides 7 through 10. Where Persona earns its keep, where it slows you down, what drift looks like, what Persona is not.
~5 min
4
Q&A, Power Preview, references, quiz, close
Slides 11 through 22. Ten dispatcher emails in one persona, references, five-question quiz, Week 19 teaser.
~11 min
No timer runs. The minute estimates are honest averages. Slower is fine.
The Workshop · Why You Should Care
Ten emails. Ten different voices. Same person supposedly writing.
Two things happen every week that a Persona prevents. Neither one looks like a mistake in the moment. Both are avoidable with a persona block at the top of the session.
Scenario 1
The tone that swung
A leasing pro asks AI for a batch of ten follow-up emails to prospects who toured but did not apply. Email one sounds warm and gentle. Email four sounds pushy. Email seven sounds corporate. Email ten sounds like a different property brand entirely. All ten went out with the leasing pro's name at the bottom. The persona was never set, so AI picked a fresh tone every time.
Scenario 2
The compliance persona that overstepped
A regional manager sets AI up as "a compliance-first attorney reviewing our marketing." AI now confidently states what the Fair Housing Act "requires" and what is "legally defensible." None of that is verified. Persona changed the voice. It did not turn AI into a lawyer, and it did not add legal authority AI does not have. This is the bright line failure. Slide 9.
Why this earns its own week
Every prompt structure so far has been one prompt, one answer. RTF, Chain of Thought, Few-Shot, Guardrails. Persona is the first structure that spans a session. Different job, different mechanics. If you set it once and forget it, AI drifts. If you never set it, AI guesses fresh every message. Either way, the recipient reads it and forms an impression of you.
The Workshop · Definition
PersonaPERSONA (PROMPT STRUCTURE)
The sustained character AI holds across every message in a session. Goes beyond a one-line Role (from RTF, Week 14). Four to eight lines that lock voice, vocabulary preferences, decision-making style, and refusals for the whole session, not just the next answer.
Multifamily example: A maintenance dispatcher persona: "You are a maintenance dispatcher with 15 years experience. Direct and practical. Use trade terms like return trip, punch list, callback. Never commit vendors to timelines you have not verified. When a vendor pushes for an exception, say 'let me check with the property manager' instead of agreeing." Seven lines. Runs across every follow-up email in the session. AI sounds like the same person every time. in one line.
A Persona is the sustained character AI holds across every message in a session. Not one line, not one answer. Four to eight lines at the start of the session that specify who AI is being, and how long it holds that character.
Plain-English version
Persona is asking AI to be someone for the whole conversation, not just for one answer. When your work is a session (ten emails, ten drafts, ten follow-ups), the same voice across all of it is what makes AI usable. Persona is the structure that gets you there.
Every prompt structure so far has been one prompt, one answer. Persona is the first one that spans the whole conversation. That is the shift Week 18 teaches.
The Workshop · Persona vs Role
Role (Week 14) is one line. Persona is a session.
The single most common mistake is treating Persona like a longer Role. It is not. Four distinctions separate them. Read them once, then decide which one your task needs.
Dimension
Role (Week 14)
Persona (Week 18)
Length
One line. "You are a friendly on-site property manager."
Four to eight lines. Voice, vocabulary, decision style, refusals.
Scope
Task-level. Shapes one answer.
Session-level. Shapes every answer until the session ends.
Reinforced
Not needed. One prompt, one answer.
Yes. Restate the persona when drift shows up. Usually every five or so messages in a long session.
Survives
The single answer. Then it is gone.
Ten follow-ups, if you reinforce. Longer if you build a saved persona block.
The tell
If you only need one answer, use Role. If the recipient of the tenth email should not be able to tell which AI wrote it versus the first, you need Persona.
The Workshop · Anatomy
Four things every Persona should specify.
A Persona that only says "you are a friendly leasing pro" is a Role in disguise. It will not survive ten messages. Four ingredients make a Persona hold up across a session. Miss any of the four and AI fills the gap by guessing, which is what causes drift.
1
Voice
Formal, warm, direct, playful, corporate. Pick one, name it. "Warm but direct. No corporate speak." If you leave voice unstated, AI defaults to whatever the tool's system prompt suggests, which drifts message to message.
2
Vocabulary preferences
What this character says, and what they never say. "Use trade terms like return trip, punch list, callback. Never say issue, use problem or fix." This is what makes a persona sound like a real role, not a generic helper.
3
Decision-making style
Defer to policy, escalate quickly, weigh cost against risk, verify before committing. Tells AI how the character responds when the answer is not obvious. "Never commit to a timeline without verifying. Say let me check first."
4
Refusals
What this character will not do, and what they say instead. "Never offer discounts or concessions. Say let me confirm what the offer includes." Refusals are Guardrails built into the Persona. Same pattern as Week 17, longer arc.
Where Week 17 shows up
Ingredient four (Refusals) is where the Guardrails structure lives inside a Persona. A Guardrail is one "do not X. Say Y instead" line inside one prompt. A Persona's refusals are the same idea, permanent across the session. If a task needs three Guardrails on the same theme (say, no invented stats, no hedge words, no concessions), those probably belong in a Persona instead.
The Workshop · When to use
Five times to reach for Persona. Four times to skip it.
Persona is not the answer to every prompt. It costs 4 to 8 lines up front, and it only pays off across a session. If the task is one prompt and done, skip it and use RTF. Here is where Persona earns its keep, and where it slows you down.
When to reach for Persona
1
Batch drafting across many recipients
Ten renewal emails, ten prospect follow-ups, ten delinquency notes. Same voice on all of them, so the property brand does not swing message to message.
2
A long working session on one project
Reviewing every renewal draft this month. Rewriting your resident newsletter section by section. AI is your same reviewer across the whole document, not five different editors.
3
Handoffs where the character matters
Vendor management, escalations, compliance reviews. The character in the persona is what makes AI say "let me check first" instead of committing on your behalf.
4
Coaching or role-play scenarios
Practicing a difficult resident conversation. Rehearsing a rent-increase call. AI holds the resident persona (or the difficult one) for the whole practice, not just the first line.
5
Building a saved assistant for a role
In an enterprise AI workspace, a Persona block becomes the definition of a saved assistant. The dispatcher assistant. The renewal assistant. The reviewer assistant. Set once, reused every time.
When to skip Persona
1
One prompt, one answer
Rewrite this sentence. Summarize this email. Convert this list to a table. Role in RTF is faster. Persona is overhead for a job that ends after one answer.
2
Factual lookups
You want the answer, not a character. Persona does not change what AI knows. See Slide 9. Ask the question directly.
3
Brainstorming
A Persona narrows AI's range. That is the opposite of what brainstorming needs. Drop the character and ask AI to think broadly.
4
Anything a compliance-first persona might make you overtrust
Setting AI as "a Fair Housing attorney" or "a compliance officer" makes AI sound authoritative. It does not make AI right. Skip the character, cite the source, or ask a real reviewer. Slide 9.
The Workshop · Build one
Building a Persona: the maintenance dispatcher.
Four ingredients from Slide 6, one worked example. This is the same persona used in the Power Preview on Slide 14. Read it once. Notice which lines specify voice, which specify vocabulary, which specify decision style, which specify refusals. All four have to be there.
The persona block (paste at the start of the session)
Persona for this session:
You are a maintenance dispatcher with 15 years of experience. Direct and practical, never fussy. You have coordinated with plumbers, HVAC techs, electricians, and painters your whole career, and you know how they talk.
Voice: warm but businesslike. Complete sentences. No exclamation points. Never corporate-speak.
Vocabulary: use trade terms like return trip, punch list, callback, time-and-materials, not-to-exceed. Say problem or fix, never say issue. Say vendor or trade, never say partner or resource.
Decision-making style: verify before committing. When a vendor asks for a schedule change, a scope change, or an exception, say "let me check with the property manager and confirm" instead of agreeing. When you do not know a cost or a lead time, say "let me confirm."
Refusals: never authorize work over $500 without written confirmation. Never commit to a same-day service window unless the vendor has confirmed availability. Never share resident personal information beyond apartment number.
Hold this persona for the whole session unless I say otherwise.
You are a maintenance dispatcher with 15 years of experience. Direct and practical, never fussy. You have coordinated with plumbers, HVAC techs, electricians, and painters your whole career, and you know how they talk.
Voice: warm but businesslike. Complete sentences. No exclamation points. Never corporate-speak.
Vocabulary: use trade terms like return trip, punch list, callback, time-and-materials, not-to-exceed. Say problem or fix, never say issue. Say vendor or trade, never say partner or resource.
Decision-making style: verify before committing. When a vendor asks for a schedule change, a scope change, or an exception, say "let me check with the property manager and confirm" instead of agreeing. When you do not know a cost or a lead time, say "let me confirm."
Refusals: never authorize work over $500 without written confirmation. Never commit to a same-day service window unless the vendor has confirmed availability. Never share resident personal information beyond apartment number.
Hold this persona for the whole session unless I say otherwise.
What this buys you
Ten vendor follow-ups later, the tone is the same. The vocabulary is the same. The decision-making style is the same. The refusals are still in place. If AI starts drifting, one line brings it back: "ReinforcementREINFORCEMENT (IN PERSONA)
A short mid-session message that restates the persona to lock character back in when drift starts. Usually one or two lines. Cheaper than restarting the session. Every long-session persona needs a plan for reinforcement: you decide up front when to send it (every five or so messages, or when you notice drift).
Multifamily example: Long renewal-outreach session, seven emails deep. AI has started dropping the "let me confirm" language you specified in the persona and is committing to answers instead. Send: "Reminder: staying in the persona from the start of this session. When you do not have data, say let me confirm." AI re-anchors and the next email lands right.: staying in the maintenance dispatcher persona from the start of this session."
A short mid-session message that restates the persona to lock character back in when drift starts. Usually one or two lines. Cheaper than restarting the session. Every long-session persona needs a plan for reinforcement: you decide up front when to send it (every five or so messages, or when you notice drift).
Multifamily example: Long renewal-outreach session, seven emails deep. AI has started dropping the "let me confirm" language you specified in the persona and is committing to answers instead. Send: "Reminder: staying in the persona from the start of this session. When you do not have data, say let me confirm." AI re-anchors and the next email lands right.: staying in the maintenance dispatcher persona from the start of this session."
The Workshop · The bright line
What Persona does not do.
The Week 18 bright line
Persona changes how AI writes. It does not change what AI knows. A compliance-first persona is not a lawyer.
This is the line that has to hold. Persona shapes voice, vocabulary, decision-making style, and refusals. Every one of those is about how AI writes. None of them adds knowledge AI does not have or authority AI does not have.
Failure 1
Legal authority
Setting AI as "a Fair Housing attorney" makes AI sound like one. It does not make AI one. Any legal claim AI makes has the same reliability with or without the persona. The persona just makes overconfident answers sound more credible, which makes overtrust more likely.
Failure 2
Domain expertise
A "senior HVAC diagnostician" persona will confidently describe repair sequences AI has no way to verify. A "seasoned accountant" persona will confidently state accounting treatments AI has not checked. Persona increases confidence, not accuracy.
Failure 3
Accountability
The persona does not sign the email. You do. Whatever AI wrote in character still leaves your inbox with your name on it. If the persona convinced you to skip your usual review because "the dispatcher persona knows what they are doing," that is on you, not on the persona.
Failure 4
Facts and numbers
A "market-savvy leasing pro" persona still invents market comps AI does not have. A "compliance-focused regional manager" persona still cites Fair Housing language incorrectly. The Week 17 Guardrails still apply. Persona plus Guardrails plus human review, not any one alone.
Persona plus Guardrails plus human review is the full stack. Do not let a convincing persona replace any of the other two.
The Workshop · Default rule
If the work is a session, use a Persona.
The default rule for Week 18
If the work fits in one prompt: use RTF Role. No persona.
If the work is a session (multiple messages, same character): use a Persona with all four ingredients. Voice, vocabulary, decision-making style, refusals.
If the session runs long (past 5 or so messages): plan for Reinforcement. Restate the persona in one line when drift shows up.
Always: Persona plus Guardrails plus human review. Never let the persona replace review, even if the character sounds authoritative.
If the work is a session (multiple messages, same character): use a Persona with all four ingredients. Voice, vocabulary, decision-making style, refusals.
If the session runs long (past 5 or so messages): plan for Reinforcement. Restate the persona in one line when drift shows up.
Always: Persona plus Guardrails plus human review. Never let the persona replace review, even if the character sounds authoritative.
Five prompt structures in your hands now. RTF, Chain of Thought, Few-Shot, Guardrails, Persona. Two more weeks in Phase 3 and you have the full seven. Then role-based application starts in Phase 4.
The Workshop · Q&A
Five questions attendees actually ask.
Tap to expand. Tap again to close. These are the five that come up every time.
Answer: A Role in RTF describes who AI is for one answer. A Persona describes who AI is for the whole session. Length is a symptom of the real difference, which is scope. A one-line Role does not need Voice, Vocabulary, Decision-making style, and Refusals stated separately, because AI only has to hold character for one message. Four to eight lines are what a Persona needs because it has to hold character across ten.
Example of use
Role: "You are a friendly leasing pro. Draft a welcome email." One answer. Done. Persona: "You are a leasing pro with 8 years experience. Warm and direct. Never over-promise on amenities. When you do not know a fee, say let me confirm." Then across the next ten prospect emails, the voice and refusals hold. Same person, ten messages.
Answer: Persona DriftPERSONA DRIFT
When AI slips out of the character you set, usually after 5 to 10 messages in a long session. Voice softens or hardens, vocabulary preferences leak out, decision-making style shifts, refusals get forgotten. Not a failure of AI. It is a limitation of how much of the earlier context still carries weight later in a conversation. The fix is Reinforcement, a short restatement of the persona mid-session.
Multifamily example: Set a compliance-first regional manager persona at message 1. By message 9, AI has started using casual language and stated a market rate without qualifying it. Send: "Reminder: staying in the compliance-first regional manager persona from earlier. When a number is not verified, say let me confirm." Character locks back in and message 10 lands right. is AI slipping out of character across a long session. You catch it by rereading the last message against the persona block. If vocabulary you specified is missing, if refusals you set are being ignored, if the voice is different, that is drift. The fix is a one-line Reinforcement message. Cheaper than restarting the session.
Example of use
Renewal outreach session, seven emails deep. AI has started dropping the "let me confirm" language and committing to answers. You send: "Reminder: staying in the persona from the start of this session. When you do not have data, say let me confirm." AI re-anchors. Email eight lands right.
Answer: You can set the persona. AI will start sounding like an expert. That is not the same as AI being one. This is exactly what the bright line on Slide 9 addresses. A confident-sounding Fair Housing persona will make wrong answers sound more credible, not more accurate. If the answer matters, cite the source (the actual HUD guidance, the actual statute) or route it to a real reviewer. The persona is not a substitute for either.
Example of use
Marketing team wants AI to check every listing against Fair Housing. A "Fair Housing attorney" persona sounds authoritative but AI will confidently state things that are wrong. The right approach: a "marketing reviewer" persona that flags anything mentioning family structure, age, disability, or protected classes for a human review. AI flags. The human decides.
Answer: Save the persona block as a text file or a saved prompt. Paste it at the start of every new session where you need that character. In an enterprise AI workspace like mPro Digital Edge, a Persona block becomes the definition of a saved assistant, so the paste step goes away. In a general AI tool without saved assistants, keep the block in your notes and paste it manually. A well-tuned persona is a reusable asset, not a one-time thing.
Example of use
A maintenance dispatcher persona (Slide 8) saved as one text file. Every Monday when the vendor follow-ups start, paste it as message one. Ten emails later, the voice is the same as last Monday. Consistency across weeks, not just across a session.
Answer: The pattern (Voice, Vocabulary, Decision-making style, Refusals, plus Reinforcement when drift shows) is what transfers most reliably. I am not certain any specific persona holds character identically across every tool on every task, and behavior can change with tool updates. The honest answer: test the persona in the tool your team actually uses, and expect to tune the wording, especially the refusals, per tool. Save the ones that work.
Example of use
Same maintenance dispatcher persona pasted into three different tools. Two hold character cleanly. One drifts by message four. That tool needs Reinforcement every three messages, or needs the refusals tightened. Same persona, different tuning. Find out empirically, once, per tool.
Power Preview · Setup
Ten emails. Same dispatcher. Same day.
Now watch a Persona do the whole job. The scenario is real: a maintenance dispatcher works through ten vendor and resident emails on a Tuesday morning. Run 1 uses Role in RTF only, one line per prompt. Run 2 uses the full maintenance dispatcher Persona from Slide 8, set once, held across all ten.
Watch for
Voice consistency across the ten. Same warmth, same directness, same complete sentences.
Vocabulary consistency. Trade terms (return trip, punch list, callback) used the same way every time.
Decision-making consistency. "Let me confirm" when the answer is not verified. Every time, not just the first email.
Refusals held. No commitments to same-day service without vendor confirmation. Ten emails in, still holding.
Vocabulary consistency. Trade terms (return trip, punch list, callback) used the same way every time.
Decision-making consistency. "Let me confirm" when the answer is not verified. Every time, not just the first email.
Refusals held. No commitments to same-day service without vendor confirmation. Ten emails in, still holding.
The setup is the same both runs: same ten inbound emails, same task (draft the reply), same brackets on real data. Only the persona changes. The output difference is Persona's job.
Power Preview · Run 1
Run 1: Role only. Ten emails, ten voices.
One line of Role at the top of each prompt. No sustained character. Watch what happens over ten messages.
The prompt (same one, ten times)
Role: You are a maintenance dispatcher.
Task: Draft a reply to this vendor email.
Format: Brief, professional.
[inbound email pasted here]
Task: Draft a reply to this vendor email.
Format: Brief, professional.
[inbound email pasted here]
What comes back over ten emails
Email 1
Warm and specific
"Hi Tom, thanks for the update on the punch list. Let me know when you can circle back for the return trip." Reads right. Uses the trade terms.
Email 4
Corporate creep
"Thank you for reaching out regarding the outstanding issue. We appreciate your partnership." Trade vocabulary gone. Warmth gone. Sounds like an autoresponder.
Email 6
Overcommitment
"We can definitely have someone on-site this afternoon. Please confirm the address." No verification. AI committed on the dispatcher's behalf. The refusal was never set because it is not in the Role.
Email 9
Different person entirely
"Hey! Sounds great, let's get it scheduled! Looking forward to it!" Exclamation points. Casual voice. Would fit a marketing email, not a maintenance dispatcher. Ten emails in, AI has forgotten what the character was.
The pattern
Role degrades. Not because AI got worse. Because Role was never designed to hold character past one answer. If ten emails go out signed by the same dispatcher and read like ten different people wrote them, the brand takes the hit, not AI.
Power Preview · Run 2
Run 2: Full Persona. Ten emails, one voice.
Same ten inbound emails. Same task. The Persona block from Slide 8 pasted once at the start of the session. Then just the task on each follow-up.
Message 1 of the session (the Persona block, then the first task)
[Full maintenance dispatcher persona block from Slide 8 pasted here]
Task: Draft a reply to this vendor email.
[first inbound email pasted]
Task: Draft a reply to this vendor email.
[first inbound email pasted]
Messages 2 through 10 (just the task each time)
Task: Draft a reply to this vendor email.
[next inbound email pasted]
[next inbound email pasted]
What comes back across ten
Email 1
Same warmth, same terms
"Hi Tom, thanks for the punch list update. Let me know when you can schedule the return trip." Warm, businesslike, trade vocabulary present.
Email 4
Persona still holds
"Got the callback request. Before we send someone out, let me confirm the parts situation with the property manager and get back to you." Same voice. Same "let me confirm" language from the decision-making style.
Email 6
Refusal held
"I hear you on the timing pressure. Let me check with the property manager on same-day availability and confirm before I commit either of us to a window." The refusal from the persona block (no unverified same-day commitments) held at message six, without repeating the block.
Email 9
One Reinforcement kept it locked
Between email 7 and 8, drift started (voice softened slightly, one exclamation point crept in). One line of Reinforcement ("Staying in the dispatcher persona from the start of this session") locked it back. Email 9 reads like email 1. Same person, same dispatcher, ten emails later.
The pattern
Persona plus one Reinforcement mid-session equals ten consistent emails. The vendor cannot tell email 10 apart from email 1. Trade terms, decision-making style, refusals, all held. This is what a saved dispatcher assistant would do on autopilot in a governed workspace.
The Workshop · Try This Today
Try This Today: build one Persona and use it three times.
Small step, real payoff. Pick one role in your work where you write more than one message a session. Build the Persona block. Use it three times this week. Notice what changes.
Step 1
Pick the role
Whichever one of your recurring sessions has the most drift right now. Vendor follow-ups. Renewal outreach. Delinquency notes. Newsletter drafts. Pick one where the voice matters and the volume is high.
Step 2
Write four to eight lines
Voice (one line). Vocabulary preferences (one or two lines). Decision-making style (one or two lines). Refusals (one or two lines). Model it on Slide 8. Save it as a text file.
Step 3
Use it three times this week
Paste the block as message one of every session where that role applies. Do not change it between sessions. See if the third session sounds like the first.
Step 4
Tune the refusals after session two
If AI committed to something you did not want it to, add that as a refusal. If AI used vocabulary you did not want, add that too. Every tuning round improves the reusable asset.
Where the platform picks up
Inside mPro Digital Edge, a Persona block becomes a saved assistant. The dispatcher assistant. The renewal assistant. Set it once, and every team member on that role has the same voice, the same vocabulary, the same refusals. Consistency across the property, not just across your session. That is Week 18 at scale.
Reminder
Persona plus Guardrails plus human review. Do not let the character in the persona replace review, even if it sounds like it knows what it is doing. See the bright line, Slide 9.
Reference · Bookmark this slide
Vocabulary cheat sheet: 49 core terms.
Every AI term this curriculum has introduced so far, in one place. Plain-language definitions. This is the same term list on the PDF download at the bottom of this slide and on Slide 21.
Foundational (from Weeks 8-10)
Generative AI
AI that creates new content (text, images, audio, code) based on training data patterns.
Large Language Model (LLM)
The kind of AI behind ChatGPT, Claude, Gemini, and Copilot. Trained on vast text and predicts one word (token) at a time.
Token
A small chunk of text an LLM reads and writes. Roughly 0.75 words per token; 1,000 tokens is about 750 words.
Training Data
The vast collection of text an LLM learned patterns from. Not a searchable database, a compressed pattern archive.
Prediction
How AI decides what to say next: the most likely word based on the pattern of what came before, one word at a time.
Hallucination
AI producing a confident, fluent answer that is factually made up. Especially common when it does not know something you did not tell it.
Sycophancy
AI agreeing with what you just said instead of pushing back. Feels helpful, is often wrong.
Prompt anatomy (from Week 11)
Prompt
What you type to AI. Your input. The one part of the container you fully control.
System Prompt
A hidden instruction the AI tool sends the model before your prompt. Shapes tone, role, and constraints. You do not write it and cannot see it.
Conversation History
Every earlier message in the current chat, yours and AI's. AI re-reads all of it every turn. This is how AI "remembers."
Context Window
The ceiling on how much text AI can read in one conversation. When it fills up, the oldest content silently drops off.
Prompt Anatomy
The five parts of a well-formed prompt: instruction, context, input data, output format, examples.
Output Format
The shape you want the answer in: length, structure, tone. One line in your prompt saves you rewriting the whole reply.
Prompt quality (from Week 12)
Specificity
The single biggest lever on prompt quality. Naming the audience, the format, and the constraints so AI is not guessing.
Ambiguity
What makes AI fall back on average defaults. The parts of a prompt AI has to guess. Iteration cannot rescue what specificity should have prevented.
Iteration (from Week 13)
Iteration
Refining a prompt after reading the answer, in the same chat. Your first prompt is a draft. Your second, written on purpose, is almost always better.
Refinement
The act of sending a targeted follow-up: add what was missing, correct what was wrong, or show an example of right. One deliberate change beats a full rewrite.
The Three Moves
ADD (give it the context it did not have), CORRECT (tell it what to stop doing), SHOW (give it an example of right). Every refinement is one of these three.
Diminishing Returns
When each new round of refinement barely changes the answer. Your signal to stop prompting and finish the piece by hand.
Loop
The Ask → Read → Refine → Stop cycle. The four-step rhythm that turns first-answer disappointment into a usable draft.
RTF (from Week 14)
Prompt Structure
A repeatable pattern for writing prompts. RTF is a prompt structure. Chain of Thought is another. Seven of them are taught in Phase 3 (Weeks 14 through 20), each for a different kind of job.
RTF
Role, Task, Format. The workhorse prompt structure. Three parts in one line that make the first prompt land closer to what you actually needed.
Role (in RTF)
The first line of an RTF prompt. Tells AI who it is being right now. "You are a friendly on-site property manager." Changes what AI draws on and how it talks.
Task (in RTF)
The second line. What AI is actually doing. A verb, an object, and enough detail that two people could not interpret it differently. Not a topic.
Format (in RTF)
The third line. What shape the answer should take. Length, structure, tone. The part almost everyone skips, and the reason so many AI answers feel "close but not quite."
Chain of Thought (from Week 15)
Chain of Thought (CoT)
A prompt structure where you tell AI to show its reasoning step by step before it gives the final answer. Improves accuracy on tasks with real math, comparisons, or judgment. Costs a few extra seconds and a few extra tokens.
Reasoning
The steps between the question and the answer. AI can produce reasoning that sounds correct and is not, which is why CoT works: you get to check the steps against reality before you act on the conclusion.
Step-by-Step
The most common CoT trigger phrase. "Think step by step" added to any RTF prompt turns on the reasoning walk-through. Simple, portable across every AI tool.
Show Your Work
The everyday-English version of CoT. What a teacher asks a student. When you ask AI the same thing, you get the same benefit: visible arithmetic, visible assumptions, catchable errors.
Few-Shot (from Week 16)
Few-Shot Prompting
A prompt structure where you give AI a small set of examples of the answer you want (usually two to four), then ask AI to produce a new one that matches the same pattern. The examples do the teaching. Faster than describing a voice, format, or category rule in words.
Zero-Shot Prompting
The default: you describe the task and give no examples. What most people do without thinking about it. Fine for factual questions and quick edits. Weak when you want AI to match a specific voice or pattern.
Example (in prompting)
A concrete sample of the answer you want. In Few-Shot, the examples together form the pattern AI will match. Consistent, sanitized, and labeled the same way every time. Also called a "shot."
Pattern Matching
What AI does when you give it examples: it looks for what stays constant across them and reproduces that constancy in its new answer. This is the mechanism Few-Shot relies on. If your examples do not share a pattern, there is nothing to match.
Guardrails (from Week 17)
Negative Instruction
A "do not" line in a prompt. The core building block of a guardrail. Tells AI what to leave out, refuse to do, or fall back on when it does not know. The opposite of positive instruction, which tells AI what to do.
Positive/Negative Pairing
The most reliable guardrail pattern: "Do not X. Say Y instead." The negative side sets the boundary. The positive side gives AI a lane to move into. Negative-only leaves AI without a fallback and often gets ignored.
Hedge Word
Softener words AI uses when it is uncertain: approximately, roughly, about, kind of, sort of, probably. Fine in personal notes. A liability in resident, leadership, or vendor communications, because they turn into commitments the reader remembers as facts.
This week · Persona (Week 18) *NEW
Persona
The sustained character AI holds across every message in a session. Goes beyond a one-line Role (from RTF, Week 14). Four to eight lines that lock voice, vocabulary preferences, decision-making style, and refusals for the whole session, not just the next answer.
Persona Drift
When AI slips out of the character you set, usually after 5 to 10 messages in a long session. Voice softens or hardens, vocabulary preferences leak out, decision-making style shifts, refusals get forgotten. Not a failure of AI. It is a limitation of how much of the earlier context still carries weight later in a conversation.
Reinforcement
A short mid-session message that restates the persona to lock character back in when drift starts. Usually one or two lines. Cheaper than restarting the session. Every long-session persona needs a plan for reinforcement.
Governance and data (running through the curriculum)
Data Classification
Sorting a piece of information by which AI tool it is allowed to enter. Tier 1 (safe for any AI), Tier 2 (enterprise only), Tier 3 (never in AI). Full treatment: Week 1.
Tier 1 · Safe
Public or fully de-identified information. Any approved AI tool is fine.
Tier 2 · Enterprise Only
Sensitive but useful information. Only in an approved enterprise AI workspace with data protections in place.
Tier 3 · Never
Information that never enters an AI tool. Real names tied to sensitive details, financial identifiers, protected-class data, and screening or eviction information.
Use Case
A specific way you plan to use AI: who is doing it, what task, what data goes in, what comes out. The Council approves use cases, not tools.
Guardrail
A rule that limits what AI can do. Two forms: (1) governance guardrails are rules built into your workspace or Council-approved use case; (2) prompt guardrails are the "do not" lines you add inside a single prompt (Week 18). Same idea, different scope. The prompt-line version is what a leasing pro can write today without a policy update.
Human-in-the-Loop
A person reviews AI output before it goes to a resident or a decision. The RETTC principle behind almost every safe use case.
AI Council
The group in your organization that approves use cases, reviews new tools, and owns the response when something goes wrong. Full treatment: Week 7.
NIST AI RMF
National Institute of Standards and Technology AI Risk Management Framework. The four functions: GOVERN, MAP, MEASURE, MANAGE.
RETTC AI Governance Framework
Real Estate Technology and Transformation Center. Eight principles specific to rental housing. Detailed on Slide 18.
Reference · Bookmark this slide
Persona: the reference card.
Everything Week 18 taught, on one screen. Bookmark this slide. Come back to it whenever a session needs to hold character across more than one message.
The four ingredients of a Persona
1. Voice. Formal, warm, direct, playful, corporate. Pick one and name it.
2. Vocabulary preferences. What this character says. What they never say.
3. Decision-making style. Defer to policy, escalate, verify before committing, weigh cost vs risk.
4. Refusals. What this character will not do, and what they say instead.
2. Vocabulary preferences. What this character says. What they never say.
3. Decision-making style. Defer to policy, escalate, verify before committing, weigh cost vs risk.
4. Refusals. What this character will not do, and what they say instead.
When to reach for Persona
Batch drafting across many recipients · a long working session on one project · handoffs where character matters (vendor, escalation, compliance) · coaching or role-play · building a saved assistant for a role.
When to skip Persona
One prompt, one answer (use RTF Role) · factual lookups · brainstorming (Persona narrows range) · anything a compliance-first persona might make you overtrust.
The Week 18 bright line
Persona changes how AI writes. It does not change what AI knows. A compliance-first persona is not a lawyer.
The drift and reinforcement pattern
Expect drift somewhere between message 5 and message 10 of a long session. Watch for: voice softening, vocabulary you specified going missing, decision-making style shifting, refusals being ignored. When you see it, drop one line of Reinforcement: "Staying in the persona from the start of this session." Character locks back in.
Permanent · Standards Library
NIST AI Risk Management FrameworkNIST AI RMF (NATIONAL INSTITUTE OF STANDARDS AND TECHNOLOGY, AI RISK MANAGEMENT FRAMEWORK)
The US federal framework for managing risk in AI systems. Four functions: GOVERN, MAP, MEASURE, MANAGE. Voluntary but widely adopted.
Multifamily example: A regional manager wants every property team to use the same maintenance dispatcher Persona for vendor communications. The AI Council walks NIST's four functions: GOVERN (who owns and approves the Persona block for that role), MAP (which failure modes matter most: overcommitment, dropped refusals, drift after long sessions), MEASURE (spot-check outbound emails against the Persona for one month), MANAGE (revise the Persona when a new failure mode shows up, or when drift patterns change).
Full document at nist.gov/itl/ai-risk-management-framework & RETTC 8 PrinciplesRETTC AI GOVERNANCE FRAMEWORK (REAL ESTATE TECHNOLOGY AND TRANSFORMATION CENTER)
Eight principles for AI governance in rental housing. Anchored to Fair Housing, resident privacy, and human accountability.
Multifamily example: A leasing team saves a "renewal outreach" Persona and reuses it across every property. RETTC's Establish an Organizational Philosophy principle says a shared Persona reflects team standards. Uphold Transparency & Explainability says the Persona should be documented like any templated communication. Persona as a governance asset, not a personal preference.
Published at rettc.org/frameworks/ai-governance.
Two frameworks. The NIST AI Risk Management Framework (AI RMF) from the National Institute of Standards and Technology, and the AI Governance Framework from RETTC (Real Estate Technology and Transformation Center). US-anchored. Every Week 18 concept about Persona lands somewhere in one or both.
Week 18 · Persona
NIST AI RMF anchor
RETTC AI Governance anchor
The four-ingredient Persona block
GOVERN function: which Persona becomes the standard for which role is a governance decision. A shared Persona for the maintenance dispatcher role is a governed asset, not a personal preference.
Establish an Organizational Philosophy on AI Use and Oversight: a saved Persona that reflects team standards is a lever for the philosophy, not a workaround around it.
Persona drift and reinforcement
MEASURE function: tracking whether the Persona held across a session. If a Persona is drifting on average by message 5 in one tool and message 10 in another, you cannot tune it without measuring where drift starts.
Uphold Transparency & Explainability: outputs going to residents or vendors should reflect the same voice and refusals every time. Documented drift patterns are how the team maintains that consistency.
Compliance-first personas and the bright line
MAP function: identifying the failure modes that matter for this task. A confident-sounding compliance Persona increases overtrust risk, which is a mapped failure, not a design goal.
Promote Fairness & Consumer Protection: operators are responsible for outputs that comply with Fair Housing and consumer protection law. A Persona does not add legal authority. Cite the source or route to a real reviewer.
Personas as saved assistants in a governed workspace
GOVERN function: which Personas become saved assistants, who can edit them, and how they are versioned are governance decisions. This is where a Persona becomes an operational asset.
Protect Privacy & Promote Data Integrity: a saved Persona defined at the workspace level does not pull real resident data into the Persona block itself. Bracketed inputs at the task level, not real data embedded in character.
Persona plus Guardrails plus human review
MANAGE function: the incident response plan covers what happens when a convincing Persona causes overtrust and a decision goes out without review. Rehearse it before it happens.
Maintain Outcome-Based Accountability & Human Involvement: the person who sent the email owns the outcome. The Persona does not sign it. Human review does not disappear because AI is now in character.
#
RETTC Principle (verbatim, in the RETTC-listed order)
1
Establish an Organizational Philosophy on AI Use and Oversight. Adopt a clear philosophy that defines how AI will be used, governed, and monitored across operations.
2
Promote Fairness & Consumer Protection. Ensure AI systems comply with the Fair Housing Act, Fair Credit Reporting Act, Americans with Disabilities Act, and state and local law.
3
Uphold Transparency & Explainability. Promote a general understanding of when AI influences key decisions, and provide clear explanations to renters and applicants of decisions affecting them as required by law.
4
Protect Privacy & Promote Data Integrity. Adhere to federal, state, and local consumer privacy laws, with an emphasis on data minimization, security, and informed consent.
5
Maintain Outcome-Based Accountability & Human Involvement. Engage humans in significant decisions, establish processes to review model outcomes, provide escalation channels for complaints and appeals, and educate staff on AI use and limitations.
6
Promote Positive Renter Experiences and Outcomes. Elevate technologies that enhance affordability, access, and housing stability, and use AI to improve resident experiences.
7
Strive for Innovation with Responsibility. Encourage development of AI tools that benefit residents, prospects, guests, and operators, without compromising trust.
8
Encourage Third-Party Partner Accountability and Due Diligence. Incorporate robust due diligence and risk management controls for third-party AI tools. Invite transparency from vendors regarding data sources, model limitations, and known risks.
Read the source
NIST AI Risk Management Framework: nist.gov/itl/ai-risk-management-framework
RETTC AI Governance Framework: rettc.org/frameworks/ai-governance
RETTC launch press release: rettc.org/press-release/rettc-unveils-first-of-its-kind-ai-governance-framework-for-rental-housing-and-technology-partners
RETTC AI Governance Framework: rettc.org/frameworks/ai-governance
RETTC launch press release: rettc.org/press-release/rettc-unveils-first-of-its-kind-ai-governance-framework-for-rental-housing-and-technology-partners
Reference · Spot the Tier
Spot the Tier: 8 Persona scenarios.
For each scenario, tap the tier you think it belongs to. Feedback shows immediately with the correct tier and the reason. Reset the whole set with the button at the bottom.
Scenario 1
You paste the four-ingredient maintenance dispatcher Persona at the start of a Tuesday vendor session in a general AI tool. All vendor references, apartment numbers, and resident names are bracketed. Ten emails, same character, no real data touched.
Green. Sanitized inputs plus a bounded Persona. The character is the reusable asset. The data is bracketed. Nothing here needs an enterprise workspace.
Not quite. Everything sensitive is bracketed out. The Persona itself does not contain real data. Green.
Scenario 2
To make AI "sound more like the team," you paste three real residents' names and delinquency histories into the Persona block as "examples of the vocabulary and tone to match." You then use this Persona across a general AI tool.
Red. Real resident names plus delinquency detail are Tier 3 anywhere. Embedding them in the Persona does not sanitize them, and the whole session now carries that data forward. The character does not launder the data.
Not quite. The Persona structure is fine. The problem is what you put in it. Real resident PII in the Persona block is Tier 3, same as pasting it into any prompt. Red.
Scenario 3
A renewal outreach Persona (voice, vocabulary, decision-making style, refusals, all bracketed placeholders) is saved as a text file. You reuse it every Monday morning for the week's outreach batch. Ten sessions in, still the same voice.
Green. The Persona is a reusable asset. It contains no real data. Every session uses fresh bracketed inputs. This is exactly the "save it once, use it every week" pattern the workshop teaches.
Not quite. Nothing in the Persona is sensitive, and every session uses bracketed inputs. Reuse across weeks is the intended pattern. Green.
Scenario 4
A compliance-first regional manager Persona is used to review the property team's draft marketing content for Fair Housing risk. The Persona reviews internal drafts. No resident data involved. Any flagged issue routes to a real Fair Housing reviewer before publish.
Green. Internal drafts, no resident data, and the Persona flags for a human reviewer rather than deciding. This is the right way to use a compliance-oriented Persona: as a first-pass screen, not a decision-maker. Slide 9.
Not quite. The Persona is doing screening, not deciding. Every flag routes to a human. No real resident data touches the session. Green.
Scenario 5
You set up a "Fair Housing attorney" Persona in a general AI tool and use it to decide which rental applicants to approve. Real applicant names, income, and background summaries are pasted in to get the "attorney" recommendation.
Red. Two failures stacked. First, real applicant data (name, income, background) is Tier 3 in any general AI tool. Second, the Persona does not make AI a lawyer. Slide 9 bright line, plus data classification. Decisions here need real legal review, not an AI character.
Not quite. Real applicant PII in general AI is Tier 3 regardless of Persona. And the Persona does not add legal authority. Red on both counts.
Scenario 6
A vendor management Persona (voice, trade vocabulary, decision-making style, refusals) is used across a batch of ten vendor follow-ups. All vendor names, invoice amounts, and property addresses are bracketed. Standard general AI tool.
Green. Bracketed inputs plus a Persona that specifies refusals (no unverified commitments, no unauthorized amounts). Nothing sensitive enters the session. This is what a saved dispatcher assistant does in a governed workspace, run manually.
Not quite. Every real detail is bracketed. The Persona does not contain sensitive data. This is a clean Green.
Scenario 7
You paste your delinquency dispatcher Persona followed by three real residents' unbracketed names, apartment numbers, and past-due balances so AI can draft outreach emails using the Persona's voice.
Red. The Persona is fine. The unbracketed real data is not. Real name plus apartment number plus balance in a general AI tool is Tier 3, and the Persona does not sanitize the input. Same lesson as Week 17's bright line on Guardrails. Bracket first, then Persona.
Not quite. The Persona did not sanitize anything. Real resident data pasted into a general AI tool is Tier 3 regardless of what character AI is playing. Red.
Scenario 8
Inside your enterprise AI workspace (SOC 2, data protections in place), a saved "renewal assistant" Persona pulls real resident tenure and renewal history from the PMS to draft personalized outreach. All within the governed workspace.
Yellow. Real resident data is involved (tenure, renewal history), so this is Tier 2, not Tier 1. The enterprise workspace with SOC 2 is what makes this appropriate. Never do this in a general AI tool. Enterprise-only.
Not quite. Real resident data is in play. That takes it out of Tier 1. The enterprise workspace with data protections is what makes it acceptable at all, which is Tier 2. Yellow.
Take Action · Quiz
Anonymous quiz: five questions.
Tap the answer for each. Feedback shows immediately. No score is stored anywhere and no name is attached. If you want, submit the anonymous results at the bottom to help tune the next workshop.
Question 1 of 5
What is the single clearest difference between a Role (RTF, Week 14) and a Persona (Week 18)?
Teach: Length is a symptom of the real difference, which is scope. Role shapes one answer. Persona holds voice, vocabulary, decision-making style, and refusals across ten or more messages in the same session. Slide 5.
Question 2 of 5
A Persona should specify four things. Which set is complete?
Teach: The four ingredients: Voice (formal, warm, direct), Vocabulary preferences (trade terms in, corporate words out), Decision-making style (verify, escalate, defer), Refusals (what this character will not do). Miss any of the four and AI fills the gap by guessing. Slide 6.
Question 3 of 5
You notice AI drifting from the Persona around message 7 of a long session. What is the right response?
Teach: Reinforcement is the fix. One line: "Staying in the persona from the start of this session." Cheaper than restarting. Faster than re-pasting. This is the drift-plus-reinforcement pattern the workshop teaches. Slide 8, Slide 11, Slide 17.
Question 4 of 5
You set up a "Fair Housing attorney" Persona to review your marketing copy. What is the risk?
Teach: This is the Week 18 bright line (Slide 9). A confident-sounding legal Persona does not add legal authority. It makes wrong answers sound more credible, which is exactly the wrong direction for compliance work. Cite the source or route to a real reviewer. Slide 9.
Question 5 of 5
A general AI tool session where the Persona itself is fully bracketed placeholders (no real names, no real balances). Which tier?
Teach: Green. Data classification is about what enters the session, not what character AI is playing. Bracketed inputs plus a bounded Persona is safe use in any approved tool. The Persona is the reusable asset. See the Slide 19 scenarios for the pattern.
Anonymous submission (optional)
Only your five answers (correct or incorrect) are sent. No name, no email, no identifying data. Used to tune the workshop across the tour.
Submitted. Thank you.
Take Action · Downloads
Take it with you.
Two downloads. Both are yours to keep, reuse, and share with your team. Print them, save them, put them in a shared drive so anyone using AI on your properties has them.
Download 1
Persona Quick Reference (PDF)
One-page landscape reference. Four ingredients of a Persona, the maintenance dispatcher Persona template, when to reach for Persona, when to skip, the bright line, the drift-plus-reinforcement pattern. Bookmark and reuse.
Download PDF
Download 2
AI Vocabulary Cheat Sheet, 49 core terms (PDF)
Every prompt-structure and governance term the workshop has taught so far, plus the three new Week 18 terms: Persona, Persona Drift, Reinforcement. Two-page reference. Matches Slide 16.
Download PDF
Send this to your supervisor
If your property or company is building a shared saved-assistant library, a Persona block is the source material. Send this workshop to your supervisor so they can pick which Personas your team should have saved and standardized. You fill in your supervisor's address.
Compose email to my supervisor
Prefer the readable version?
The full 49-term cheat sheet lives on Slide 16 for browsing, or use the PDF download above to print or share.
Take Action · Close
Session complete.
What you walked away with
The Persona definition · the four ingredients (voice, vocabulary preferences, decision-making style, refusals) · when to reach for Persona and when to skip it · the maintenance dispatcher walkthrough · the drift-plus-reinforcement pattern · the Week 18 bright line · two downloadable references · NIST and RETTC anchoring for Persona as a governed asset.
Good job. Five weeks ago you had RTF. Then Chain of Thought. Then Few-Shot. Then Guardrails. This week you added Persona. Five of the seven prompt structures for Phase 3 are in your hands. Two left. The curriculum board is right here, and this is the map for the next two weeks.
Phase 3 · The Seven Prompt Structures
Week 14 · RTF ✓ done
Week 15 · Chain of Thought ✓ done
Week 16 · Few-Shot ✓ done
Week 17 · Guardrails ✓ done
Week 18 · Persona ✓ this week
Week 19 · Structured Output → next Tuesday
Week 20 · Iterative Prompting → the week after
Coming up next Tuesday
Week 19 · Structured Output. Getting AI to return answers in a specific shape (a table, a list, a JSON block, an email with named fields) instead of a paragraph. The structure that turns AI output into something your workflow, your report, or your handoff can actually use. See you Tuesday.
Where the whole workshop lives at scale
Everything Week 18 taught (Persona blocks, drift-plus-reinforcement pattern, refusals as governed refusals) is exactly what a governed AI workspace does automatically for a saved assistant. That is mPro Digital Edge. Personas become saved assistants. Refusals become policy. Drift becomes something a supervisor sees, not something you catch by rereading email 7. Trust me.
See you Tuesday.
