Warmup

The Workshop

Power Preview

Resources & Tools

Take Action

Week 16 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 Chain of Thought. Tap each to expand, tap again to close. Just for you, no scoring.

Answer: One added line at the end. "Think step by step. State your assumptions before you recommend" is the workhorse version. Six trigger phrases work; any of them do the job. That is the whole trick.

Example of use

Role: You are a regional weighing a vendor decision. Task: Compare Bid A and Bid B, recommend one. Format: One paragraph with annual costs shown. Chain of Thought: Think step by step. State assumptions before you recommend. Four extra words, whole different answer.

Answer: On tasks where the reasoning does not help you: writing tasks with no math or judgment, quick edits, short factual questions, or drafts you are going to send as-is. CoT makes AI produce more text. If you are not going to read and check the reasoning, you have made yourself slower for no gain.

Example of use

"Draft three subject line options for a re-engagement email" is a writing task. Adding "think step by step" makes AI explain how it thought about writing subject lines, which is not useful. Send the RTF version, take the three options, pick one. Save CoT for the vendor comparison later.

Answer: No. Reasoning transparency changes how you check the answer. It does not change what data belongs in a general AI tool. Bracket sensitive slots, get the reasoning, fill in the real values yourself.

Example of use

Wrong: "Sarah Martinez in unit 214 owes $2,847. Walk through step by step whether to file for eviction." Right: "Resident with [X months] tenure and $[Balance] past due, walk through step by step how to weigh eviction versus a payment plan." Same CoT quality. Real details stay in your PMS.

Missed last week?

Open Week 15 · Chain of Thought before you start this one. Password on that page is 1552. This week builds on it.

The Workshop · Opening

Show. Do not tell.

Two weeks in a row we told AI what to do. RTF told it the shape of the answer. Chain of Thought told it to show its work. This week we take a different route: we show AI what a good answer looks like, then ask for one more just like it.

This is Few-Shot. You paste in two or three examples of the answer you want, then ask for a new one that matches the pattern. When you are trying to get AI to match your team voice, a specific format, or a category rule your team already knows, showing is faster than telling. Every time.

This week in one line

Paste in the answer you want to get. Then ask AI for another one just like it.

Read this workshop as

Tap the role closest to yours. A few examples throughout will lean toward that context. Change your mind anytime.

The Workshop · Agenda

Six parts. About 25 minutes. No timer.

Self-guided. Pause whenever you need. Your notes and progress save to this device automatically.

What Few-Shot is

Slides 4 and 5. Show, do not tell. Anatomy of a good example.

~4 min

When to use it and when to skip it

Slides 6 and 7. Great for voice-matching and category rules. Not great when you have no examples.

~3 min

Three structures, one prompt

Slides 8 through 10. RTF plus CoT plus Few-Shot, stacked.

~4 min

Power Preview: two examples

Slides 12 through 15. Voice-matching a leasing follow-up. Teaching AI your emergency vs routine triage.

~7 min

Reference and reinforcement

Slides 16 through 19. Vocab (43 terms), Few-Shot reference card, standards, Spot the Tier.

~4 min

Quiz, downloads, and close

Slides 20 through 22. Five questions, two downloads, Week 17 teaser.

~3 min

No timer runs. The minute estimates are honest averages. Slower is fine.

The Workshop · Why You Should Care

Some things are faster to show than to describe.

Try to write instructions for AI that capture your team's voice, in words. Warm, professional, direct, uses contractions, no exclamation points, sign-off is "See you soon," not "Best." That was 20 words. And you have already forgotten half of it. And AI still will not quite get it right.

Now try this instead: paste three of the actual emails your team sends, and say "write one more like these." That takes 10 seconds. And AI gets it right, first try.

Scenario 1

The voice you cannot describe

Your best leasing pro writes follow-up emails that convert. Warm but not chatty. Direct but not pushy. Uses specific detail from the tour. You want the whole team writing that way. You have tried to train them by describing what makes it good. It has not worked. Because "good" is not describable in words. It is describable in examples.

Scenario 2

The rule everyone knows and no one wrote down

Your maintenance team knows what counts as an emergency versus a routine request. They know it after ten years of doing the job. But if you asked them to write it down, you would get three paragraphs of edge cases that would not quite capture it. Show AI ten past requests with the correct label on each, and you can get the classifier your team has been living without.

Here is the thing

Most of your team's best practices are not documented. They live in the work itself. Few-Shot is the fastest way to lift them out of the work and hand them to AI without rewriting them into a training manual first.

The Workshop · Teaching A

Few-Shot in one line.

The definition

Few-Shot is when you give AI a small set of examples of the answer you want, then ask AI to produce one more that matches the same pattern.

The name comes from the AI research world. A "shot" is an example. Zero-Shot is what you have been doing your whole life with AI: no examples, just a description of the task. Few-Shot is what happens when you paste in two, three, or four examples of the answer you want and then say "now write one for [new case]."

Two things happen when you do this. AI locks onto the pattern in your examples. Then it produces a new answer that matches the pattern. Not just the topic. The pattern. The rhythm, the length, the tone, the format, the sign-off, the category rule underneath.

Zero-Shot (no examples)

You describe. AI guesses.

"Write a warm, professional follow-up email to a prospect who toured yesterday. Direct but not pushy. Mention a specific detail from the tour."

Result: generic corporate email. Sounds like every other property. Uses "wonderful" three times.

Few-Shot (with examples)

You show. AI matches.

"Here are three follow-up emails our team has sent. [Example 1] [Example 2] [Example 3] Now write one for a prospect who toured yesterday and mentioned they wanted a quiet apartment."

Result: matches your voice. First try.

Pay attention

Two examples is the minimum. Three is the sweet spot. Four or five is fine. One example is not Few-Shot, it is imitation, and AI will often just rewrite your one example instead of learning the pattern.

The Workshop · Teaching B

Anatomy of a good example.

The examples do the teaching. If the examples are inconsistent, AI learns the wrong pattern. Four rules for the set you paste in.

The examples agree with each other

If two examples end with "See you soon" and the third ends with "Best regards," AI will pick one and be unpredictable about which. Pick examples that agree on tone, length, structure, and sign-off. The pattern has to be visible in every example, not average across them.

The examples cover the variety AI will see

If you are teaching AI to classify service requests as emergency versus routine, do not show only emergencies. Show two emergencies and two routines, and pick examples that would fool someone who did not know your rules. The interesting cases are the ones near the boundary.

The examples are labeled the same way every time

Use a consistent label. "Example 1:" "Example 2:" is fine. "Request: [text] Classification: emergency" is fine. AI needs to see where the input ends and the output starts. If your labels are inconsistent, AI will not know which part of your example is the pattern to reproduce.

The examples are sanitized

Real resident names, unit numbers, phone numbers, and balances get replaced with placeholders before you paste. [Resident], [Unit], [Balance], [Address]. Your best examples are still real correspondence. The pattern in them is what you want AI to learn. The private details are not.

The Workshop · Teaching C

Five times to reach for Few-Shot.

Any time the words "this, but more like ours" would answer the question, Few-Shot is what you want.

Use case 1

Matching a voice

Team voice, your voice, Tami's voice, your best leasing pro's voice. Three examples and AI can match it. Describing voice in words almost never works.

Use case 2

Replicating a format

Rent renewal letters. Work order confirmations. Move-in packets. Tour recap notes. If your team already has a format that works, show three, ask for one more.

Use case 3

Teaching a category rule

Emergency vs routine. Priority vs standard. Escalate vs handle. Any classification your team knows how to do but has not written down. Four labeled examples give AI the rule.

Use case 4

Standardizing across the team

If different team members write inquiry replies differently, pull three examples of the version you want to be the standard. Now everyone using AI writes the standard version.

Use case 5

Anything with a "house style"

Newsletter tone. Social post length. Property description structure. If "we always write it this way" is true, three examples teach AI "this way." Faster than every attempt to describe it.

The tell

When to pick Few-Shot

If your first instinct is to describe what "good" looks like, and you catch yourself using vague words like "warm" or "professional," stop describing. Go find three examples.

The Workshop · Teaching D

Four times to skip Few-Shot.

Few-Shot is not free. Finding three good examples takes time. If the task does not need pattern-matching, skip it.

Skip 1

You do not have examples

If you would have to write the examples yourself first, that took as long as just writing the thing. Use RTF, describe what you want, live with the result. Or make one, review it, and next time you have your first example.

Skip 2

You want variety, not consistency

Brainstorming five different subject lines. Coming up with three different angles. Getting fresh ideas. Few-Shot narrows AI to your pattern, which is the opposite of what you want when you are looking for new directions.

Skip 3

Factual questions and quick edits

"What is the security deposit rule in Florida?" does not need examples. Neither does "fix the grammar in this paragraph." Zero-Shot handles both cleanly.

Skip 4

Your examples contradict each other

If you pick three examples that use three different tones, three different formats, or three different sign-offs, you have taught AI nothing. Better to write one careful RTF prompt than to feed AI three inconsistent examples.

Trust me

The temptation is to use Few-Shot on everything once you learn it. Do not. On half the tasks a leasing professional does in a day, RTF is faster and works fine. Pick Few-Shot when the pattern is the point.

The Workshop · Teaching E

Three structures, one prompt.

RTF gave you the shape. Chain of Thought gave you the reasoning. Few-Shot gives you the pattern. When they stack, you get an answer that is the right shape, arrived at by the right logic, matching the pattern your team already uses.

Stacked prompt example: renewal outreach
Role: You are a senior leasing professional writing renewal outreach for a resident whose lease ends in 60 days.
Task: Draft an email that invites the conversation without pressure, references something specific from their tenure, and offers two paths (renewal at current rate vs upgrade to a different apartment).
Format: Under 150 words. Three short paragraphs. Subject line first.
Chain of Thought: Before you write, think step by step. State what you assume about the resident's situation given the tenure length. Explain what tone would fit given those assumptions.
Few-Shot: Match the pattern of these three examples of our team's renewal outreach:
Example 1: [Paste renewal email 1, sanitized]

Example 2: [Paste renewal email 2, sanitized]

Example 3: [Paste renewal email 3, sanitized]
Now write: A renewal outreach email for a resident with 3 years of tenure who has always paid on time and never opened a service request.

Do not memorize the labels. RTF, CoT, Few-Shot is a checklist to run through, not a form to fill out. On any given task you might use one, two, or all three. When you use all three, they compound.

The Workshop · Teaching F

The bright line.

Week 16 bright line

Examples with real resident details are still real resident data. Sanitize before you show.

Few-Shot has one risk that RTF and CoT do not have. The examples you paste in are the pattern you are teaching AI. That pattern is often lifted straight out of your best past work: real emails to real residents about real balances, real complaints, real move-out disputes.

The reasoning transparency rule from last week still applies. The RTF-shape rule from the week before still applies. Adding Few-Shot to the toolbox does not change either of them. But it adds one new step: before you show, sanitize.

Wrong

"Match the pattern of these three emails I sent last month:
Example 1: To Sarah Martinez in unit 214...
Example 2: To Marcus Chen in unit 512...
Example 3: To the Williams family in unit 302..."

Right

"Match the pattern of these three emails I sent last month:
Example 1: To [Resident] in [Unit]...
Example 2: To [Resident] in [Unit]...
Example 3: To [Resident] in [Unit]..."
Same pattern. Zero resident data.

Trust me

Sanitize once. Save the sanitized versions in a folder called "Few-Shot examples." Reuse forever. The five minutes you spend cleaning three emails is the last time you have to do it.

The Workshop · Default Rule

The default rule for Few-Shot.

The default rule

If you are trying to match a voice, a format, or a category rule, use Few-Shot. If you are trying to explore, ask a factual question, or make a quick edit, do not.

You have three prompt structures now. RTF for shape. Chain of Thought for reasoning. Few-Shot for pattern. Any real task will use one, two, or all three of them together. The tell for which to reach for is what the task asks of the answer.

Shape the answer → RTF. Role, Task, Format. When you know what you want the output to look like.

Reason through the answer → Chain of Thought. When math, judgment, or comparison matter.

Pattern-match the answer → Few-Shot. When you can point to two or three examples of "like this."

The Workshop · Q & A

Six things I get asked about Few-Shot.

Tap any question to expand. Tap again to close.

Answer: Two is the floor. Three is the sweet spot. Four is fine. Five is fine. Past six you are wasting your time. AI has learned the pattern after three consistent examples. More examples rarely make the answer better, and they make the prompt longer than it needs to be.

Example of use

Voice-match on renewal letters: three sanitized examples of the letter you want AI to write. Done. Do not paste in ten. AI is not learning more from ten than from three, and you have burned 20 minutes finding them.

Answer: You can. It is called One-Shot, not Few-Shot, and it is weaker. With one example, AI often just rewrites your example instead of learning the pattern behind it. With two or three, AI can see what stays constant across examples (the pattern) and what changes (the specific content). If you only have one, use RTF and describe what you want.

Example of use

You have one perfect welcome letter you sent last month. Show AI just that one and ask for a new welcome letter for a different resident, and you will get almost the same letter with the name swapped. Show AI three sanitized welcome letters from different residents and you get a real new letter that matches the pattern.

Answer: Depends on what you are teaching. If you want the answer to match your team's voice, the examples must come from your team. If you want the answer to match a general pattern (a standard format for a rent renewal letter, a standard structure for a service request response), any consistent good examples work. If you want the answer to match your team's category rules, the examples must be labeled the way your team labels them.

Example of use

Trying to draft social captions in your property's voice? Use captions your property has posted. Trying to write a standard NSF letter that has to match state notice requirements? Any three good NSF letters will do, because the pattern lives in the law and not in your voice.

Answer: Then you have three different patterns and AI does not know which one to match. Either pick three examples that agree with each other, or accept that this is not a Few-Shot task and use RTF instead. The mistake is pasting three inconsistent examples and blaming AI when the answer is inconsistent too.

Example of use

Move-in emails from your team: one is 80 words and warm, one is 200 words and formal, one is 50 words and casual. That is not Few-Shot material. Pick a lane. Show three that are all 80 words and warm, or three that are all 200 words and formal.

Answer: The prompt structure works with text. The AI models that support image input can accept image examples too, and they will pattern-match on those (layout, color use, visual style) the same way they pattern-match on text. Support varies by tool and by model version, and it changes fast. You may want to verify what your specific tool supports before you plan a workflow around it.

Example of use

A marketing team pastes three past property flyers as images and asks for a new flyer in the same style. Works on the tools that accept image input. Does not work on tools that only accept text. Check your tool.

Answer: Usually yes. RTF sets the frame. Few-Shot fills it. Without RTF, AI has to guess whether you want a new example, an analysis of the examples, or a summary of them. State the Role, state the Task, state the Format, then say "match the pattern of these examples," then paste them. Stack it.

Example of use

Bad: "Match these three [paste examples]." AI might summarize them. Good: "You are a leasing pro. Task: draft a new prospect follow-up. Format: under 100 words. Match the pattern of these three: [paste]." Now AI knows the frame and the pattern.

Power Preview · Setup

Two side-by-side examples.

Two tasks. The Zero-Shot version on the left. The Few-Shot version on the right. Same task, same AI, different structure. Read both.

Example 1

Voice-Matched Follow-Up

Every property manager has watched a great follow-up email from their best leasing pro and thought "why does everyone not write like that." Here is how you get AI to write like that. Show it three.

Example 2

Emergency vs Routine Triage

Your maintenance team knows the difference between an emergency and a routine ticket. New team members do not, not yet. Four labeled examples turn AI into a triage assistant that follows your team's rules.

Read this first

The examples on the following two slides use placeholders like [Resident] and [Unit] instead of real details. That is the same sanitization step your team should use on your own examples. Nothing on this slide is invented AI wizardry, this is exactly what a real prompt should look like in your tool tomorrow.

Power Preview · Example 1

Voice-matched leasing follow-up.

Same prospect. Same tour. Same AI. Left is what most people type. Right is Few-Shot.

Zero-Shot (describe)

Write a warm, professional follow-up email to a prospect named [Name] who toured a two-bedroom yesterday. Direct but not pushy. Mention something specific from the tour.

What you get back

"Dear [Name], It was such a pleasure showing you around our wonderful community yesterday. I hope you enjoyed learning about all of the amazing amenities we have to offer. I would love to answer any questions you may have as you consider making [Property] your new home. Please do not hesitate to reach out at your earliest convenience. Warm regards, [Your Name]."

Sounds like every property. Uses "wonderful." Uses "amazing." Nobody would ever say "at your earliest convenience" out loud. Prospect skims and deletes.

Few-Shot (show)

Match the pattern of these three follow-up emails our team has sent:

Example 1: Hi [Name], really glad you came by yesterday. I know the courtyard-facing two-bedroom checked a lot of your boxes, and I wanted to send you the floor plan I promised. If it helps, I can hold that one until [Day]. Let me know. Sarah

Example 2: Hi [Name], quick note after your tour, thanks for stopping in. You mentioned the commute was a big factor, so I pulled the average drive times on our resident survey. Two of them work at your building. Happy to connect you if you want to hear from a neighbor. Sarah

Example 3: Hi [Name], I've been thinking about our conversation about the dog park. Wanted to let you know we have a resident meet-up this Saturday if you want to see it in action. No pressure at all. Sarah

Now write a follow-up email for a prospect named [Name] who toured a two-bedroom yesterday and mentioned they wanted a quiet apartment.

What you get back

"Hi [Name], appreciate you stopping in yesterday. You said quiet was the big thing, so I looked at what we have on the interior side of the courtyard, which stays quieter than the street-facing apartments. I can hold one for you through [Day] if that would help. Let me know. Sarah."

Matches the voice. Same length. Same rhythm. Same sign-off. Prospect reads it.

The lift

Sanitize three real follow-ups once. Save them in a "Few-Shot examples" folder. Every follow-up email your team sends from now on can start from those three. That is a two-hour task with a payoff that runs for years.

Power Preview · Example 2

Teaching AI your triage rules.

Your team knows what counts as an emergency. New hires do not, not yet. Show AI four labeled examples and it can help apply your team's rules to new incoming requests.

Zero-Shot (ask cold)

Is this an emergency or a routine service request? "The AC in [Unit] is not cooling below 78, outside temp is 91."

What you get back

"This appears to be a routine service request. The AC is functioning but underperforming, which is typically categorized as a comfort issue rather than an emergency..."

Wrong. Your team classifies this as an emergency in Florida in July. AI does not know your team's rules. It guessed.

Few-Shot (labeled examples)

Classify service requests as EMERGENCY or ROUTINE using our team's rules. Here are four past examples with correct labels:

Request: "No hot water in [Unit]." Classification: EMERGENCY

Request: "AC not cooling below 78, outside temp above 85." Classification: EMERGENCY

Request: "Dishwasher will not drain, food is sitting in it." Classification: ROUTINE

Request: "Front-door lock is sticking but still works." Classification: ROUTINE

Now classify this new request: "The AC in [Unit] is not cooling below 78, outside temp is 91."

What you get back

"Classification: EMERGENCY. This matches Example 2 (AC not cooling below 78 with outside temperature above 85)."

Right. AI now applies your team's rule instead of guessing.

Pay attention

AI classifying a service request is not AI dispatching a service request. The AI Council still owns whether the tool sees resident data at all, and a human still owns the actual dispatch. This is help, not replacement. Same rule as every week.

Power Preview · Try This Today

Twelve minutes. Voice-match your own writing.

One exercise you can do today, before this week ends. You will have a Few-Shot prompt of your own by the time you finish.

Pick a writing task you do every week (4 min)

Follow-up emails. Renewal notices. Inquiry replies. Move-in confirmations. Whatever task you personally do most. Not a task, an exact type of message.

Find three you have written that you were proud of (5 min)

Not three at random. Three that you would be proud to hand a new hire and say "write like these." Consistency matters more than volume.

Sanitize them (2 min)

Replace names with [Resident]. Replace unit numbers with [Unit]. Replace balances with [Amount]. Save the sanitized versions in a folder called "Few-Shot examples." Bright line reminder from Slide 9.

Run the prompt on a real task (1 min)

Next time you sit down to write one of these messages, paste your three examples first, then ask AI for a new one for your actual case. See what you get.

Good job

You now have a Few-Shot prompt for one type of message. Do it once more this month for a second type. By the end of Phase 3, you have a folder full of Few-Shot templates that turn every AI session into a voice-matched draft.

Reference · Bookmark this slide

Vocabulary cheat sheet: 43 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.

This week · Few-Shot (Week 16) *NEW

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.

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 inside your prompt or your workspace that limits what AI can do. "Do not invent numbers" is a guardrail. Full treatment: Week 17.

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.

💾 Download AI Vocabulary Cheat Sheet (PDF)

Reference · Bookmark this slide

Few-Shot, on one page.

Read the left column when you are writing a prompt. Read the right when you want to see it in the wild.

The template

Start with your RTF prompt. Role, Task, Format. Same first three lines as always.

Then paste examples. Label them consistently. Two minimum, three is best, five is the ceiling.

"Match the pattern of these examples: Example 1: [paste] Example 2: [paste] Example 3: [paste]. Now write one for [new case]."

The four rules for the example set are on Slide 5: agree with each other, cover the variety, label the same, sanitize.

Filled in

Role: You are a leasing pro at [Property].

Task: Draft a follow-up email to a prospect who toured yesterday.

Format: Under 100 words. Casual sign-off.

Match the pattern of:

Example 1: [sanitized real email]

Example 2: [sanitized real email]

Example 3: [sanitized real email]

Now write: A follow-up for [Name], who toured yesterday and mentioned wanting a quiet apartment.

When to reach for it

Use Few-Shot for voice-matching, format replication, category rules your team already applies, and any "write more like this" task.

Skip Few-Shot for factual questions, brainstorming that needs variety, quick edits, and any case where you would have to write the examples yourself first.

Watch for these traps

Unsanitized examples. Your best past emails are still real resident correspondence. Replace names, apartment numbers, balances with brackets before you paste. Save the sanitized versions in a folder called "Few-Shot examples."

Examples that contradict each other. Three follow-up emails in three different tones is not a pattern. AI cannot match what is not consistent. Pick a lane.

Reaching for Few-Shot with no examples. If you have to write the examples from scratch, you have already written the thing. Use RTF instead and save Few-Shot for tasks where the examples already exist.

The bright line, always

Examples with real resident details are still real resident data. Sanitize before you show. See Slide 9.

💾 Download Few-Shot Quick Reference (PDF)

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 the whole team using Few-Shot for renewal outreach. The AI Council walks NIST's four functions: GOVERN (who approves what may go in the example set), MAP (Fair Housing risks in the source correspondence and the pattern being taught), MEASURE (are the sanitized examples still triggering resident-data hits in the tool's audit log), MANAGE (what happens when someone pastes an unsanitized example). One framework, four questions, a Few-Shot policy on the record.

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 property uses Few-Shot to teach AI to classify service requests as emergency versus routine. RETTC's Maintain Outcome-Based Accountability & Human Involvement principle says a human still owns the classification when it matters (a habitability call, an ADA-related request). Few-Shot helps AI assist. It does not replace the human who signs off.

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, at rettc.org). US-anchored. Together they cover the risk side (NIST) and the operating side (RETTC) of governing AI in rental housing. Every Week 16 concept about Few-Shot lands somewhere in one or both.

Teaching AI a pattern through examples

GOVERN function: policy defines what examples may be used to train AI behavior at the prompt level. A sanctioned example set is a governed asset, not a personal folder.

Establish an Organizational Philosophy on AI Use and Oversight: a Few-Shot library that reflects team standards is a lever for the philosophy, not a workaround around it.

Sanitizing examples before pasting

MAP function: identifying where PII lives in the inputs an AI system receives. Prompt-side inputs are AI inputs. Sanitization is the mitigation.

Protect Privacy & Promote Data Integrity: data minimization at the point of prompt. The pattern in an example is what you want to share. The identity attached to it is not.

Using labeled examples to teach a category rule

MEASURE function: the labels in your examples become the ground truth AI is matching against. Whether that ground truth is fair, complete, and consistent is what you must be able to measure.

Promote Fairness & Consumer Protection: a category rule taught by example carries any bias in the examples forward. Fair Housing risk lives in the pattern, not just in the individual decisions.

Standardizing team voice through Few-Shot

GOVERN function: which voice becomes the standard is a governance decision, not a personal preference. Whose examples get elevated matters.

Uphold Transparency & Explainability: if AI-drafted communication is going to residents, the pattern behind it should be documented in the same way any templated communication is.

The bright line (examples are still data)

MANAGE function: the incident response plan covers what happens when someone pastes an unsanitized example. Rehearse it before it happens.

Maintain Outcome-Based Accountability & Human Involvement: the person who chose to paste the example owns the outcome. Human accountability does not disappear because AI now uses the pattern.

The RETTC AI Governance Framework · all 8 principles

Official titles as published at rettc.org/frameworks/ai-governance. These are the 8 principles the Framework asks every rental housing provider and technology partner to adopt.

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.

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.

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.

Protect Privacy & Promote Data Integrity. Adhere to federal, state, and local consumer privacy laws, with an emphasis on data minimization, security, and informed consent.

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.

Promote Positive Renter Experiences and Outcomes. Elevate technologies that enhance affordability, access, and housing stability, and use AI to improve resident experiences.

Strive for Innovation with Responsibility. Encourage development of AI tools that benefit residents, prospects, guests, and operators, without compromising trust.

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 RMF: 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

Reference · Spot the Tier

Few-Shot does not change the data rule.

Eight Few-Shot prompts a real person might write this week. Your job: read each one, then tap the tier that matches what the person put into the prompt. The structure never changes the tier. Only the examples do.

Tier 1 · Green: general information, sanitized examples, no PII. Tier 2 · Yellow: real business context but no resident identifiers or financials. Tier 3 · Red: real resident names, real balances, real medical, real applicant status, real portfolio financials embedded in the examples.

Scenario 1

"Match the pattern of these three follow-up emails: Example 1: To [Name], thanks for touring the two-bedroom... Example 2: To [Name], following up on the one-bedroom you saw... Example 3: To [Name], quick note after your visit... Now write one for [Name], who toured yesterday."

Green. Every name is bracketed. Examples are fully sanitized. This is what "show, do not tell" looks like when the bright line is respected.

Not quite. Every name is a placeholder. No identifiers anywhere in the examples. Green.

Scenario 2

"Here are three renewal outreach emails I sent last week. Example 1: To Sarah Martinez in 214, thanks for three great years... Example 2: To Marcus Chen in 512, your lease is up March 15... Example 3: To the Williams family in 302, ready to talk renewal? Write one more for Jessica Rivera in 108."

Red. Four real resident names and four real unit numbers in one prompt. This is exactly what the Slide 9 bright line prohibits.

Reread the prompt. Four full names and four unit numbers. Real residents. Red.

Scenario 3

"Classify these service requests as EMERGENCY or ROUTINE. Example 1: ‘No hot water in [Unit].’ = EMERGENCY. Example 2: ‘AC not cooling below 78, outside above 85.’ = EMERGENCY. Example 3: ‘Dishwasher will not drain.’ = ROUTINE. Example 4: ‘Front-door lock is sticking.’ = ROUTINE. Now classify: ‘Bathroom fan not working, no ventilation.’"

Green. Apartments bracketed as [Unit]. No names. Pure category rule taught by example. This is Few-Shot at its cleanest.

Look again. Every apartment is bracketed as a placeholder. No names. Just the classification pattern. Green.

Scenario 4

"Here are three ADA accommodation approval letters we sent. Example 1: Approved [Resident]'s emotional support animal request based on letter from Dr. Patel dated 3/12. Example 2: Approved [Resident]'s ramp modification request based on physical therapy documentation... Now draft one for a new accommodation request."

Red. The residents are bracketed but the disability details, the medical providers, and the dates are not. ADA-related accommodation records are Tier 3. Bracketing the name is not sanitization; the whole example is the sensitive story. Draft the letter in a Council-approved enterprise workspace, not a general AI tool.

The residents are bracketed. The medical, provider, and disability details are not. ADA-related detail is Tier 3 regardless of the resident bracket. Red.

Scenario 5

"Match the pattern of these three social media captions from our property's Instagram feed. Example 1: . Example 2: . Example 3: . Now write a new caption for our fall resident event."

Yellow. The captions themselves are public marketing content, but they carry your property's specific voice and reference the property name. That is real business context. Enterprise workspace is the right home for it. Not Red, not fully Green either.

The captions are public but they carry property-specific voice and identifiers. That is Yellow, not Green (business context is present) and not Red (no PII). Yellow.

Scenario 6

"Draft a rent renewal letter. Match the pattern of these three: Example 1: [sanitized renewal letter, no names, no unit numbers, no balances]. Example 2: [sanitized]. Example 3: [sanitized]. Now write one for a renewal at market rate."

Green. Fully sanitized examples, generic new task. The pattern gets taught and no resident data goes anywhere. This is the pattern to build a library of.

Every example is labeled sanitized. No PII in the new task either. Green.

Scenario 7

"Here are the last three eviction filings we processed. Example 1: [Full complaint text with resident name, address, balance, history]. Example 2: [Full complaint text, resident 2]. Example 3: [Full complaint text, resident 3]. Draft the next one for a resident 60 days past due."

Red. Three full eviction filings with resident names, addresses, balances, and payment histories. This is the highest-sensitivity category and the pattern is legal filing language, which does not need Few-Shot in a general AI tool. Legal work belongs with counsel and, if AI is involved at all, in a Council-approved enterprise environment.

Three full eviction filings with everything identifiable. Legal filings on real residents are Tier 3 without exception. Red.

Scenario 8

"Match the pattern of these three vendor comparison paragraphs. Example 1: Vendor A (bracketed rate, bracketed terms) vs Vendor B (bracketed rate, bracketed terms), recommended B because... Example 2: [same pattern, generic]. Example 3: [same pattern, generic]. Now write one comparing [Vendor A] and [Vendor B] for a new pest-control decision."

Green. All rates bracketed, all vendors bracketed. The pattern gets taught. Real vendor names and figures get substituted after AI writes the paragraph. Every week, this pattern works.

Every rate, every vendor is bracketed in both examples and the new task. Nothing here identifies a real business relationship. Green.

Take Action · Anonymous Quiz

Five questions. No name, no email attached.

Tap A, B, C, or D for each question. Read the feedback that appears. Nothing is stored with your name. Nothing you type here identifies you. The whole class's results roll up so Tami can spot where the sticky points are for Week 17.

Question 1 of 5

What is the minimum number of examples that makes a prompt Few-Shot instead of One-Shot?

Answer: B. Two is the floor. Three is the sweet spot. Past five you are usually wasting your time.

One example is called One-Shot and AI often just rewrites your one example instead of learning the pattern behind it. With two or three consistent examples AI can see what stays constant (the pattern) and what changes (the specific content), and match the constancy in the new answer. More than five rarely helps and it burns time you spent finding them.

Example of use

Voice-matching your team's renewal outreach: three sanitized past letters is right. Not one (AI copies it). Not ten (AI is not learning more than it did at three, and you burned 20 minutes finding them).

Question 2 of 5

You have exactly one great example of what you want. No time to find or write two more. What is the best move?

Answer: C. Use RTF and describe what you want. Save Few-Shot for when you have two or more real examples.

One-Shot is weaker than Few-Shot; AI usually just rewrites the single example instead of learning the pattern. Pasting the same example three times teaches AI nothing new. Asking AI to generate its own examples and then match them is circular, you are just Few-Shotting AI on itself. RTF is what you built the last three weeks for; it is the right tool when examples are not available.

Example of use

You have one great move-in email you sent last month. Use RTF: "Role: leasing pro. Task: draft a move-in email to [Resident] moving in [Date]. Format: under 100 words, warm, casual sign-off." When you have three great move-in emails you like, that is when Few-Shot earns its place.

Question 3 of 5

Which of the following tasks is the strongest fit for Few-Shot?

Answer: A. Voice-matching an established team style.

Voice is the classic Few-Shot use case: hard to describe in words, easy to show with three examples. B is a variety task (Few-Shot narrows AI, the opposite of what you want when brainstorming). C is a factual question and needs no examples. D is a mechanical edit and needs no examples. Slide 6 lists the five strongest use cases; Slide 7 lists the four times to skip Few-Shot.

Example of use

If your best leasing pro writes follow-ups a certain way and you want AI to match that voice for the whole team, three sanitized examples of her follow-ups will teach AI more in 30 seconds than any RTF prompt describing "warm but direct, uses casual sign-offs, references a specific tour detail" can.

Question 4 of 5

What does the Week 16 bright line say about the examples you paste into a Few-Shot prompt?

Answer: B. Examples with real resident details are still real resident data. Sanitize before you show.

The whole point of Few-Shot is that the examples are the pattern. That pattern is often lifted straight out of real correspondence with real residents. Structure never lowers the sensitivity of what is inside the structure. Names, unit numbers, balances, medical, and applicant status get replaced with brackets before you paste. Same rule as every other week: Slide 9 is the canonical version.

Example of use

Wrong: "To Sarah Martinez in 214, following up on the balance issue..." Right: "To [Resident] in [Unit], following up on the [Topic]..." Same pattern gets taught. Zero resident data leaves the building.

Question 5 of 5

Which situation is the clearest signal to skip Few-Shot?

Answer: C. You would have to write the examples from scratch yourself first.

If you have to invent your own examples, you have already done the work Few-Shot was supposed to save. Use RTF instead, get the first draft out, and next time you have your first real example. A and B are exactly the cases Few-Shot is for. D might be Few-Shot material if you have three past examples and it might be RTF material if you do not; the tell is whether the examples already exist.

Example of use

A brand-new task type (say, a first-ever community-partnership announcement your property has never sent before): no examples exist. Do not force Few-Shot. Use RTF, write the first one, sanitize it, and file it in the Few-Shot examples folder. In a month you have three and Few-Shot works.

Take Action · Downloads

Two files to take with you.

The Quick Reference is Slide 17 on one page. The Vocabulary Cheat Sheet is every AI term the curriculum has introduced through Week 16. Print them, save them to your phone, share them with the team.

Download 1

Few-Shot Quick Reference

The template. The filled-in version. When to reach for it. When to skip it. The three traps to watch for. The bright line at the bottom. Slide 17, in your pocket.

💾 Download (PDF)

Download 2

AI Vocabulary Cheat Sheet

Every AI term the curriculum has introduced through Week 16. 43 terms in plain language. Foundational, prompt anatomy, prompt quality, iteration, RTF, Chain of Thought, Few-Shot, governance, and data. Same list as Slide 16.

💾 Download (PDF)

Send this to your supervisor

If your supervisor has not seen this curriculum yet, drop them a note. Every week people find out their peers were doing this while they were not. This week's note writes itself: Few-Shot is the fastest way I know to make AI sound like our team. Enter your supervisor's email below (nothing is stored, nothing is sent to anyone else) and the button opens a pre-written email in your mail app.

Take Action · Session complete

Session complete.

What you walked away with

• The definition of Few-Shot and how it differs from Zero-Shot and One-Shot.
• The four rules for a good example set (agree, cover, label, sanitize).
• Five times to reach for Few-Shot; four times to skip it.
• The stacked prompt: RTF for shape, CoT for reasoning, Few-Shot for pattern.
• The Week 16 bright line: examples with real resident details are still real resident data. Sanitize before you show.
• The default rule for picking which structure to use when.

Good job. Two weeks ago you had RTF. Last week you added Chain of Thought. This week you added Few-Shot. Three of the seven prompt structures for Phase 3 are in your hands. The curriculum board is right here, and this is the map for the next four weeks.

Phase 3 curriculum board

Week 14 · RTF (Role, Task, Format): the shape
Week 15 · Chain of Thought: the reasoning
Week 16 · Few-Shot: the pattern (this week)
Week 17 · Guardrails: telling AI what NOT to do (next week)
Week 18 · Persona: sustained character across a session
Week 19 · Structured Output: forcing the format
Week 20 · Iterative: refining across turns

Next Tuesday · Week 17

Guardrails: tell AI what NOT to do.

Every week we have added structures for what AI should do. Next week we set the rails inside a prompt so AI stays inside bounds you define. Never make up a stat. Never quote a policy you have not seen. Never say "approximately." This is the safety valve of prompting, and it is the last of the three "defensive" structures.

Bookmark: multifamilynext.com/52-ai-tuesdays/week-17-of-52 (goes live next Tuesday).

Where the platform picks up

Everything you learned this week runs inside mPro Digital Edge. The platform's Chat Assistant lets you save Few-Shot example sets by task type, so you paste them once and reuse them every time. If you have not seen the platform yet, we can walk your team through it. It is the difference between knowing a structure and having it built into how your team works.

See you Tuesday.

Tami Siewruk
CEO, Multifamily NEXT and mPro Digital Edge

Week-16-of-52: Few-Shot in one line

Welcome back. Let us warm up first.

Show. Do not tell.

Six parts. About 25 minutes. No timer.

Some things are faster to show than to describe.

Few-Shot in one line.

Anatomy of a good example.

Five times to reach for Few-Shot.

Four times to skip Few-Shot.

Three structures, one prompt.

The bright line.

The default rule for Few-Shot.

Six things I get asked about Few-Shot.

Two side-by-side examples.

Voice-matched leasing follow-up.

Teaching AI your triage rules.

Twelve minutes. Voice-match your own writing.

Vocabulary cheat sheet: 43 core terms.

Few-Shot, on one page.

The RETTC AI Governance Framework · all 8 principles

Few-Shot does not change the data rule.

Five questions. No name, no email attached.

Two files to take with you.

Session complete.

OR CALL US