Prompting is not one skill. It is five, and most commercial real estate teams only ever practice the first one. That is why the same firm can watch a colleague get a clean lease summary out of ChatGPT on Monday and a mediocre one out of the same tool on Tuesday, then conclude the technology is unreliable. The technology is fine; the skill set is uneven. The best-known prompt template in real estate, A.CRE’s R.O.D.E.S. formula (Role, Objective, Details, Examples, Sense-check), teaches one of those five skills well and leaves the other four unaddressed. This is the full set: a prompt skills framework a 4–20 person firm can teach itself, anchored to the LOIs, lease summaries, and deal screens your team already produces, and scored so a principal can tell whether it worked.
Why “prompt engineering” is the wrong mental model
Calling it “prompt engineering” makes people picture a formula, and a formula is not a skill set. A 2023 study by Harvard, MIT, Wharton, and BCG researchers (Dell’Acqua and colleagues) put 758 consultants through AI-assisted work. Inside the range of tasks AI handles well, consultants using it finished 12.2% more work, moved 25.1% faster, and produced output rated 40% higher in quality. The detail that matters for a CRE firm: those given prompting guidance beforehand outperformed those handed the tool with none. Skill was the variable, not access.
That study also named two learned working styles: “centaurs” split tasks cleanly between human and AI, while “cyborgs” interleave with the tool constantly. Both beat unstructured use, and both are practiced behaviors, not personality types.
A single template underdelivers because a real CRE task exercises several distinct abilities at once. Getting a usable deal screen out of an offering memorandum means giving the model the right source material, telling it what shape the answer should take, correcting it when the first pass misses, checking every number it produced, and knowing whether this is a one-off or a workflow worth systematizing. Those are five separable skills. Teach them as one and people plateau at “write a decent prompt,” roughly a third of the capability. The wider argument for why small firms can build this faster than institutional giants runs through the small CRE firm AI manifesto; this framework is the skills layer underneath it.
The five prompt skills
The skills build in order. Each one raises the ceiling on the next, so a firm that drills them in sequence gets compounding returns rather than five disconnected tips.
Skill 1: Context loading
The skill: giving the model the specific material it needs before asking for anything. This is the single highest-impact prompting skill and the one people skip most, because typing a one-line request feels faster than assembling inputs.
Context is the deal terms, the comp set, a prior example of the output you want, the actual lease PDF. A broker who writes “draft an LOI for a retail space” gets generic filler. A broker who pastes the deal terms, the landlord’s standard clauses, and last month’s executed LOI as a model gets a first draft they can send after light edits. The gap between those two outputs is entirely context, not phrasing.
How to practice: before every prompt, ask “what would a new analyst need to do this task?” and paste that in. Failure mode: treating the model as if it already knows your deal, your market, and your house style. It knows none of them until you supply them.
Skill 2: Role and format framing
The skill: telling the model who to act as and exactly what shape the output should take. This is where R.O.D.E.S.-style templates earn their keep, and where most published “prompt tips” stop.
Framing has two moves. The role sets the lens (“act as a commercial leasing broker reviewing this for a landlord client”). The format sets the container (“return a one-page screen with a numbered risk list and a table of the key economic terms”). For a small firm, the format half compounds fastest: define your lease-summary template once and every teammate reuses it, so outputs stop looking like ten different people wrote them.
How to practice: keep a shared document of your firm’s output formats and paste the relevant one into the prompt. Failure mode: vague asks like “summarize this lease,” which invite the model to guess at length, order, and emphasis, and to guess differently every time.
Skill 3: Iterative steering
The skill: treating the first output as a draft to correct, not a verdict to accept or reject. The value usually lives in the second and third prompt, not the first.
Steering is specific redirection: “tighten the rent escalation paragraph and cite the clause number,” “you missed the CAM reconciliation terms in section 7.” People who one-shot a prompt, get an 80%-right answer, and rewrite the last 20% by hand are leaving the tool’s best work on the table. A precise follow-up is faster than manual editing.
How to practice: commit to at least one correction on every task before you touch the output yourself. Failure mode: the binary reflex — accept it whole or throw it out — which wastes the model’s ability to revise on command.
Skill 4: Verification
The skill: checking every fact, figure, and clause reference against a source before the output leaves the building. In commercial real estate this is not a nicety. It is the skill that makes all the others safe to use.
Language models produce fluent, confident text that can contain an invented comp, a wrong cap rate, a misremembered square footage, or a clause number that does not exist. The cost of one of those reaching a client or an investor is not a typo; it is credibility, and sometimes liability. Verification is a distinct, teachable habit: the acquisitions analyst checks every number in a generated summary against the source rent roll, the property manager checks each clause reference against the actual lease. A useful summary of what “trained staff” actually means puts this plainly — fluency is as much about catching the model’s errors as producing its wins.
How to practice: end every task by tracing each number and citation back to a document. Make it the last step, every time, until it is reflex. Failure mode: trusting fluent prose. Confident and correct are unrelated properties in a generated draft.
Skill 5: Workflow chaining
The skill: using one output as the input to the next, and recognizing when a repeated chat sequence has become a workflow worth building. This is the bridge from personal productivity to firm capability.
Chaining looks like this: Tuesday’s lease summary becomes the source for Wednesday’s tenant email, which becomes the basis for the CAM notice. The person doing it stops treating AI output as a finished product and starts treating it as working material. The higher-order version is pattern recognition: once you have run the same three-prompt sequence on forty leases, you are no longer prompting, you are running an unbuilt pipeline. That recognition is the on-ramp to automation; until then, chaining by hand is where a lean team gets outsized throughput.
How to practice: notice when you repeat a sequence and log it. Failure mode: running the same manual chain hundreds of times without ever flagging it as a candidate to systematize.
Mapping the skills to CRE roles
All five skills matter for everyone, but the weighting differs by seat. A 10-person firm is really three or four small teams sharing an office, each leaning on a different subset.
| Role | Highest-impact skills | Anchoring deliverable |
|---|---|---|
| Brokerage (sales/leasing) | Context loading, format framing | LOIs, property narratives, market write-ups |
| Property management | Format framing, verification | Lease summaries, tenant communications |
| Acquisitions / investment | Context loading, verification | Deal screens, OM analysis, investor updates |
| Ops / admin | Format framing, chaining | Meeting notes to action items, CRM hygiene |
The pattern is worth naming. Client-facing producers live on context and framing because their output is persuasive writing built from deal specifics. Anyone touching numbers or legal language lives on verification because their errors are expensive. Ops staff live on chaining because repetitive work rewards systematization first.
A proficiency scale you can measure
A framework you cannot measure is a slogan. Score each person on each skill against a plain three-level scale, at the start of training and again ninety days later. The point is not precision; it is a shared language for “getting better.”
| Skill | Novice | Competent | Fluent |
|---|---|---|---|
| Context loading | One-line prompts, generic output | Pastes deal terms and an example | Assembles full context by reflex; output is send-ready |
| Format framing | Accepts whatever shape comes back | Uses a saved firm template | Adapts the template per audience without losing consistency |
| Iterative steering | One-shots, then edits by hand | Gives one or two corrections | Steers precisely; rarely rewrites manually |
| Verification | Trusts the draft | Spot-checks obvious numbers | Traces every figure and citation to source, every time |
| Workflow chaining | One task at a time | Occasionally reuses an output | Chains routinely; flags repeat sequences for automation |
Fluency is not scoring “fluent” on all five. For most people it is competent-to-fluent on the two skills their role leans on, plus non-negotiable fluency on verification if they touch numbers or contracts. Define your firm’s bar per role, then measure against it.
How to build these skills without a course
You do not need a curriculum to teach this. The most reliable method is weekly drilling on your own live work, which is what the firm-wide 90-day training program is built to deliver. Each session takes one real task, has everyone attempt it, then compares outputs out loud. The gap between the best and worst result in the room is almost always a gap in one of the five skills, which makes the lesson concrete.
Sequence the drills to the framework. Spend the early weeks on context loading and format framing, because they produce the fastest visible wins. Layer in steering once people stop one-shotting. Make verification the closing step of every drill from week one so it becomes muscle memory. Introduce chaining last, once individual tasks are solid, because chaining a shaky skill just compounds the shakiness.
Two adoption realities decide whether this sticks. The reps have to be on genuine deals, not toy exercises, or senior producers tune out. And the rollout is a social problem before a technical one: the ten rules for rolling out AI to a skeptical team cover the buy-in mechanics, and the lessons from a dozen AI workshops cover what non-technical teams actually retain versus what washes off by the next week. Retention is far higher when the skill is tied to a document someone had to produce anyway.
What the framework deliberately leaves out
A framework is as much about exclusions as inclusions, and this one deliberately omits much of what passes for “advanced prompting” online.
It leaves out prompt-injection tricks, jailbreak techniques, and elaborate role-play scaffolds — a CRE firm drafting LOIs and abstracting leases needs none of them. It leaves out chasing the newest model each month; the current generation of ChatGPT, Claude, Gemini, and Microsoft Copilot are all more than capable for CRE language work, and switching tools quarterly resets your team’s muscle memory for no gain. And it leaves out treating prompting as a substitute for judgment. The framework makes drafting and review faster; it does not make pricing calls, read a counterparty, or hold a relationship. Those stay human, and get more of your team’s hours once the drafting load drops.
Where a workshop fits
Everything here is runnable self-serve; this framework is written so a motivated principal can teach it from the weekly-drill method above. The honest case for outside help is speed: an experienced facilitator knows which failure modes hit CRE document types and compresses the trial and error. Market pricing for hands-on AI team workshops generally runs $2K–15K depending on depth and team size, against which the fair comparison is the billable hours your principals would spend designing the drills themselves.
For transparency, since this framework is the method we teach: SFAI Labs runs LLM fluency workshops for small CRE teams, scoped to prompting applied to your firm’s real work — LOIs, lease summaries, market write-ups, and email handling, with your documents and your people in the room. That is the whole of the workshop; custom automation and written AI policies are separate projects, not part of it.
If you want a read on where your team’s five skills stand today, and whether training, automation, or neither is the right next move, the free AI-readiness assessment is the place to start.
Book your free AI-readiness assessment →
FAQ
What is the prompt skills framework for CRE professionals?
It is a way of teaching AI prompting as five separate, buildable skills rather than one formula: context loading (giving the model the right source material), role and format framing (setting the lens and output shape), iterative steering (correcting the first draft), verification (checking every fact against a source), and workflow chaining (using one output as the next input). Each maps to a real CRE deliverable like an LOI or lease summary, and each is practiced and measured on its own.
How is this different from the R.O.D.E.S. prompt framework?
R.O.D.E.S. (Role, Objective, Details, Examples, Sense-check) is a strong template for writing a single prompt, and it maps neatly onto the second skill here, format framing. The difference is scope: R.O.D.E.S. optimizes one prompt, while a skills framework covers the full competency — supplying context, steering across multiple turns, verifying output, and chaining tasks into workflows. Use a template like R.O.D.E.S. as the on-ramp to skill two; treat the other four skills as the rest of the training.
Which prompt skill should a CRE team learn first?
Context loading. It is the highest-impact skill and the one people skip most. Most disappointing AI output in a real estate firm traces back to a prompt that asked for something without giving the model the deal terms, the comp set, or an example of the desired result. Teams that fix only this one habit see the biggest single jump in output quality before they touch any other technique.
How do we know if our prompting skills are actually improving?
Score each person on a three-level scale (novice, competent, fluent) for each of the five skills, once at the start and again after about ninety days of weekly practice. Pair that with a time comparison on a fixed set of real tasks measured on day one and day ninety. If verification is not at fluent for anyone touching numbers or contracts, the training is not done regardless of how the other scores look.
Do CRE professionals need technical skills to learn prompting?
No. Prompting is a language and judgment skill, not a coding skill. It rewards the abilities good brokers, property managers, and analysts already have: knowing what a good LOI looks like, spotting a number that seems off, understanding what a client needs to read. The one genuinely new habit is verification discipline, and that is closer to proofreading than to programming.
How long does it take to build these five skills?
Individual basics come within two weeks of deliberate practice. Firm-level fluency, where the whole team defaults to AI-first drafting and everyone verifies reliably, is a habit-formation timeline of roughly ninety days. Verification and chaining take longest — the first fights the instinct to trust fluent text, the second needs the earlier skills solid before it works.
Is it safe to put confidential deal data into these tools while practicing?
On business-tier plans, the major providers state that your inputs are not used to train their models by default; on free personal accounts, protections are weaker. Practice on a firm-wide business subscription, anonymize anything under NDA before pasting, and name one person to answer edge-case questions. Verify the current data-handling terms of the specific plan at purchase time, because terms change.
Which AI tool is best for building these skills?
Any one current general-purpose assistant — ChatGPT, Claude, or Gemini — on a firm-wide business plan. Pick one and standardize so the team shares prompts and lessons. Microsoft Copilot is worth considering if your firm lives in Outlook and Word. Avoid switching tools every few months; the skills transfer, but the shared prompt library and muscle memory do not.
Key takeaways
- Prompting is five skills, not one: context loading, role and format framing, iterative steering, verification, and workflow chaining. Templates like R.O.D.E.S. teach one of them.
- Skill matters more than access. The BCG-Harvard study found guided users far outperformed those simply handed the tool, with 40% higher quality output inside AI’s capable range.
- Context loading is the first and highest-impact skill; most disappointing output is a context problem, not a phrasing problem.
- Verification is the skill that makes the rest safe in CRE, where an invented number or misread clause carries real cost. Make it the closing step of every task.
- Weight the skills by role: producers on context and framing, number and contract handlers on verification, ops staff on chaining.
- Measure with a novice-competent-fluent scale per skill at day one and day ninety, and define your firm’s bar per role rather than demanding fluency in all five from everyone.
Arthur Wandzel
