Turn a Site-Visit Voice Memo Into a CRE Investment Summary with AI

You just walked the asset. Standing in the parking lot, you have a sharper read on it than any offering memorandum gave you: the roof that is two years from a replacement, the anchor tenant whose space looked half-dark, the corner that catches the morning traffic. The problem is that read lives in your head for the drive home and then evaporates. The fix is simple: record a 2 to 3 minute voice memo on the asset, transcribe it, and have AI structure your observations into the documents the deal actually needs.

This is a drafting and structuring job, not an analysis job. The AI works only from what you said into the recorder. It does not appraise the building, underwrite the deal, or invent a tenant roster you never mentioned. For the analysis layer that comes after, see underwriting multifamily deals with AI. To install the reusable skills that run this and the rest of the deal grind, the free 7 Claude Skills for Real Estate pack ships them ready to run.

Capture the memo while you walk

The whole method depends on a good raw recording, so talk the way you would brief a partner who could not make the visit. Walk a consistent loop and narrate five things: physical condition (roof, parking, mechanicals, the obvious deferred maintenance), tenancy and occupancy (who is in place, what looks dark, signs of recent move-outs), the location read (traffic, surrounding uses, access, what the block feels like), the value-add angle (what you would change and why), and the red flags (anything that made you hesitate). Two to three minutes is plenty. You are capturing observations, not writing prose, so ramble honestly and let the structuring happen later.

Transcribe it with whatever you already use: your phone's voice-to-text, a transcription app, or by uploading the audio straight into Claude or ChatGPT. The transcript is the only ground truth the AI gets. If you did not say it on the asset, it should not appear in the output, which is exactly what the next step enforces.

The grounding rule: no invented facts, flag every gap

This is the part that makes the workflow trustworthy, and it is one instruction you paste in front of the transcript every time: use only what is stated in this memo. Do not estimate square footage, cap rates, rents, year built, or any number I did not say. Where a document normally needs a fact I did not provide, insert a [CONFIRM: ...] tag describing what is missing instead of guessing. Left to its defaults, a language model will helpfully fill a blank with a plausible-sounding number, and a plausible-sounding number in an investment summary is worse than a blank one.

In practice the output comes back with tags like [CONFIRM: in-place rents and lease expirations] or [CONFIRM: actual year built and rentable SF] wherever your memo was silent. That is the feature, not a failure: the tags become your follow-up list. This no-invent discipline is the same backbone behind every reliable skill in the Claude Skills real estate investors actually use. The AI structures and drafts; it never fabricates the facts you did not observe.

Output one: the internal site-visit and investment summary

The primary deliverable is a clean internal write-up your investment committee or your partner can read in two minutes. Ask for a fixed structure: a one-line deal summary, property and location condition, tenancy and occupancy notes, the value-add thesis in two or three sentences, identified risks and red flags, and a facts-to-confirm block that simply collects every [CONFIRM: ...] tag in one place. Because it is built only from your narration, it reads like you wrote it, just organized.

This summary is also the seed for the next document in the deal. When a property clears your first read, the same observations feed straight into a first offer, and you can hand the summary to AI to draft the terms: see how to draft a CRE letter of intent with AI. If your memo mentions financials you were handed on site, run them through the Rent Roll and T-12 Normalizer before any number reaches the summary, so the in-place NOI is honest rather than the seller's headline.

Output two: the short OM-teaser blurb

From the same transcript, in the same pass, ask for a short marketing-flavored teaser: three or four sentences you could drop into a deal email, a partner pitch, or the top of your own one-pager. It leads with the strongest grounded selling point from your memo (the location, the value-add angle, the tenant story) and stays honest, because it still only draws on what you said. Where the teaser wants a number you did not give, it keeps the [CONFIRM: ...] tag rather than inventing a return figure to make the deal sound better.

The honest boundary matters most here: this is a drafting aid, not a representation of fact. Every claim in a teaser that goes to a third party is yours to verify before it leaves your hands. The AI gets you from a voice memo to a polished first draft in seconds; the diligence stays human.

Output three: the underwriting follow-up punch list

The third output turns the gaps into action. Ask the AI to read the memo and produce a punch list of the diligence items a real underwrite would need, split into what to request from the broker or seller (rent roll, trailing twelve, tax bills, service contracts), what to verify independently (zoning, true rentable area, deferred-maintenance bids), and what your own model needs as inputs. Every [CONFIRM: ...] tag from the summary lands here as a line item, so nothing you flagged on the asset slips through.

This is where a voice memo becomes a workflow rather than a note. The punch list points at the next real task, and for proactive investors that same instinct extends upstream into sourcing the deals worth walking in the first place: see AI owner-propensity scoring for off-market CRE.

Where to start

On your next site visit, record the 2 to 3 minute memo, paste the transcript under the grounding rule, and ask for all three outputs in one prompt: the investment summary, the teaser, and the punch list. Read the [CONFIRM: ...] tags first, because they are the AI telling you exactly where its draft stops and your diligence begins. Install the free 7 Claude Skills for Real Estate pack so the prompt is reusable instead of retyped each time.

When this becomes a habit across a team, the question shifts from whether AI can draft the document to how you wire it into your deal flow. That is the advisory work we do with investors and developers: mapping the voice-memo-to-summary loop, and the rest of the analyst grind, into workflows that run on your CRM and data room with human review on every deal-quality call.