Automating Property Viewing Reports with AI in n8n

A real estate agent runs through four viewings in one afternoon. Between each one, they drive, take calls, prepare for the next. By evening, they try to remember: who loved the open kitchen at 12 Lilac Street, who balked at the price of the 3-room flat on Foch Avenue, who asked to see the property again with their partner. Human memory isn't built to hold onto this level of detail over several hours — and it's exactly that level of detail that separates a follow-up that converts from a generic email the prospect ignores. This guide builds an n8n pipeline that captures the report on the spot, right after leaving the property, and automatically turns it into a personalized follow-up and a CRM update.

Why capture the information right away, not in the evening

The modern replication of Ebbinghaus's forgetting curve, published by Murre and Dros (2015), Replication and Analysis of Ebbinghaus' Forgetting Curve, in PLOS ONE, confirms with controlled data what every real estate agent notices empirically: most forgetting happens in the first few hours after an event, well before the end of the day. A report dictated "cold" in the evening, after three or four viewings blurred together, mechanically loses the nuances that matter — the tone of an objection, the facial expression in front of the bathroom, the exact phrasing about budget. The right moment to capture the information isn't at the end of the day in front of a CRM — it's in the car, right after leaving the property, while it's still fresh.

The architecture in one paragraph

The pipeline has five steps: the agent records a voice memo from their phone (a hosted n8n form or a simple file share) right after the viewing; this audio file is sent to the OpenAI node in transcription mode (whisper-1); the raw text goes through a Basic LLM Chain paired with a Structured Output Parser that extracts a structured JSON — property visited, prospect, positive points, objections, interest level, next step; this JSON feeds both a personalized follow-up email generated by a second LLM call and a CRM update; finally, a grouped digest recaps the day's viewings for the agency. If n8n's AI nodes are still new to you, our guide to the AI Agent node and chains covers the fundamentals needed for the rest.

Step 1 — Capture the voice memo from the field

Two realistic options for an agent on the move:

• n8n Form Trigger with a file field, accessible from a bookmarked link on the phone — no third-party app to install. Our guide to the Form Trigger and multi-step forms covers the full setup, including adding an optional text field (property address, prospect name) to make the downstream matching more reliable.
• Shared folder (Google Drive, OneDrive) that the phone's voice recorder app automatically syncs to — a "new file" trigger then fires the workflow with no extra action from the agent.

Either way, keep the file under 25 MB (OpenAI's transcription API limit) — a two- to three-minute memo in MP3 stays well under that, even for several viewings recorded back to back. Larger files (a video memo, for instance) follow the same rules covered in our guide to large files and binary data in n8n.

Step 2 — Transcribe with Whisper

The OpenAI node, Audio resource, Transcribe a Recording operation, with the language set explicitly rather than left on auto-detect for accuracy and latency — exactly the setup detailed in our guide to transcribing meetings with Whisper, transposed here from meeting minutes to viewing reports. Real estate vocabulary (energy performance diagnostic, shared wall, easement, co-ownership board) comes through correctly as long as the language is explicitly forced rather than left on auto-detection.

Step 3 — Structure the report with an LLM

The raw transcription text — often disjointed, with hesitations and restarts — goes through a Basic LLM Chain wired to a Structured Output Parser, forcing a usable JSON rather than a paragraph someone has to re-read manually. Our guide to the Structured Output Parser covers the mechanics; the schema for this use case looks like:

{
  "property": "12 Lilac Street, 3-room, 68 sqm",
  "prospect": "Mr. Dupont",
  "positive_points": ["natural light", "open kitchen", "close to transit"],
  "objections": ["price seen as €15,000-20,000 too high"],
  "interest_level": "high",
  "next_step": "send a counter-offer before Friday"
}

The interest_level field (a simple scale: low / medium / high) then lets you automatically route hot leads to an immediate alert for the agent, instead of letting them sit in the evening digest — the same routing principle detailed in our guide to qualifying inbound leads with AI.

Step 4 — Generate the follow-up and update the CRM

Two branches come out of this structured JSON:

• Follow-up email: a second LLM call, fed the JSON and a tone template (professional, warm, not overdone), drafts an email specific to this prospect and this viewing — not a fill-in-the-blank template where only the first name changes. The node then produces a draft rather than sending directly, following the same cautious principle detailed in our guide to AI-generated reply drafts: the AI prepares, the agent reviews and approves before sending, which prevents an awkward phrasing or a mistranscribed detail from going out unchecked.
• CRM update: a structured note (property, prospect, key points, interest level) gets pushed to HubSpot or Pipedrive via the API, following the setup detailed in our guide to syncing a CRM with n8n. The prospect's record reflects the viewing within minutes, without the agent needing to open the CRM themselves at the end of the day.

Step 5 — The end-of-day digest

A Schedule Trigger in the late afternoon queries the day's reports and sends a grouped summary over Slack or email to the agency: number of viewings, hot leads to prioritize, recurring objections coming up on the same property (often the signal that it's the price or the listing that needs adjusting, not the viewings themselves that are at fault). It's the same grouped-digest pattern used in the daily digest workflow from the Inbox AI Pack, applied here to viewings instead of incoming emails.

Common pitfalls

• Leaving the memo for the evening. As the forgetting curve cited above shows, a report dictated several hours after the viewing loses the nuances that make a follow-up effective. A reminder scheduled one hour after each viewing slot (via the synced calendar) reduces this risk.
• Sending the follow-up without review. An imperfect transcription (a misheard name, a truncated address) can slip into the final email. Routing through a draft rather than a direct send remains the simplest safeguard.
• Not specifying the language on the transcription node. Auto-detection works, but less reliably on technical real estate vocabulary and regional accents — forcing the language explicitly noticeably improves reliability, at no extra cost.

Going further

This pipeline — transcription, AI structuring, follow-up draft, and logging — reuses building blocks already packaged on the FlowKit side: the Inbox AI Pack (€79) for AI draft generation and the daily digest, and the Compliance & Audit Pack (€149) if you also need to keep a timestamped audit trail of every exchange with a prospect or client — for instance under a listing mandate. The Complete FlowKit Bundle (€269) brings all of it together with the RAG Assistant Pack, useful if you also want your mandates and diagnostics to be searchable through an internal chatbot. If your agency is still running through viewings with nothing but a memory that fades as the day goes on, it's an afternoon of setup away from never losing the detail that would have closed the deal.