The worked example
Cowboy Ridge, six photos through the engine
A real Roam Free property in Austin. Blue tags are what the vision pass recorded — only what's visibly verifiable. The green arrow line is the caption that ships.
Cowboy Ridge — gallery captions
62 photos · 62 captioned · pushed with receiptsThe method
Two passes, on purpose
The pass that looks is never the pass that sells. That separation is what keeps thousands of captions honest.
Record only what is visibly verifiable
- Objects, finishes, and layout actually in the frame
- Counts that can be counted — seats, beds, chairs
- No building-material guesses, no access details, no "probably"
Turn observations into a guest benefit
- Leads with what the guest gets to do there
- Uses amenity words guests — and AI search — actually query
- Cross-checked against the verified fact base before publish
Why it compounds
AI search reads captions
Assistants and channel ranking systems index caption text. "Full-size pool table" and "cold plunge" become findable answers instead of unlabeled pixels.
Nothing to walk back
Captions only describe what's visible and verified — no caption can overpromise, the #1 source of "listing didn't match" reviews.
Verified live, per photo
On the 123-unit engagement, every caption pushed carried a per-photo receipt: 4,147 of 4,148 matched the live gallery byte-for-byte.
Your titles stay yours
PM-authored photo titles are never touched. The engine adds; it doesn't overwrite your team's work.