Water damage, a fire, a burglary. You're on site, you photograph the damage and you describe what you see at the moment you see it. KontexVoX engraves every description into the EXIF, with date, time and geolocation. Now you need to structure it all into a claim file for the insurer.

The problem with claim files isn't the photo. Everyone takes photos. It's memory: three weeks later, filling in the form, you no longer remember whether that crack was there before, whether it was the tap leak or the pipe, in which room exactly. The descriptions you dictated on site, though, don't forget.

Prerequisites

Before you start, you need:

1. Photos enriched with KontexVoX. The descriptions live in each file's EXIF Description field, with capture date and geolocation. See What is an EXIF file?.
2. An AI agent with file-system access: Claude Code, Codex CLI, or Gemini. Not a web chatbot.
3. The photos exported to your computer via AirDrop or the Files app (the channels that preserve EXIF — see How to share without losing the EXIF).
A claim file is several photos to cross-reference with their dates and geolocation. A web chatbot doesn't read your local files. To assemble the file, you need an agent working inside your folder.

What you dictated: an example

Here's what the EXIF of four photos taken during water damage contains:

IMG_1422.jpg → "Bathroom, south wall. Damp stain, 40 cm, behind the sink. Wallpaper peeling at the bottom." — 03/14/2026 09:42 — GPS: 48.3567, 2.4012 IMG_1423.jpg → "Bathroom, ceiling. Brown mark in the corner, water still beading." — 03/14/2026 09:43 IMG_1425.jpg → "Hallway, floor. Parquet warped over 2 meters from the bathroom. Hollow sound underfoot." — 03/14/2026 09:45 IMG_1428.jpg → "Kitchen, ceiling light. Water running from the fixture, I've cut the electricity." — 03/14/2026 09:48

These descriptions, you dictated while walking through the apartment, while the water was still running. KontexVoX transcribed and engraved them with timestamp and geolocation. Three weeks later, filling in the insurance form, you don't have to guess: it's all there, in order, with exact times.

The prompt

Read all the JPEG images in this folder. For each image, extract: the EXIF Description field, the capture date and time, the GPS geolocation. Generate a structured claim file in Markdown with: 1) Claim reference (to fill in), 2) Date and location (extracted from the EXIF), 3) List of observed damage (each item = one photo with description, timestamp, geolocation), 4) Visual severity estimate (low, moderate, severe), 5) Appendix: summary table of all photos with reference. The format must be submittable to an insurer as-is.

What the AI produces with this prompt

A claim file in Markdown, structured in five sections. Each damage item is documented with its reference photo, the exact description you dictated, and the timestamp and geolocation extracted from the EXIF. A severity estimate is assigned to each item. The summary table in the appendix lists every photo with its reference.

The report can go to your insurer as-is. Attach it to your online declaration, send it by email attachment (the channel that preserves the photos' EXIF), or print it for the mail. The metadata recorded by KontexVoX — date, time, place, description — gives weight to your declaration: these aren't context-free photos, they're timestamped evidence.

Recommended tool: Codex CLI

Codex CLI suits structured formats and normalized data. A claim file needs rigor in its structure, which is Codex's strong point.

Frequently asked questions

Does the insurer accept a Markdown file?

The insurer expects a readable, documented file, not a specific format. Markdown reads as plain text and exports to PDF in one click. If your insurer expects a specific format (PDF form, Word), ask the AI agent to generate the file directly in that format by adjusting the prompt.

Do EXIF metadata have evidential value?

EXIF metadata (date, time, geolocation) are technical information embedded in the file at the moment of capture. They don't replace an official report or appraisal, but they document the chain of evidence: when, where, and what you saw. In a dispute, original photos with intact metadata are stronger than photos without context.

Does it work if I have no signal on site?

Yes. KontexVoX engraves the description into the file locally, without a connection. Geolocation may be missing if GPS can't lock, but the description and timestamp are always there. That's actually the most common claim scenario: you're in a basement, an attic, somewhere with no signal.

Can I do the same thing without KontexVoX?

Technically yes — you can take photos and write the descriptions by hand afterwards. But the problem is memory: three weeks after the loss, in front of the insurance form, the details are gone. Voice engraves the context while it's fresh, while the water is still running. That's why I created KontexVoX: not for insurance, but so I'd never again forget what I saw.