Jobsite Photo Field Note: Minimum Facts Before an AI Documentation Report

Why a photo field note must exist before the AI report

A routine progress email can lean on the work order and the crew schedule. A photo documentation report leans on shot purpose, capture time, what is in frame, what must stay out of the prompt, and who decides the customer-facing or insurer-facing position. Treating a phone dump like a marketing gallery is how offices send a polished AI summary that claims "pre-existing damage documented on day one" when the note only supports a blurry hallway shot with no timestamp and no named purpose.

Read the job pointer and the shot list before anyone prompts a model. Classify the report purpose: progress documentation, change-order support, pre-existing condition baseline, punch-list evidence, billing support, warranty visit record, insurance packet draft, internal QC, or other office-defined purpose. Many packs contain two purposes. Write both down. The purpose decides which facts must be real and which claims the owner must approve.

Treat the AI as a writer working from a field note and a curated photo set, not as a site witness who "was there." It can organize verified captions, keep tone steady, and list gaps. It cannot know what was behind the camera, whether a shadow is a crack, or whether the owner wants a soft internal summary versus a firm customer-facing evidence letter.

Photos without a field note also create privacy risk. Faces, children, neighbors, license plates, payment cards, medical documents on a kitchen table, and Wi-Fi passwords on a router sticker are common accidental captures. The field note's withhold list is what keeps those out of the prompt and out of the draft.

The minimum facts to collect first

Before prompt time, fill each field or write UNKNOWN with the person who can resolve it:

• Job pointer — job or work-order ID, site address shorthand on file, and customer or property label the office already uses.
• Report purpose — progress, change-order support, baseline/pre-existing, punch list, billing support, warranty, insurance packet draft, internal QC, or other. Quote the office language, not marketing language.
• Photo set pointer — folder name, album link the office already controls, or file count and device source. Do not invent a cloud URL that does not exist.
• Capture window — date and approximate time range for the set, or per-shot timestamps if the office already has them. Mark UNKNOWN rather than guessing from memory.
• Shot purpose list — for each included photo or photo group: what it is meant to show (north elevation before demo, wet spot at baseboard after rain, panel label close-up, material delivery ticket). Purpose first; aesthetics never.
• In-frame facts (verified) — what a reviewer can actually see without interpretation. Separate "stain on ceiling tile" from "roof leak from flashing."
• Out-of-frame / UNKNOWN — what the photo does not show and must not be invented (behind cabinets, attic not accessed, neighbor property).
• Withhold / redact list — faces, children, plates, cards, documents, passwords, unrelated household items, or anything the office policy keeps out of AI prompts and customer packets.
• Related baseline — prior photo set, scope pointer, change-order ID, warranty claim ID, or UNKNOWN if none.
• Office position — internal QC only, customer progress update, change-order evidence draft, billing support memo, insurance packet draft pending owner review, or not yet decided.
• Promises and next steps — re-shoot needed, customer access ask, or DO NOT PROMISE items (cause certainty, claim approval, fixed repair dates).
• Approval owner — who must approve customer language, insurer language, and any fee or concession wording that the photos are used to support.
• Channel and timing — how the report or update will be sent and when a reply is reasonably expected.
• PII check — confirm the prompt excludes payment card data, full account numbers, government IDs, and anything not needed for this documentation purpose.

Office position is the field that prevents accidental guarantees. If the owner has not decided whether this is an internal QC note or a customer-facing evidence letter, a model will often fill the gap with polished certainty. That sentence can create a liability implication nobody authorized. An internal summary that lists verified in-frame facts and marks cause as UNKNOWN is safer than a dramatic "we proved the leak started before we arrived" line.

Shot purpose beats file name. A phone dump named IMG_4821.jpg tells the model nothing useful. A one-line purpose ("kitchen ceiling stain, looking up from island, after Saturday rain") is what keeps the draft inside the evidence.

An illustrative example

Three gaps still matter even in a "simple" progress pack: EXIF times are not exported yet, stain cause is unknown, and a change-order conversation may start later. The safe AI draft at this stage (1) names the job pointer and capture window as recorded, (2) describes only verified in-frame facts for the three selected shots, (3) states that attic and roof were not photographed, and (4) avoids cause, claim, and completion promises. That draft makes no insurance conclusion and no invented day-1 comparison unless the approver explicitly opens the baseline album and authorizes the comparison language.

Once the owner keeps the progress-update position and the superintendent confirms the three shot purposes, the field note stays stable for send. If the customer replies asking for a price to "fix the leak," stop using this sheet as a courtesy progress template and open the matching change-order or scope note before the next AI draft. If they dispute an invoice that already billed related waterproofing, open the invoice dispute note instead.

A short sequence before anyone prompts the model

1. Name the job and the report purpose — one sentence the approver can hear without opening the album.
2. Curate before you prompt — select the minimum photos that serve the purpose; reject the rest into a non-prompt folder.
3. Apply the withhold list — redact or exclude faces, plates, cards, documents, and anything policy keeps out.
4. Write shot purposes and verified in-frame facts — purpose first; interpretation last; mark cause UNKNOWN when it is.
5. Fill office position and DO NOT PROMISE — if position is undecided, draft only an internal checklist, not a customer letter.
6. Prompt with the field note only — attach or describe the curated set; do not paste the entire phone dump.
7. Review by role — field lead checks in-frame accuracy; PM or office checks job pointer and related IDs; owner checks position and promises; sender checks recipients and attachments.
8. Save the packet — field note, selected photo list, draft, edits, and sent message with the job file.

This sequence is short on purpose. Most photo-report failures are not writing failures; they are curation and position failures. A beautiful draft from an uncurated dump is still a liability.

Prompt rules that keep the model inside the note

When the note is complete enough to draft, give the model rules as strict as the facts:

• Use only the field note and the curated photo set. If a fact is UNKNOWN, say the office is checking it; do not fill the gap.
• Describe only verified in-frame facts; do not invent defects, measurements, code violations, or causes the note does not state.
• Do not invent timestamps, weather claims, prior-condition conclusions, or "standard industry" findings.
• If promises are DO NOT PROMISE, omit them entirely—even in a P.S.
• If office position is "not yet decided," draft only an internal holding summary or a non-sent checklist, not a customer or insurer email.
• Do not identify people in photos, speculate about neighbors, or include withheld items "for context."
• If a change-order, dispute, or warranty sheet is the real next step, say so and stop expanding this progress draft into pricing or blame.
• Keep length limited; one clear purpose; one named contact; one reply path.
• After the draft, list every sentence that depends on a field a human must verify against the album and job file.

That final checklist turns review into matching claims to photos and the field note, instead of rereading a smooth report for vibes.

Common failure patterns

The first failure is the invented defect: the model adds "clear water intrusion at flashing" because documentation reports often sound that way. If the note only supports a ceiling stain, the draft may only say ceiling stain.

The second failure is timestamp fiction: asserting a capture day or sequence the office never exported. Mark EXIF UNKNOWN until someone pulls it.

The third failure is prompt oversharing. Dumping fifty unredacted photos into a chat to "let the model pick" parks faces, plates, and household details in a log the office may not control. Curate first. The FTC's protecting personal information guide and Start with Security are practical baselines: keep only what the task needs, limit access, and avoid parking sensitive customer data in chat logs.

The fourth failure is purpose creep, where a progress pack silently becomes an insurance argument or a change-order accusation without a matching sheet and approver. Keep office position explicit.

The fifth failure is skipping related process sheets. Photos that prove extra work still need a change order field note before pricing language. Photos on a warranty visit still need the warranty service visit field note before customer promises. After-hours intake is still a callback note problem, not a photo-report problem.

Finally, keep the record. Save the field note, selected photo list, withhold decisions, draft, reviewer edits, and sent message with the job file. If the account escalates into a dispute, claim, or complaint later, that packet shows what the business knew when it documented the site.

How this note connects to related field guides

Use this sheet when the immediate job is to document what the camera shows before an AI writes. Use the AI jobsite photo documentation guide for the wider workflow shape. When photos are evidence for unpaid scope, open change-order or scope sheets before the model writes money language. When photos sit inside a warranty visit, invoice dispute, or estimate objection, open those notes for position and approver fields instead of stretching a progress caption into a legal conclusion.

The habit across Contractor AI Hub field notes is the same: collect minimum facts, mark UNKNOWN, name the approver, then draft. Photos do not relax that habit; they make it more important, because a confident caption under a wrong picture travels farther than a wrong paragraph alone.

A copyable one-page template

Keep the filled sheet with the job file beside the selected photo list. A one-line post-send note is enough when the report was clean; when it was not, the note should say which field was incomplete rather than which employee was "bad at photos."

Assumptions and limits

All examples are hypothetical. Photo retention rules, insurance practices, consumer protection laws, licensing claims, and contract language vary by jurisdiction and by your agreements; this page does not interpret them and is not legal, insurance, engineering, or financial advice. Do not use AI drafts to invent defects, causes, claim outcomes, or fixed calendar results. If a customer mentions a lawyer, a regulator, a formal insurance dispute, or a formal complaint, stop casual AI drafting and follow your advisor's guidance. AI vendors differ in data retention; review their terms before pasting customer photos and contact data. This guide claims no measured claim-approval, collection, or satisfaction improvement.

Sources and method

• FTC: protecting personal information — retrieved 2026-11-07
• FTC: start with security — retrieved 2026-11-07

AI-assisted writing with automated source, calculation and editorial checks. No hands-on testing or professional certification is claimed. Check the stated assumptions before applying this guide.