Turn today's photo and notes into a 5-line customer update

The job

You took a photo on the job and jotted some notes. The homeowner wants to know where things stand. This prompt gives you a 5-line text: what got done, what's next, what they need to decide, what's still unknown, and your sign-off. Anything you marked "maybe" stays a maybe.

Who it's for

• Remodelers, handymen and plumbers on jobs that run more than a day.
• Solo owners and small crews who text the homeowner at the end of the day.

The prompt

Start a new chat for each update. Attach today's photo first.

Prompt 1: the update text

I run a small home-service business. Turn today's job photo (attached) and my notes into a progress update text for my customer. Rules:
- 5 short lines max, plain words, no trade jargon.
- Line 1: what got done today. Line 2: what's next. Line 3: anything the customer needs to decide or do. Line 4: anything still unknown, said plainly as unknown. Line 5: sign-off with my first name and business.
- Only state facts that are in my notes or plainly visible in the photo. If my notes say "maybe", "check" or "not sure", keep it uncertain.
- No dates, prices, warranties, guarantees or promises that aren't in my notes.
- Greet the customer naturally using the name below.

Customer: [CUSTOMER NAME]
Me: [YOUR FIRST NAME] at [YOUR BUSINESS]
Notes:
[PASTE NOTES]

Paste your notes as written. Write "maybe" or "not sure" wherever you mean it; the prompt tells the AI to keep them.

Worked example

Input (fake)

Photo: an opened shower wall with a new valve in it (a free-license photo, not a real job).

Customer: Janelle
Me: Dev at Brightwater Plumbing
Notes:
shower valve swap. old valve out, new pressure-balance valve in, pex tied in, pressure tested ok, no leaks at test. added 2x blocking for backer board. tile patch: she has 6 spare tiles, MAY not be enough. if not, we find a match or do an accent strip, her choice. patch fri if tiles work. water back on, shower NOT usable till patch cures 24h after.

The traps: a "maybe" tile count, a Friday that depends on the tiles, and a shower that can't be used yet.

Output

ChatGPT · Thinking effort: High (model not shown) · plan not shown · tested 2026-10-29 · run 2, job C. AI output from fake data, verbatim; passed U1-U7.

Hi Janelle, the new shower valve and pipes are in, wood supports added, and the pressure test showed no leaks. Water is back on.
Next is the tile patch Friday if your spare tiles work. The shower can't be used until 24 hours after the patch.
If the tiles aren't enough, you'll need to choose between matching tile or an accent strip.
We don't know yet whether your six spare tiles will be enough.
Dev, Brightwater Plumbing

What we liked: 77 words, no "pex" or "blocking," and Friday keeps its "if." It says the shower is off-limits and asks for the one decision that's hers.

Tool scores

A pass means usable as-is or after one small edit. Each reply was judged on seven criteria (U1-U7, listed under How we tested). Two runs on 3 fake jobs, so 6 chats per tool. Claude had one run, so 3 chats.

ChatGPT doesn't show which model answers; tested on the High setting.

The main finding: in 21 replies, no tool used a dollar figure, priced the can light, dated the painter, called the crack fixed or promised the work would last. The misses were smaller but could still mislead. Gemini turned "patch fri if tiles work" into "patching the tile wall on Friday," both runs. ChatGPT and Gemini each softened "electrician needs to look" once, and Gemini turned "plaster mostly off" into "finished taking off the kitchen back wall plaster." "Lath," "furring strips" and "mud" came straight through from the notes. And 8 of 21 replies ran over 90 words, including all three of Claude's.

Claude's replies read the most naturally, and it flagged on its own that the notes said "thurs" on a Thursday. Its score rests on 3 chats, not 6.

How we tested

• Date: 2026-10-29, about 2:10 to 2:55 AM CT.
• Route: each tool's website, signed in, a new chat for every run, with the job photo attached.
• Tools, model labels and plans (as the apps showed them):
  • ChatGPT: Thinking effort: High. No plan label shown. ChatGPT doesn't show which model answers; tested on the High setting. The Thinking effort slider normally sits higher on our account; we moved it to High for these runs and moved it back afterward.
  • Claude: "Fable 5.1 High", mode "Auto", plan Pro. One run per job (3 chats) to save usage credits.
  • Copilot: "Auto", plan Personal (Upgrade button visible).
  • Gemini: "Flash". Plan not shown (Upgrade button visible).
• Runs: 3 fake jobs (drywall ceiling for a couple, kitchen wall demo, shower valve), 2 runs each; Claude 1 run.

A pass means usable as-is or after one small edit.

• Pass criteria. One small fix (like a trim) made a pass with a small edit. Two or more fixes made a fail.
  • U1: 5 lines or fewer, plain words, 90 words or less.
  • U2: every claim comes from the notes or is plainly in the photo; no invented work.
  • U3: "maybe," "check" and "not sure" items stay uncertain, and nothing unpriced becomes included work.
  • U4: no invented dates, prices, warranties, guarantees or durability promises.
  • U5: the customer's decision or open question is clearly flagged.
  • U6: right greeting and sign-off: owner's name, no [brackets], not the AI account's name.
  • U7: sendable with one regeneration or less.
• Checking: a script flagged dollar amounts, trigger words and word counts, then a person read every reply.
• Fake data only. All test data was made up. No real customers, jobs or prices were used.