To write site visit reports with AI without losing rigour, the method has four steps: capture well on site (notes, voice and one photo per issue), ask the AI for a draft based on the practice’s template with a rule never to make anything up, check it against a fixed list, and sign and send it the same day. The report stops being an hour’s job in the afternoon and becomes a review of a few minutes. AI drafts; the professional checks and signs.

This guide covers what a good report contains, how to capture information on site, three ways to get from notes to a draft (with an instruction ready to copy), what to check before signing, and the most common mistakes.

What a good site visit report contains

Every practice has its own format, but a good report usually records:

  • Project details: name, address or reference, client and report number.
  • Date, time and attendees, with their role on the project.
  • Progress, by area or by work package.
  • Issues, each with its location and, where possible, a photo.
  • Instructions given, each with who is responsible and by when.
  • Open items from previous visits: what’s been resolved and what’s still open.
  • Next visit and signatures.

The two parts that cause the most trouble later are instructions with no owner or deadline, and open items that get lost from one report to the next. A good template forces you to fill them in.

One important point: if an instruction has to be issued formally under the building contract or recorded in an official site book, the report doesn’t replace that. The report records the visit; formal instructions have their own place.

Step 1: capture well on site

The quality of the report depends on what you record during the visit, not on the tool. A few habits make the difference:

  • One photo per issue, taken so it’s clear where it is: a wide shot first, then the detail.
  • Say the location out loud if you record a voice note: “Level 2, stair B, temporary handrail missing.”
  • Note instructions with owner and deadline on the spot: “Contractor, by Monday.”
  • Copy measurements and references exactly, without rounding.
  • Go over open items from the last report before you leave site.

You don’t need to write well on site. You need everything to be there.

Step 2: from notes to draft

There are three ways to do it, from least to most automated.

With an assistant and the practice’s template

With a business plan for ChatGPT, Claude, Gemini or Copilot, you can save a fixed instruction and paste in the visit notes each time. This instruction works well; adapt it to your practice:

You write the site visit reports for our practice. Using the notes, voice transcript and photos I give you, write the report following exactly the structure and tone of the example report. Rules: don’t make up any data; if something needed isn’t in the notes, write [OUTSTANDING]. Copy measurements, references and names exactly. Separate progress, issues, instructions (each with owner and deadline) and next steps. Include the open items from the previous report and say whether they’re resolved or still open. At the end, list any questions you have about the notes.

Paste a real report from the practice below it, with personal data removed, as the example. Before uploading site notes with clients’ or workers’ names, check what you can upload and with which account in our guide on ChatGPT and data protection.

Advantage: you can start today. Drawback: each person copies, pastes and formats by hand, and the result depends on always using the same instruction. You’ll find more instructions like this in 15 AI prompts for architects and engineers.

With an automation

The notes, voice memos and photos the architect sends by WhatsApp or email go straight into a system that prepares the draft in the practice’s template, with photos in place, ready for review. The copying and pasting disappears, and every team member produces reports in the same format.

With a specialised product

NotaFiel, our product for site documentation, does exactly this: it learns from reports the practice has already signed, shows every piece of data for the professional to confirm before writing it, and delivers the report in the practice’s format, with the photos in place.

Step 3: check before signing

Checking is what turns a draft into a reliable report. A fixed list, always the same:

  1. Date, project and report number correct.
  2. Attendees complete, with their roles.
  3. Measurements and references identical to the notes.
  4. Every instruction has an owner and a deadline.
  5. Nothing that wasn’t in the notes. If a sentence sounds right but you don’t remember it, delete or check it.
  6. No unresolved [OUTSTANDING].
  7. Previous open items carried over correctly.
  8. Every photo matches the issue it illustrates.

With practice, this check takes a few minutes. And it’s exactly the step not to skip: an AI that is almost always right is still sometimes wrong.

Step 4: sign, send and file the same day

A report sent the same day carries more weight than one sent a week later: the contractor acts sooner and nobody argues about what was said. File it in the project folder, under the record of the works: at completion, it’ll be where it should be. We explain this in our guide on completion certificates and handover files.

Common mistakes

  • Asking for the report without an example. Without one of the practice’s reports as a model, AI writes generic text that has to be redone.
  • Not forbidding it to make things up. If you don’t say so, it fills gaps with whatever sounds likely.
  • Incomplete notes. AI doesn’t know what wasn’t recorded. No tool makes up for a badly recorded visit.
  • Mixing projects. One conversation per project, or a system that keeps projects apart, avoids carrying data from one to another.
  • Skimming the check. Measurements and deadlines are read one by one.

How to tell whether it works

Before you start, note how long a typical report takes, from getting back from site to sending it. After a month, compare the time, how many corrections the drafts needed and whether reports go out sooner. If checking time doesn’t drop, the problem is usually in the capture on site or in the instruction, not in the AI.

Frequently asked questions

Is a report written with AI valid?

Yes. A report’s validity depends on its content and on who signs it, not on the tool used to write it. That’s why the professional’s check and signature are essential.

Can I dictate the notes instead of typing them?

Yes, and on site it’s usually easier. A voice transcript works as input just like written notes. Say locations, measurements and owners out loud.

What if the AI gets a measurement wrong?

That’s why measurements are checked one by one against the notes. The instruction should ask for them to be copied exactly; even so, checking is your job.

Do I have to tell attendees I use AI?

The EU AI Act doesn’t require it for writing reports, but if you upload personal data to a tool, the GDPR applies. National rules can go further: in Italy, professionals must tell clients which AI tools they use. We explain this in our guide to the EU AI Act.

What about photos that show people?

If someone can be recognised, they contain personal data. Documenting an issue almost never needs it: frame or crop before uploading.