Works acceptance record with ChatGPT: method and checks

Install the ChatGPT skill: Works acceptance record →

An architect drafts, inside ChatGPT, the handover record for the final site visit on a single-family home extension. The contractor announces the works are finished; two points are still visible on site: a door rubbing against the floor in the study (joinery trade) and a paint touch-up in the hallway that was never done. The client has not yet stated any decision on acceptance itself. ChatGPT can help format the handover record and its accompanying defects register; it cannot decide acceptance on the client's behalf, nor turn a technical observation into a legal agreement.

This article details exactly what the ChatGPT pack dedicated to this document produces — a Word handover record and an Excel defects register linked by the same IDs —, what the Civil Code says about handover and the three legal warranties that follow from it, the method for preparing both files without blurring findings and decisions, and what sets this ChatGPT version apart from the equivalent method already published for Claude.

Summary

1. What this pack produces: a linked record and register, not a stand-alone document
2. Handover: a legal act ChatGPT can never decide on your behalf
3. The three legal warranties that start on the handover date
4. Preparing the record and register with ChatGPT, step by step
5. What sets this method apart from the Claude version
6. When this pack is not enough
7. Training to make site handovers more reliable
8. Frequently asked questions

What this pack produces: a linked record and register, not a stand-alone document

The stated format for this skill, DOCX + XLSX, means the request produces two files in a single session: the draft handover record and the resulting defects register, sharing the same IDs across both documents. That is the most visible practical difference from how this kind of file is usually built trade by trade.

Why link both files from the first version

A signed handover record with no linked register forces, at the first defect update, a reconstruction of the correspondence between the paper document and a table built afterwards — work that takes longer than the table itself. Requesting both files together, with a stable defect ID (R01, R02…) repeated identically in the record and the register, avoids that reconstruction and lets each defect be tracked through to its resolution without losing the thread.

What the pack does not do

The pack decides nothing: it organizes observations supplied by the user, separating what was observed, what was communicated as a decision, and what remains to be confirmed. A defect with no precise location, or a decision not explicitly communicated, must stay flagged as such in both files, never filled in with a default value.

Handover: a legal act ChatGPT can never decide on your behalf

Handover of the works is the act by which the client — never the architect assisting them — declares acceptance of the works, with or without reservations, as set out in Article 1792-6 of the French Civil Code. No tool, ChatGPT included, can make that decision: it can only prepare the document that will record it once it has been explicitly made.

Acceptance with or without reservations: what actually changes

Acceptance without reservations, in principle, closes off the ability to claim under the parfait achèvement (making-good) warranty for a defect apparent on the day of the visit and not recorded in the handover document. Acceptance with reservations, conversely, preserves the client's right to demand that recorded defects be fixed. That is why the pack always asks for the decision and its date to be supplied separately from the findings: neither should ever be inferred from the other in the generated document.

The risk of tacit handover when the question stays open too long

A client who takes possession of the premises and starts occupying them without having formally pronounced handover may have that occupation interpreted as a tacit handover depending on the circumstances. That risk grows with the time that passes without a formalized decision — one more reason to prepare the record quickly after the visit, rather than leaving the file open while waiting for a client's answer.

The three legal warranties that start on the handover date

The handover date recorded in the document is never a mere administrative formality: it is the common starting point of three legal warranties, the longest of which protects the client for ten years.

Why a date error in the generated record is costly

All three periods run from the handover date recorded in the signed document, never from the end-of-site date announced by the contractor nor from the last invoice. An incorrect date in the file produced by ChatGPT — back-dated or post-dated, even by a few days — genuinely changes the three periods available to the client. It is a field to check line by line before signature, never reconstructed from memory once the document has been circulated.

Preparing the record and register with ChatGPT, step by step

Step 1: Gather inspection findings, labelled photos and decisions actually communicated. Without these, ChatGPT can only produce an empty template; the pack never replaces the site observation itself.

Step 2: Import the documents into a ChatGPT project together with the pack's instructions. Name files with their date and revision, and explicitly state which are approved and which are only templates carried over from an earlier site.

Step 3: Request the record and register in a single session, with shared IDs. A defect must carry the same number in both files from the first version, so later updates remain traceable.

Step 4: Have the acceptance decision and its date explicitly confirmed, separately from the findings. That decision belongs to the client alone; it is never inferred from the visit itself, nor from a general impression of how the site is progressing.

Step 5: Open both files, check their cross-consistency, then send them for signature. Verify that every defect in the register exists in the record with the same location wording, and that no field has been filled with a plausible value that was never supplied.

Fictional example for testing the method: a door rubs against the floor in the study (joinery trade), photo P03, with no client decision or repair date supplied at this stage. The correct request to ChatGPT is to prepare the finding and the draft record without marking acceptance as approved, leaving the decision and its date open in both files — never filling them in with a plausible guess.

What sets this method apart from the Claude version

Educasium also publishes an equivalent method for preparing a handover record with Claude, which covers the legal framework of handover at greater length. Both methods share the same legal foundation — the Civil Code does not change depending on the tool used — but differ on two practical points.

The first concerns the scope of the deliverable: the Claude skill dedicated to handover produces only the record, while the defects register belongs to a separate skill, defect tracking, installed and invoked independently. The ChatGPT pack presented here combines both outputs in a single request, which simplifies the first version of the file but means checking two files instead of one before circulation.

The second concerns installation and continuity across sessions. Claude offers a native toggle to enable a skill once and for all; ChatGPT, in this pack, relies on a project to which the instruction files are attached. In both cases, a file already downloaded to your computer is not automatically available in a new conversation: it must be resupplied with the site's current data, whichever tool is used.

When this pack is not enough

A properly prepared record does not cover every situation encountered after a final site visit.

A disagreement over the very principle of handover — the client considering the works unfinished, the contractor considering them compliant — is not resolved with a better-worded document: it calls for a negotiated agreement between the parties or, failing that, judicial handover, a route beyond what a document-generation pack can prepare. A site involving several clients — a co-ownership body, a self-build collective — requires organizing the visit and signature in a way suited to several decision-makers, best checked with a legal professional before the visit. A record prepared with AI assistance, ChatGPT like any other tool, remains in every case a draft: it must be reviewed and approved by the client before signature, and referred to a legal professional as soon as the situation departs from a standard case, before any final issuance or signature.

Training to make site handovers more reliable

A well-designed pack is not enough on its own: it also needs adapting to every site without losing time or missing a defect. For an architect practising as a self-employed professional (NAF code 7111Z in France), the FIFPL fund covers part of the cost of Qualiopi-certified training under 2026 criteria set at €300 per day and €900 per year, with e-learning capped at 50% of the daily rate — funding that must be arranged before training starts, not after.

Frequently asked questions

Can ChatGPT decide whether handover should be granted with or without reservations?

No: that decision belongs exclusively to the client. ChatGPT can prepare the document that will record it once explicitly communicated, but must never infer accepted or refused handover from the visit's findings alone. If that decision has not yet been made at drafting time, the draft record must stay explicitly marked as provisional until the client communicates it.

Does the pack generate the defect-tracking register directly after handover?

Yes, in the same session: this pack's DOCX + XLSX format combines the record and the defects register, with shared IDs from the first version. That register must then be updated at every follow-up visit; a repair reported by the contractor is never proof of resolution until a joint inspection has taken place. For that ongoing tracking, our dedicated defect-tracking method details how to update that same workbook visit after visit, without ever renumbering rows already created.

What happens if a defect is not fixed one year after handover?

The parfait achèvement warranty, set out in Article 1792-6 of the Civil Code, runs for one year from handover. A defect not fixed within that period does not automatically disappear, but enforcing it becomes more complex and generally depends on a case-by-case analysis, often with the help of a legal professional. In the meantime, the register must keep displaying that defect as open, with its original handover date, rather than being closed to tidy up the document.

Must the ChatGPT pack be reinstalled for every new site handover?

No, as long as the ChatGPT project stays active with its instruction files: what needs updating is the content specific to each site — findings, photos, communicated decisions — never the pack itself. Keep a separate ChatGPT project per site, however, so a finding from an earlier handover does not creep into the current document. If the project must be recreated regardless, re-import only the documents for that specific handover before generating the file.

A reliable handover record, whether prepared with ChatGPT or any other tool, rests on three constant elements: a decision explicitly made by the client and never inferred from the visit alone, located defects linked by a stable ID between the record and the register, and an accurate handover date, since it fixes the starting point of the three legal warranties.

Before handover, our ChatGPT method for DCE consistency checking helps secure contract documents ahead of the site, and after the visit, our ChatGPT method for defect tracking details how to trace every defect through to resolution.

Training 100% fundable via OPCO/FIFPL. Qualiopi-certified programme. To structure your site handovers with AI as part of our AI for architectural management training, contact Educasium and specify your status (employee, self-employed, business owner) and your goal. This pack is also available on its dedicated ChatGPT skill page.