Construction defects tracker with ChatGPT: method and checks
Defect tracking with ChatGPT: the one-year parfait achèvement warranty period, and why a reported fix is not proof of resolution.
By Educasium
Install the ChatGPT skill: Construction defects tracker →
Six weeks after handover on a single-family home extension, the contractor reports having "come by to sort out" defect R12, paint that needs redoing in the hallway. No one has gone back to check on site. The client, keen to move on, asks the architect to update the register with ChatGPT and close the line. The question at stake is not one of formatting: it is whether a contractor's statement is enough to prove a defect has actually disappeared, and how long remains before the right to enforce it becomes harder to exercise.
This article details the legal period protecting every defect, the difference between a reported intervention and a proven resolution — the most common error in an otherwise well-kept register —, the method for updating this register with ChatGPT without ever turning a statement into proof, and what sets this ChatGPT version apart from the equivalent method already published for Claude.
Summary
- The period protecting every defect: the parfait achèvement warranty
- Reported intervention, resolved defect: two statuses ChatGPT must never confuse
- Updating the register with ChatGPT, step by step
- A table of the four possible statuses and their proof
- What sets this method apart from the Claude version
- When this register is not enough
- Training to make defect tracking more reliable
- Frequently asked questions
The period protecting every defect: the parfait achèvement warranty
A defect recorded as a reservation in the handover document benefits from a one-year period from handover for the contractor to fix it, under Article 1792-6 of the French Civil Code. This period also covers defects notified in writing after handover, not only those recorded on the day of the visit itself. The starting point of this period is never the end-of-site date announced by the contractor, nor the last invoice date: it is the handover date recorded in the document, a figure the register must reproduce exactly rather than infer.
What remains owed after the period expires
The one-year period expiring does not make the defect itself disappear: it simply makes claiming it more complex, and generally dependent on a case-by-case analysis, often with the help of a legal professional. That is one more reason never to let a "reported intervention" status silently turn into "resolved" for lack of active tracking, while the legal clock keeps running in the background.
Reported intervention, resolved defect: two statuses ChatGPT must never confuse
A contractor reporting that they intervened on a defect has not, by that fact alone, resolved it: resolution requires verification by someone other than the contractor, not a statement. A register generated by ChatGPT only reproduces this distinction if the user explicitly supplies it in their request — the tool can never guess that a sentence like "the contractor says they came by" does not constitute proof of correction.
Why this confusion is the costliest of all
Confusing a reported intervention with a proven resolution prematurely closes a file that, in reality, remains open: if the defect resurfaces later, the client discovers the register had shown it as resolved for months, with no inspection ever having taken place. This is the most common error in an otherwise well-kept register, precisely because it stays invisible until the defect comes back.
Updating the register with ChatGPT, step by step
Step 1: Gather the last approved register and the new, dated observations. Without this foundation, no before/after comparison is possible, and an already-resolved defect risks being reopened by mistake, or the reverse.
Step 2: Check every proposed status change against evidence, before asking ChatGPT for it. A contractor's statement is never, on its own, proof of resolution; a dated inspection, ideally documented with a photo, is.
Step 3: Confirm the actual handover date before any period calculation. The register must never recalculate a warranty deadline from an assumed end-of-site date: the only date that matters is the one recorded in the handover document.
Step 4: Ask ChatGPT to recalculate totals and the summary after every change, without rewriting the detail. A summary not built from formulas linked to the register's actual rows will silently drift away from the real detail over time.
Step 5: Circulate a dated version and archive the previous one separately. Keeping the version history allows reconstructing, in the event of a later disagreement, what was known and on what date — including what was requested from ChatGPT at each update.
A table of the four possible statuses and their proof
| Status | What it means | Evidence required before applying it |
|---|---|---|
| Open | Defect found, no action taken | None: the default status |
| Reported intervention | The contractor states they intervened | Nothing at this stage: the status stays "open" until the inspection happens |
| Resolved | The defect has been verified as fixed, on site, by someone other than the contractor | A dated joint inspection, ideally with a photo |
| To qualify (period expired) | The one-year warranty has expired for this specific defect | The actual handover date, never an estimate |
Fictional example for testing the method: R12 concerns lobby paint. The contractor reports a repair without a closure inspection. The correct request to ChatGPT is to add that report to the history and keep the defect open; a missing inspection date remains pending, never invented to fill an empty cell.
What sets this method apart from the Claude version
Educasium also publishes an equivalent method for tracking construction defects with Claude, built around a native skill installed once and for all. Both methods share the same central risk — confusing a reported intervention with a proven resolution — and the same legal foundation, since Article 1792-6 of the Civil Code does not change depending on the tool used.
The ChatGPT pack presented here combines naturally with the handover record pack, which produces an initial register already sharing the same defect IDs from handover onward. The register from this tracking pack then reuses that same workbook to update it visit after visit, never renumbering existing rows. As with the Claude version, a file already downloaded to your computer is not automatically available in a new conversation: it must be resupplied with its current data at every register update, whichever tool is used.
When this register is not enough
A properly kept register does not resolve every situation encountered after handover.
A persistent disagreement between the client and the contractor over whether an intervention actually took place — one asserting it happened, the other disputing it — is not resolved with a better-formatted table, but with a joint inspection or, failing agreement, an expert assessment. If the handover date itself is disputed — for example in the case of an unformalized tacit handover — the period's starting point must be clarified before labelling a defect "out of period," to avoid setting a deadline on an uncertain basis. A register prepared with AI assistance, ChatGPT like any other tool, remains in every case a tracking document: the decision to treat a defect as genuinely resolved always belongs to the professional who carried out or supervised the inspection, never to the tool that updated the table.
Training to make defect tracking more reliable
A well-designed register is not enough on its own: it also needs adapting to every site, over several months, without losing track of a single 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
Does a reported intervention from the contractor count as resolving the defect in the ChatGPT-generated register?
No: a contractor's statement that they intervened is never, on its own, proof the defect was corrected. Resolution requires a joint inspection carried out by someone other than the contractor, before the status can be changed in the register. In the meantime, the register must show an intermediate status — "reported intervention" rather than "resolved" — so as not to suggest the inspection has already taken place.
How long does a defect remain enforceable after handover?
The parfait achèvement warranty, set out in Article 1792-6 of the Civil Code, runs for one year from the handover date recorded in the document. After that period, the defect does not legally disappear, but enforcing it becomes more complex and generally depends on a case-by-case analysis. The register should therefore display, for every defect, the number of days remaining before the one-year deadline, rather than a bare handover date stripped of context.
Should a defect be reopened if it resurfaces after being resolved?
Yes, in most cases: a defect closed by mistake, or one that resurfaces after an incomplete fix, should be re-entered with its original ID rather than created as a new row, to preserve the traceability of its full history since handover. The status must then revert to "open" or "reported intervention" as appropriate, and must never keep showing as "resolved" once the recurrence is confirmed. The history of the first resolution must stay visible in the register, rather than being deleted, to document that the initial inspection was not sufficient.
Can the ChatGPT pack generate the initial register directly after handover?
Yes, via the combined handover record pack, which delivers the record and an initial register sharing the same defect IDs. This tracking pack then lets you update that same workbook, visit after visit, never renumbering existing rows. Both packs should therefore be used within the same ChatGPT project, so the IDs generated at handover stay consistent with those later carried into tracking.
Reliable defect tracking, whether prepared with ChatGPT or any other tool, rests on three constant elements: a verified handover date, since it fixes the start of the one-year legal period, a strict distinction between a reported intervention and a proven resolution, and a stable ID per defect that survives every update.
Before any defect is opened, our ChatGPT method for the handover record details how to produce the shared initial register, and our ChatGPT method for site meeting minutes revisits the document that traces decisions made in meetings.
Training 100% fundable via OPCO/FIFPL. Qualiopi-certified programme. To structure your defect tracking 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.