A file before the conseil de prud'hommes (France's labour court) plays out as much on the valuation of the claims as on the argument, and that is precisely where tools differ. A generalist AI writes a convincing paragraph but can neither reconstruct three years of overtime back-pay nor place a compensation figure within the range of the barème Macron (France's statutory scale capping compensation for dismissal without real and serious cause). The right selection criterion is therefore not the quality of the prose: it is the ability to produce accurate, traceable amounts.
Before comparing tools, look at the work. An ordinary conseil de prud'hommes file means gathering, head by head, sums that are not calculated the same way:
| Claim | What it takes to value it |
|---|---|
| Overtime back-pay | The week-by-week count, with the applicable uplifts |
| Compensation for dismissal without real and serious cause | Length of service, reference salary, position within the barème Macron range |
| Statutory or contractual dismissal compensation | Length of service and the applicable collective agreement |
| Compensation in lieu of notice and paid leave | The notice period and accrued leave |
| Miscellaneous back-pay | Bonuses, compensatory rest, reclassification |
Each of these sums depends on a reference salary and a length of service that must first be established from the file exhibits: contract, payslips, dismissal letter, amendments. It is a document-reading job before it is a calculation job, and it is the part that takes the most time.
There is also a timing constraint. Actions relating to the termination of the contract are time-barred twelve months from notification of the dismissal (article L1471-1 of the Labour Code), whereas the action to recover salary is time-barred after three years (article L3245-1). Two distinct periods, on the same file, with immediate consequences for the scope of the claims that can be valued.
1. Does the tool calculate, or only write? This is the main dividing line. A generalist language model produces plausible text; it does not run a reliable count of three years of back-pay, week by week, with variable uplift rates. Asking a tool built to write for a calculation is the first source of error.
2. Is each amount traceable? A figure with no supporting exhibit is unusable in submissions. The tool must be able to indicate which payslip, which amendment or which clause each retained element comes from. This is also what allows the lawyer to check it, who remains the one who decides and signs.
3. Does it absorb the whole file? A dispute with three years of payslips, work schedules and email exchanges quickly runs to several hundred exhibits. Generalist assistants cap the number of documents handled in a single analysis, which forces you to split the file and lose the overall coherence.
4. What happens to the data? The consultations a lawyer sends to their client, the correspondence with the client and, more generally, the file are covered by professional secrecy (article 66-5 of law no. 71-1130 of 31 December 1971). Loading a file's exhibits into a consumer tool, with no data-processing contract and no guarantee on the location and reuse of the data, exposes you to a risk that is not only commercial.
5. Can the tool place an amount, not just calculate it? The barème Macron (article L1235-3) gives a range between a floor and a ceiling, not an amount. Knowing where a file sits within that range means comparing it with decisions handed down on close profiles. A calculator that displays the statutory range only copies out the Labour Code.
The hallucinated reference. A generalist AI can produce an appeal number, a date and a holding that are perfectly credible for a decision that does not exist. On a conseil de prud'hommes file, this shows at the hearing. The safeguard is not to trust, it is to demand the source.
The non-reproducible calculation. If the same file, submitted twice, gives two different totals, neither can be produced. A valuation tool must be deterministic on the calculation part.
Confusing the limitation periods. Many valuation errors come from a salary claim calculated over a longer period than the limitation allows, or from a challenge to the termination brought out of time.
The scale applied where it does not apply. The cap of the barème Macron (article L1235-3) is set aside where the dismissal is void. A tool that mechanically applies the range without raising the question of nullity underestimates the file.
The most reliable method is to take a file already argued, whose outcome you know, and submit it to the tool. Three questions are enough to decide: is the total correct, is each line tied to an exhibit, and is the time saved real once the checking is done? A tool that produces a valuation you have to recalculate entirely does not save time.
Plato is built for this part of the work: reading the exhibits, calculating the back-pay and the compensation, placing the amount from comparable decisions, each figure staying traceable back to the exhibit it comes from. The tool prepares, the lawyer decides and signs.
For the detail of the calculation methods, see our articles on calculating overtime back-pay, the barème Macron 2026 and estimating compensation for dismissal without real and serious cause. For the overview of tools, see which is the best legal AI for lawyers.
Which AI tool should you use for a conseil de prud'hommes file?
One that can value the claims from the exhibits, not just write. The decisive criteria are calculation, traceability of each amount, the volume of documents handled, data protection, and the ability to place an amount within the barème Macron range.
Can ChatGPT prepare a conseil de prud'hommes file?
It can help structure a line of reasoning or a template, but it is not built to produce a reliable count or to cite verifiable decisions. The risk of hallucinated references and the professional-secrecy question make it a poor support for a file's exhibits.
Can an AI calculate overtime back-pay?
A specialised tool can: it reconstructs the count week by week with the applicable uplifts, from the payslips and the working-time data. A generalist assistant produces an order of magnitude, not a count you can put before a court.
How do you check that a valuation produced by an AI is reliable?
By requiring each line to point back to the exhibit it is drawn from, and by submitting a file already argued whose outcome you know. A valuation you cannot trace cannot be checked, and so cannot be produced.
Can a file's data be loaded into a consumer AI?
Best avoided. A lawyer's files are covered by professional secrecy (article 66-5 of the law of 31 December 1971), which means looking at the data-processing contract, the hosting location and how the data is used.
Does an AI replace the lawyer before the conseil de prud'hommes?
No. It prepares the file, calculates and documents. The legal assessment, the strategy and the signature remain the lawyer's.
What is the difference with an online simulator?
A simulator applies a formula to data entered by hand. A file-preparation tool starts from the exhibits, establishes the reference salary and length of service itself, and ties each amount to its source.
Can a tool say where the amount will sit within the barème Macron range?
It can place the file against decisions handed down on close profiles, which lets you claim a supported amount. No tool can promise the amount a judge will retain.