“Having AI read the contract” sounds easy, but whether the result is useful depends on how it is done. In this article we explain plainly what the Contract Auditor does step by step, what assurances it offers and what its limits are.
Step by step: what happens?
- Text is extracted on your device: Your PDF itself is not sent; only its text goes to the server for analysis. For scanned documents, text is read on your device with OCR.
- The document is recognised and questions are asked: The system identifies the document type and parties and asks a few questions whose answers change the analysis. You choose which side you are on; the system does not assume.
- The analysis runs: In the quick scan, the most important risks are found in a single pass. In the detailed audit the document is read several times, numbers are calculated, legal points are searched in official sources and findings go through a second check.
- The result is shown in the document: Risky spots are highlighted in your PDF; each finding states why, the impact and (in the detailed audit) suggested replacement wording.
Safeguards
- Every finding comes with a verbatim quote, and that quote is searched in the document text by software; a finding that cannot be found does not enter the report.
- Page numbers are computed from the quote's real position in the document, not guessed by the model.
- In the detailed audit, statute articles are not written from the model's memory; only content read from an official source is shown, with a source link. A point that could not be read is marked “could not verify”.
- Unknown information is stated openly in the assumptions section.
Limits you should know
- It can be wrong: it may miss a risk or overstate one. This is a preliminary assessment, not legal advice.
- It cannot judge documents it has not seen: annexes whose text is not in the document cannot be analysed; the report says so.
- The current-law check is done only in the detailed audit; the quick scan gives no statutory basis.
- Very long documents (over about 240,000 characters) must be split; for scanned documents the first 30 pages are read.
Quick scan or detailed audit?
The quick scan suits a first look at a document: it takes about 1-2 minutes and uses 8 to 20 units depending on length (monthly allowance first, then credits). The detailed audit is for an important contract you are about to sign: it takes about 6-12 minutes and uses 85 to 122 credits, from purchased credits only. The detailed audit costs more because the strongest model reads the document several times and searches legal points in official sources.
Privacy
Document text is not stored on our servers; the result is deleted within 30 minutes. Our AI provider does not use API data to train models by default; it may retain it for up to 30 days for safety review only. The result stays only on the page where the analysis finished: download the report right away.