Accuracy and reliability

Answers are produced by artificial intelligence, but we don't hand you its free-form text directly. After the language model has drafted an answer, a separate verification mechanism checks every material statement against the original text of the law. The principle is simple: cite or drop. If a statement can be confirmed by a specific provision, it stays. If no provision confirms it, it doesn't make it into the answer. Put simply: if it isn't in the law, e-zakon doesn't say it.

This check happens before you see the answer, not after the fact. That is the key difference from systems that merely generate plausible text and can invent an article, a deadline, or a fine. Where artificial intelligence tends to "fill in" convincing details, our verification mechanism works the other way — it strips out anything not grounded in the text of a provision. We deliberately trade completeness for groundedness: better to say less but accurately than more but unreliably.

We include only acts that are in force; repealed versions are excluded so that an answer never rests on a provision that has lost effect. At the same time we're honest about the limits: this is a prepared database of current legislation, not absolutely everything ever enacted in the country. If an applicable provision lies outside the database, e-zakon may not see it. We don't promise exhaustive coverage — we promise that what e-zakon cites genuinely exists and is in force.

The law changes, and an answer is useful only when it reflects the law's current form. We keep the database up to date: new and amended acts are added on a regular basis, and versions that have lost effect are removed. Because of this, answers rely on the law as it stands now, not on an outdated version.

We don't disclose the details of our update process, but we stand behind the outcome itself: at the moment of answering, e-zakon relies on the in-force versions in its database. That said, we can't guarantee that every change is reflected instantly, so for especially recent questions we recommend checking against the primary source — all the more so because a link to it appears right in the answer. The infrastructure that runs all of this is hosted on Google Cloud.

We hold that a statement about the law without a link to its source is just an opinion. So every material statement in an answer comes with a link to a specific act and article. This isn't decoration but a consequence of the same verification: the link leads to a real source you can open and read yourself. If a statement can't be tied to a specific provision, it isn't left in the answer without a link — it's dropped.

This approach makes the answer verifiable. You don't have to take our word for it: specific articles, deadlines, and fines are cited so that you can confirm for yourself that they exist. A finished answer can be saved as a PDF and shared together with the links to the provisions.

The verification mechanisms are responsible for ensuring that every statement rests on the law in force. The methodology and overall approach of e-zakon, in turn, are reviewed by a qualified lawyer. e-zakon was created by the same legal and tax practice as tax-legal.uz, with over 10 years of experience. Learn more about the team on the about page.

For all the rigor of these mechanisms, e-zakon's answers are reference information, not legal advice. Artificial intelligence can be wrong: it may misread a question, fail to find the applicable provision, or miss a nuance of your specific situation. Groundedness and source links reduce the risk of error but do not eliminate it entirely.

So for important decisions — where money, deadlines, rights, or obligations are at stake — double-check the information against the primary source and consult a qualified lawyer. We build trust through transparency, not through promises of infallibility. How to ask a question, and what the service can and can't do, are on the how it works page.