How Merito's tools work, and what their numbers mean

What you upload or write (documents, answers, drafts) is kept in your account, in private storage. Only you, and the people you choose to share a case with, can open it.

The tools do not read your raw files every time. Your documents are read once into a structured Case File, and the tools work from that file. This keeps the tools consistent with each other and makes their statements easier to check against your own documents.

Some steps use an AI model. Portions of your structured case data are processed by Anthropic through its commercial API. We do not use your case content to train, fine-tune or develop any model, and under the commercial terms we use, your inputs and outputs are not used to train Anthropic's models either. Larger document sets run in batches, which Anthropic keeps for up to 29 days and then deletes.

Text inside an uploaded document is treated as data, never as an instruction. A sentence in a document that says "ignore your rules" changes nothing, and no AI output triggers an action by itself: nothing is sent, filed or paid for without a click from you.

While your account is active we keep your case content so the tools can work. You can ask for deletion from your settings: the account enters a 14-day period in which you can cancel, and after that the request is reviewed and the content is permanently deleted. A few records are kept after deletion (for example billing records); the Privacy Policy lists them.

Merito is not a law firm and gives no legal advice. Everything it writes (letters, narratives, RFE responses, summaries, comparisons) is a draft: a starting text for you, and for your attorney if you have one, to read, correct and decide on.

Every result carries the same notice that it is a preparation aid, not legal advice.

A tool can be wrong. It can misread a document, miss a fact or phrase something too strongly. That is why a result should always be read against the documents it came from, by you or by your attorney.

The score has two parts. The first looks at how well the evidence in your Case File documents each criterion you rely on. The second looks at how the record as a whole reads, the step an officer takes after the criteria, called final merits.

• It measures how well your evidence is documented. It does not measure how good your work is.
• It shows one of five neutral bands (from "Getting started" to "Strong foundation") and a small number. In the top band the number is not shown, because it would read as more certainty than a readiness view can give.
• It is never a chance of approval, a percentage of success or a prediction of any decision. Merito does not compute one anywhere.

Anything a program can decide exactly (counts, dates, formats, whether a link opens, whether an exhibit is numbered) is checked by plain code, not by an AI. The AI is used for what code cannot do, such as reading a letter and saying which criterion it speaks to.

The AAO Decisions Explorer is built on decisions the Administrative Appeals Office has published. For each decision we keep the text and add our own fields: the criteria it discusses, the outcome, the field and the year. The short summary of each decision is written by us and checked against the decision before it is shown.

Most AAO decisions are non-precedent decisions. They show how a reviewer reasoned about one record; they are not rules, and a similar record can be decided differently. Read them as examples of reasoning, not as promises.

Every decision links to its source, so you can read the full text and see what the summary left out.

Each opportunity starts as a lead from a public list, feed or catalogue. We then open the organizer's own page and read it. The deadline, the fee, who can take part and how people are chosen come from that page. If the organizer's page cannot be found or read, the opportunity is not published, and a detail the page does not state is left blank, not guessed.

• Pay-to-present and pay-to-play events (where a fee buys the place or the honor) are rejected and hidden for everyone.
• Procurement notices, tenders and requests for services are not opportunities and are left out.
• Events and deadlines that have passed move to the next expected round, or expire.
• A second pass reads the page again; an unclear record is decided by a person, not by the program.
• Each record says how people are chosen: apply yourself, nomination by others, or by invitation.

The stars describe the activity: how selective it is, how well known the organizer is, how many criteria it can speak to, how strong the evidence it leaves you is, and how public it is. They are a guide to the opportunity, not a promise about any case.

We read only public pages, follow each site's robots.txt, never sign in, never get around a paywall or a bot check, and leave alone any source whose terms forbid collection.

The Citations tool reads free scholarly sources (OpenAlex, Semantic Scholar, Crossref and ORCID) and merges the same paper from several sources into one. Google Scholar is not read. The h-index is counted from the list the source gives, and the papers that make it up are marked in the list.

A number is only as complete as its source. If a source misses a paper, the number misses it too; the page says which source each figure comes from.

If a page, a summary or an opportunity is wrong, tell us through the contact address in the Privacy Policy. A correction is made at the source of the error (the record, the rule or the prompt), not only on the one page where it was found.