How we build our research

Every stock analysis starts from real raw data pulled from external sources (see Data Sources for the full list). Deterministic code — not AI — computes the StockIQ AI Score and its 7 sub-categories (technical, fundamental, growth, fair value, news, quality, and risk) from a fixed formula documented on the Methodology page. Only after that computation already exists does AI turn it into readable prose — it is never the thing that decides the score.

We prefer primary and regulatory sources wherever possible: SEC filings (13F, Form 4) straight from EDGAR, short-interest data from FINRA, and market/financial data from several commercial providers (FMP, Twelve Data, Finnhub, Yahoo Finance) so an outage at any single provider doesn't take down the whole page — when that happens, the page explicitly shows which categories are currently unavailable instead of guessing.

Every data point flows through a caching layer with a real timestamp ("updated X ago"), so it's always possible to see exactly how fresh what's displayed is. When a source returns a value that looks implausible (a unit or currency mismatch, for example), we prefer showing "unavailable" over a wrong number — the same reason some pages (like TASE stocks missing financial statements at our providers) are explicitly marked incomplete rather than pretending to be a full analysis.

Every AI call on this site carries an explicit system instruction: explain only the data that has already been computed and provided to it, and never invent a fact, number, or event not present in that data — including an explicit ban on "buy"/"sell" or any investment recommendation. When AI is unavailable (provider load, a network error), the page still shows every real score and data point — just without the written paragraph. See the AI Methodology page for more detail.

Market data and news refresh continuously (most caches renew anywhere from a few minutes to a few hours, depending on the data type). Written research for each stock is backfilled gradually in the background, not regenerated on every view — every page shows a real "last updated" timestamp so it's always clear how current the shown research actually is.

When different providers disagree on the same data point, we don't average or guess — we follow a fixed priority order (generally: regulatory/primary source first, then the primary commercial provider, then a fallback provider), and that order is defined directly in code rather than decided ad hoc per case.

Every score, percentage, and number on a page is a computed data point — not an opinion. Interpretation (the written paragraphs, "why this matters") always sits separately from the raw data and is always framed as one possible reading of it, not as an additional fact. When we're not confident about something, we say so instead of filling the gap with an overconfident guess.

The Performance Tracker shows real outcomes — including stocks whose score dropped or whose return was negative — not just wins. We treat that as part of the product's credibility, not a flaw to hide: a model that only ever shows successes is a model that can't actually be checked.

We deliberately avoid language like "the stock will rise," "buy now," or a promise of returns — in AI-written content and in the site's own marketing copy alike. Any page showing performance (including the hypothetical model portfolio) is explicitly labeled as hypothetical/historical tracking, with a clear note that past performance does not predict future results.

For the full process of reporting and fixing data errors, see the dedicated Data Corrections policy.