Merriam-Webster is compiled by lexicographers, so its entries carry editorial judgement about usage and are stable. Wiktionary is openly edited, which makes it far broader across languages and more variable in depth — excellent for a word no commercial dictionary covers, less dependable for a contested definition. Wolfram Alpha computes rather than retrieves: it will evaluate an expression or convert units, and it is the wrong tool for anything requiring interpretation.
Stack Exchange is worth understanding structurally. Answers are voted on by other users, and the accepted answer is chosen by whoever asked, so the top answer reflects community consensus rather than verified authority. On well-trafficked sites within the network that consensus is usually reliable; on quiet ones it can rest on very few votes.
Amnesty International and Human Rights Watch document abuses and campaign against them. Their reporting is detailed and sourced, and it is produced in support of a stated position — which is the point of an advocacy organisation, not a flaw in one. The United Nations and UNESCO publish with institutional caution and considerable process behind them, which makes them slower and more consensus-bound.
Pew Research Center is the different one: it runs surveys and publishes methodology, sample sizes and margins of error, and takes no policy positions. When a claim about public opinion needs checking, that methodological transparency is what makes it checkable. Snopes serves a narrower purpose well — tracing a specific viral claim to its origin. Every listing here is editorially reviewed, with outbound links marked sponsored or nofollow.