How Which Language App scores and compares its apps

What is compared

The dataset covers consumer language-learning products a person can buy for themselves: course apps such as Babbel or Busuu, flashcard and vocabulary apps such as Memrise or Drops, audio courses such as Pimsleur, immersion apps such as Lingopie or LingQ, tutor marketplaces such as Preply or italki, AI conversation practice such as Speak, and language-exchange apps with a paid tier such as HelloTalk or Tandem.

Each app is one record: languages offered, free tier, platforms, offline mode, speech recognition, live tutoring, placement test, refund or trial policy, CEFR levels, company, country and founding year. Each published consumer plan is one row: plan name, tier, billing period, price, currency, price per month at that billing period, and what the plan covers.

Business, school and enterprise plans are excluded. Regional prices other than the United States are excluded. In-app purchases that are not a subscription tier, such as gem packs, are excluded.

How the finder works

The finder at the top of the home page answers with two apps: one to start with, and a second that adds a different kind of practice, such as a language partner or a tutor beside a course. It reads the whole catalogue. An app is considered for a language only when it lists that language on its own site, and a course its maker lists as still in development is named but never recommended.

Nine apps also have detailed method records, taken from their official course and help pages: which languages, goals and levels each course serves, how it teaches, and whether its free access lasts. One of these that fits every choice you made counts as a match. For any other app, or a choice its record does not cover, the answer says what to check before you pay.

The order follows fixed rules, not a personal fit score: first the apps that keep your budget, then those that match your choices, then those with fewer things to check, then the usual starting points for your goal. The second app has to keep the budget too; when no other kind of app does, the answer names one app and says what more money would add. There is no affiliate weighting and no predicted learning outcome.

A stage match needs language-specific course evidence. Beginner and intermediate are broad descriptions, not proficiency assessments. Advanced study, a tutor’s curriculum or an open-ended reading library may need an individual check. A general CEFR range is never treated as a named exam-preparation claim.

Each method claim has its source address, the date it was checked and a short excerpt in the maintained data records. A record without a valid check date leaves its app out of the matches; how a course teaches does not change once the course exists, so the age of a check never removes an app. Product facts are kept apart from our advice about fit, and competitor pages supply no facts or scores. Prices come from the plan records, with their own check dates.

Your choices are written into the page address, so an answer can be bookmarked or shared, and “Copy a link to this answer” copies that address. Like any page address, it reaches the analytics described on the privacy page, and showing an answer sends one analytics event with its language, goal and level. Nothing is kept in your browser’s storage. Without JavaScript, the lesson descriptions and complete price comparison remain available.

Use the app picker on the home page; each app's page explains how its lessons work.

Two kinds of evidence

Published facts. Prices are checked against each app's own pricing page, or its help centre where the pricing page does not list a figure, as shown to a visitor in the United States. App-store listings are used for platform facts and for in-app subscription prices where the app publishes them only there. A feature is recorded as present only when the app states it on a page it controls. Every fact carries its source URL, listed under Sources at the bottom of the app's page; facts that change (prices, plans, free access, languages offered, trials and refunds) also carry the date they were checked. Features and company details do not change once an app has them, so they carry their source only.

Independent reviews and owner reports. For each app, up to four public sources were read: published tests and reviews by technology and language-learning publications, long-form owner reviews, and public forum threads. The app's own site and blog, pages that exist to sell a competitor, and Amazon review pages are not sources. Each source was read against the eight criteria below and given a score of 1 to 5 per criterion it addresses, with a one-line reason in this site's words; where a source says nothing about a criterion it gets no score, never a guess. The sources are cited at the bottom of the app's page. They are citations, not the argument: no score or ranking is copied from any of them, and no pick is taken from one publication's list.

Store reviews, counted, not scored. Each app page gives Apple's and Google's own star ratings for it, dated, and, where there are at least 20 one- and two-star reviews among the up to 500 most recent in Apple's US App Store, how many of those low reviews mention the three topics named most often, from a fixed list: the price or the subscription, limits on free use, ads, bugs, speech recognition, signing in and lost progress, customer support, AI, and lessons that are too easy or wrong. A script matches the same words every time, and a review can count under several topics. The counts say what unhappy reviewers write about, not how many people are unhappy, and they never change a score or a pick.

The eight criteria

Every app is scored 1 to 5 on each criterion. What counts for each, and what the published facts alone can say about it:

• Teaching depth: is the grammar explained, and is there a structured course that builds level by level with review? From the facts: a course app starts at 4, an audio course or a tutor service at 3, an immersion, vocabulary or AI-conversation app at 2, a language-exchange app at 1, plus one point when the app states one CEFR range for its courses as a whole. Where an app states a range course by course instead, each language page shows it, but it does not add the point. Reviews then say how well the teaching actually holds up.
• Speaking practice: can you speak into it, and does it correct you? From the facts: live lessons with a tutor as the product itself is 5; live tutoring sold beside an app, an AI conversation partner, or real conversation partners on an exchange app is 4; speech recognition on your pronunciation is 3; nothing stated is 1. Reviews say whether the practice is worth anything.
• Language coverage: how many languages, and how deep beyond the big ones? From the facts: 50 or more languages is 5, 20 or more is 4, 10 or more is 3, 3 or more is 2, fewer is 1. Reviews say whether the less common ones are thin.
• Free tier: can you learn on the free tier, or is it a demo? From the facts, using the app's own description: the whole product free with ads or limits is 5, a free account with caps is 3, the first lessons only is 2, a trial that ends or no free tier is 1.
• Price per month: what a year costs, per month, at annual billing, and whether it is worth it. From the facts: $5 or less a month is 5, up to $8 is 4, up to $12 is 3, up to $18 is 2, more is 1; a tutor marketplace where the tutor sets the rate is 3; an app with no published web price, or one sold only by a one-time purchase, is not scored on price, and its page gives the purchase price instead. Reviews add value for money.
• Trial and refund: is there a real trial, and can you get your money back? From the facts, using the app's stated terms: a free trial and a money-back guarantee is 5, a money-back guarantee alone is 4, a free trial alone or a replaced trial lesson is 3, refunds only under Apple's or Google's rules or no terms published is 2. Reviews say whether refunds are honoured in practice.
• Offline use: does it work without a connection? From the facts: offline mode stated is 4, stated absent is 1, not stated is not scored unless a review addresses it. Reviews can lift a stated offline mode to 5 or lower it.
• Staying power and bugs: do people keep using it, and does it work reliably (crashes, lost progress, billing trouble, unanswered support)? This criterion comes from reviews and owner reports only; the facts an app publishes cannot answer it.

How a score is formed

For each criterion, the published facts give one number by the rules above, and the independent sources give another: the average of the scores they were given for that criterion. Where both exist, the criterion score is the average of the two. Where only one exists, it stands alone and the app's page says which. The overall score is a weighted average of the criterion scores: teaching depth and speaking practice count double, price and staying power one and a half times, free tier and trial and refund once, language coverage and offline use half. A criterion with no score is left out of the average rather than counted as zero.

Scores are worked out from the recorded facts and review scores every time the site is built. When a published fact changes, or a fact is corrected, the scores, the ranks and the picks that depend on it change with it; nothing is held over from an earlier assessment.

Disagreement between sources lowers confidence; it never picks a winner. An app's page shows how many sources were read and whether they broadly agree. Confidence is high with three or more sources that agree within two points on every criterion, medium with two sources or with wider disagreement, low with fewer than two.

Thin evidence means no pick. An app with fewer than two independent sources is still scored from the published facts so it can sit in the tables, but it carries no pick anywhere on the site, and its page says so. Cambly has one source, and 90 Day Korean, Anki, Assimil, ASL Bloom, Auslan Wiz, Auslearn, BaseLang, Beelinguapp, Bright BSL, Bunpro, ChinesePod, Coffee Break French, Coffee Break German, Coffee Break Italian, Coffee Break Spanish, Conversation Exchange, Dinolingo, Dreaming Spanish, Du Chinese, DW Learn German, ELSA Speak, Endless Spanish, Fluent Forever, Gus on the Go, Gymglish, Hack Chinese, HelloChinese, Human Japanese, Intersign ASL, Intersign BSL, Jiveworld, Jumpspeak, Kwiziq, Langotalk, Langua, Language Reactor, Language Transfer, LingoDeer, LingoHut, Lingvano, Michel Thomas Method, Migaku, MosaLingua, News in Slow, Novakid, Ouino, Pingo AI, Pleco, Pocket Sign, Praktika, Promova, Quizlet, Readlang, Readle, Renshuu, Rockalingua, Rocket Languages, Satori Reader, SignSchool, Skritter, SpanishDict, Speakly, Speaky, StoryLearning, Studycat, SuperChinese, Talk In Arabic, Talk To Me In Korean, Talkpal, Teacher AI, Univerbal, WaniKani, Yabla and Yoyo Chinese have none yet, so none of these seventy-five carries a pick; every other app has at least two. The coverage is listed on each app's page under Sources.

The picks and the order of the tables

A pick is named for the buyer's situation: a trip, an exam or a CEFR level, a daily habit, speaking with a tutor, no budget, a commute, reading and listening, and so on. Each situation weights the criteria that decide it (for a trip: speaking practice three times, offline use twice, then trial terms, price and teaching depth) and admits only the apps that fit it (a tutor pick needs live tutoring; an exam pick needs a stated CEFR range; a free pick needs a free tier you can learn on). The app with the highest weighted score that has at least two independent sources takes the label; ties go to the cheaper app. An app is used at most once per page, with one exception: when the overall pick is also the best app for a situation, that label joins its card rather than passing to a weaker app. A situation with no eligible, evidenced app is left out rather than given to a weaker candidate. The reason on each card is written from the app's scores and its published facts, never copied from a source; a mark against the app stays in the score table on its page.

Cards and comparison tables show the same overall score and an overall rank among all covered apps; tied scores share a rank. These are rubric scores, not customer star ratings, and source counts describe evidence coverage. Situation-weighted scores choose the picks but never replace the displayed overall score. The overall pick, first on the homepage and the best-apps page, is the app with the highest overall score among those with two or more sources.

Every table groups apps by what they are (course apps, audio courses, vocabulary apps, immersion apps, AI conversation practice, tutors and live classes, language exchange) and orders them by overall score inside each group, highest first, with ties going to the cheaper app. The kind of product is the first thing a learner decides, so the groups stay; price is a sortable column and a filter, never the default order. The "best for" line on each row is a description of what the app is for, not a pick label.

The functions that score, group, sort and pick read the published facts and the review scores only; the field that records whether an affiliate programme has been joined, and the file that maps apps to Amazon listings, are not readable by them. Commercial status never affects a score, an order, a pick or which merchant is named first on a button.

How prices are made comparable

Price per month is the plan price divided by its term in months. Babbel's $107.40 12-month plan is $8.95 per month; its $17.99 1-month plan is $17.99 per month. Prices marked EUR are explicitly labelled priced in euros and excluded from dollar price sorting; no exchange rate is invented. The per-month figure at annual billing is the one the tables show first because it is the figure most people end up paying; the monthly price is shown beside it.

Each price is what the seller was charging on the day it was checked. Where a page shows a sale price next to a regular price, the sale price is recorded and the regular price is kept beside it with the same evidence; a crossed-out figure that is only a comparison, such as a year at the monthly rate, stays in the plan's notes. Where only a promotional price is visible, that price is recorded and the row says so.

Tutor marketplaces price lessons, not months, and each tutor sets the rate. On italki you buy credits and spend them lesson by lesson; on Preply you subscribe to a set number of lessons with one tutor, at least four, renewed every 28 days. For these the published per-lesson price range is recorded, low and high, and shown in place of a per-month figure rather than ranked against subscription apps by a number that would mean something different.

Prices in a currency other than US dollars are recorded in that currency and never converted, because a converted number would carry a date and a rate this site cannot stand behind.

How we check, and what a dash means

Prices are re-checked weekly against maker pages. Each price shows when its value was updated. A failed check keeps the previous price and date; after failed reads on three separate days, it is also labelled last seen. An unchanged price keeps its observation date. Open the maker page before buying, especially for regional pricing.

A dash in a table means exactly that: when checked, the app's own page did not show the figure to a visitor in the United States, or the page renders its prices only after sign-in. Such a row is sorted to the end of any price-ordered table and the reason is recorded on the app's page. An existing price stays visible with its date when a source cannot be checked. No new price is guessed.

What this site does not do

This site does not use each app for months, run lessons to completion or score speech samples, and does not claim to. What it does instead is the rubric above: the same eight criteria applied to every app, from the facts each app publishes and from the consensus of independent reviewers and owners who did use it, with every score and its reason shown on the app's page so a reader can judge the verdict without visiting anyone else's site. It shows no display advertising, sells no placement, and lets no commercial relationship touch a score or an order.

Affiliate links, Amazon and revisions

Outbound links to apps are marked rel="sponsored" wherever the app runs an affiliate programme this site could join. When a programme is joined, the link becomes a tracked referral link and the affiliate disclosure page lists the app, the network and the rate. We have not joined any app's programme, so no link to an app is tracked yet. Where an app's boxed or lifetime edition is listed on Amazon, a button links to that listing and carries this site's Amazon Associates tag. As an Amazon Associate I earn from qualifying purchases. Which merchant comes first on a button follows one rule: the app's own store first wherever the app runs a direct programme, Amazon first only where it does not. The rule never reads a score.

Every row can be traced: the source URL, and for prices and terms the date it was checked, are listed under Sources on the app's page and are included in the CSV and JSON downloads. Corrections sent to [email protected] are checked against the app's page and applied with a new date. This method was last reviewed on 2026-10-12.