Duolingo Review: strong reach, slow delivery (59/100)

Field	Value
Domain	duolingo.com
Category	Consumer language-learning and educational apps
Pricing	Unknown in the supplied inputs
Pages crawled	40
Crawl date	2026-10-11
Overall score	59/100 (fair)

Duolingo is easy to understand: the homepage presents it as a playful platform for learning languages, chess, and more, with fun and personalization at the center. That clarity is the review's strongest through-line.

The score is pulled down by three practical issues. Performance is critical: homepage real-user evidence records p75 LCP of 6.997 seconds. Accessibility is also weak, with partial foundations but material barriers across the public surface. Editorial QA finds repeated programmatic course copy and multiple-H1 defects, while the public help surface is thin for a consumer learning product.

Technical SEO is stronger. Sampled URLs return 200, canonicals generally self-reference, robots.txt exposes a sitemap, and the guaranteed 404 probe returns 404. Risk & stability also scores strongly as a vulnerability scan, though no traffic symptom or GSC data was supplied. The result is a recognizable product with a meaningful performance, accessibility, and support backlog.

Duolingo is immediately legible as a playful learning platform. The homepage H1 promises fun learning for languages, chess, and more, while the meta description adds free, effective, science-based, personalized courses. That is memorable, but broad: the supplied crawl gives limited evidence of audience segmentation or competitive framing. The practical fix is to name the priority learner job and explain the distinctive learning mechanism or trade-off beside the alternatives surfaced in Google Suggest.

The homepage communicates a fun, free learning experience with personalization, but it does not fully answer audience-specific outcome, cost, or trust questions in the supplied artifact. Google Suggest includes cost, pricing plans, and reviews, which identifies real evaluation questions without proving that the homepage omits their answers. Keep the playful promise, then connect it to a concrete learner job, a realistic progress expectation, and findable information about method and plan trade-offs.

The simulated walkthrough found a clear first step: “The most fun way to learn languages, chess, and more!” and a recorded “Get started” route to /register. The main obstacle is mobile speed: PSI reports 11.3 s LCP, 4.2 s FCP, 25.5 s Time to Interactive, 7.5 s main-thread work, and a 4,443 KiB payload. Reduce initial JavaScript and payload, then verify that the rendered mobile DOM exposes the same primary path recorded in the link graph.

Duolingo public accessibility is poor at 42/100 on the supplied automated evidence. Course and Math pages report unnamed inputs; Math and Chess report 12 and 8 missing-alt images; the homepage PSI audit reports contrast and button-name failures; and viewport settings restrict zoom. Add semantic landmarks, meaningful headings, labels, alt decisions, and user-controlled zoom, then verify focus and announcements manually.

Duolingo’s visual execution is intentional: the 390px mobile render fits without horizontal scrolling, body text is 17px, primary actions are 330×50 px, and CLS is 0. The experience loses polish when interaction constraints are included: zoom is disabled, PSI flags contrast, 67 small tap targets are reported, and the homepage outline contains three H1 elements. Preserve the compact visual treatment while enlarging hit areas, restoring zoom, correcting heading levels, and reducing the 11.3 s mobile LCP.

Duolingo’s homepage is substantially slower than the Core Web Vitals target: real-user p75 LCP is 6.997 s versus Google’s 2.5 s good threshold, while the lab measured 11.258 s and 25.5 s to interactive. Startup work includes 7.5 s of main-thread work, 5.0 s of JavaScript boot-up, 2,297.5 ms blocking time, and 1,222 KiB estimated unused JavaScript. Reduce route code and third-party work before tuning layout; CLS at 0.05 p75 is already comparatively strong.

Duolingo’s clearest writing is direct and distinctive: “The most fun way to learn languages, chess, and more!”, “Learn English in just 5 minutes a day. For free.”, and “The most human bot ever”. The weakness is system-level repetition: sampled language pages reuse “free. fun. effective.”, “backed by science”, “stay motivated”, and “personalized learning”. The crawl also reports one 260-word paragraph with 43.3-word average sentences on course pages. Keep the voice, split the prose, and add course-specific proof.

Risk & stability scores 78/100 as a vulnerability scan, not a traffic-loss diagnosis: Google Search Console is not connected and no symptom was supplied. The clearest resilience concern is a sampled English course page with 1 raw word versus 262 rendered words and js_dependent=true. Put primary course content and metadata in the initial HTML, then use GSC to compare clicks, impressions, sessions, and landing-page trends before attributing any traffic change.

The sampled pages have a recognizable editorial voice, with headings such as “Chess for the rest of us” and “Math is hard, so we made it fun.” QA weakens at scale: Spanish, Italian, German, French, Dutch, and Greek samples share the same H2 sequence and near-identical supporting copy. The crawler also reports three H1 elements on the homepage and two on course pages. Require one course-specific evidence block and one semantic H1 before publication, and source or narrow ranking claims.

The discoverable public help surface is thin at 38/100. This does not establish that private or uncrawled support content is absent. Course pages provide BreadcrumbList schema, while the llms.txt probe returns application HTML rather than a dedicated plain-text index. Expose public onboarding, task guidance, and escalation paths.

The sampled public surface has a broadly healthy crawl foundation: sampled URLs return 200, canonicals generally self-reference, robots.txt exposes a sitemap, and the guaranteed 404 probe returns 404. The main technical risk is the English course template, where 1 raw word becomes 262 rendered words and js_dependent=true. Host consolidation also uses 302 redirects for HTTP and non-www requests. Server-render core course content and metadata, then use a single-hop permanent redirect policy.

Duolingo earns its strongest marks for immediate product legibility. The homepage states the proposition plainly, and First impressions & positioning scores 78/100.

The priority is clear: reduce public-page delivery cost, remove the public accessibility barriers, and build a findable self-serve help surface. Editorial QA should also address repeated programmatic course copy and multiple-H1 defects.

The homepage presents Duolingo for broad learners seeking fun, personalized language, chess, or math practice. The evidence supports broad learner support; narrower segments and outcomes remain unestablished.

This 13-dimension review uses the supplied crawl of 40 pages from 2026-10-11, the public dimension score table, and the stated enrichment context. The review covers positioning, audience and messaging, usability, accessibility, design execution, performance, writing quality, risk and stability, editorial QA, docs and self-serve help, and technical SEO.

Decision-support surfaces (12) was excluded because it was missing. Review-content integrity (13) was not applicable. No GSC data was connected, so Risk & stability is a vulnerability scan rather than a causal traffic diagnosis. Pricing remains unknown in the supplied inputs.

Read How SiteList scores for the review method. The byline identifies this as a 13-dimension review.