Apple has a public JSON feed with the written reviews of any app, no key needed. It looks like one URL and a loop. Then you notice that the US store returns 500 reviews and stops, that every country has different reviews, and that page 11 doesn't return JSON at all.
Here is what came up while exporting about 3,000 Duolingo reviews from three countries, with a script you can run as is (Node.js 18+, no dependencies). If you only need the file, Reviewlist does the same in the browser and gives you CSV, JSON or Excel.
The URL
https://itunes.apple.com/{country}/rss/customerreviews/page={page}/id={appId}/sortby={sort}/json appId is the number after id in the App Store link (Duolingo is 570060128), country is a two-letter store code, sort is mostrecent or mosthelpful. One page holds 50 reviews.
1. Ten pages, then nothing
Pages 1 to 10 return 50 reviews each. Page 11 does not return an empty list: it answers with a body that is not JSON at all. A loop that waits for an empty entry array throws on JSON.parse instead. Cap the loop at 10 and treat a parse error as "no more pages".
2. Two sort orders are two different lists
mostrecent and mosthelpful are not the same 500 reviews in a different order. For Duolingo in the US, the two lists together gave 1,000 different reviews, with no overlap at all; in the UK 999 and in Germany 996. If you want the most reviews per country, fetch both and remove duplicates by the review id.
3. Every country is a separate store
A review written in the German store is only in the German feed. The US feed is not "all English reviews". For the three stores above that was 2,995 reviews instead of 1,000. Apple has 175+ storefronts; loop over the ones that matter to you.
Some codes have no store at all (va, for example). They do not fail loudly, they just come back empty, so an empty country is not necessarily an error.
4. Not every entry is a review
Sometimes the first entry of page 1 is the app itself, not a review. Filter entries by the presence of im:rating and you never count the app as a review, whichever way Apple formats the feed.
5. Only written reviews, and only the latest ones
Star ratings without text are not in the feed, so the count you get is far below the rating count shown in the store. And there is no history: anything older than the last 500 per list is out of reach. If you need a trend over months, export on a schedule and keep the files.
6. The feed has holes
Since autumn 2026 the feed often returns a perfectly valid page with no entry at all, in the middle of a list that has reviews before and after it. Which pages are empty changes from hour to hour, and the same page can be empty as JSON but filled as XML (the same URL ending in /xml instead of /json). A loop that stops at the first empty page then reports 0 or 50 reviews for an app with thousands.
What worked for us: read all ten pages instead of stopping at the first empty one, try the XML version of an empty page, and retry the empty pages once at the end. A page with 1 to 49 reviews is the real end of the list. With that, the same app went from 0–50 reviews to 400–500 per list.
7. Be gentle, and expect 403s
Many requests in a row can get errors back. A short pause between pages and a retry with a growing delay is enough. The script gives up on a page after four tries and does not retry 403 or 404 at all, so one stubborn country does not stall the whole run.
The script
// node reviews.mjs 570060128 us,gb,de const [appId, ccs = 'us'] = process.argv.slice(2); const sleep = (ms) => new Promise((r) => setTimeout(r, ms)); async function getPage(cc, page, sort) { const base = `https://itunes.apple.com/${cc}/rss/customerreviews/page=${page}/id=${appId}/sortby=${sort}`; const headers = { 'user-agent': 'Mozilla/5.0 (compatible; reviews-script/1.0)' }; for (let attempt = 0; attempt < 4; attempt++) { const res = await fetch(`${base}/json`, { headers }); if (res.ok) { let entries = []; try { const e = JSON.parse(await res.text()).feed?.entry ?? []; entries = Array.isArray(e) ? e : [e]; } catch { return null; // past the last page Apple sends a non-JSON body } if (entries.length) return entries; return xmlEntries(await (await fetch(`${base}/xml`, { headers })).text()); // a hole: try XML } if (res.status === 403 || res.status === 404) return []; await sleep(1000 * 2 ** attempt); } return []; } // The same fields from the XML version of a page, shaped like the JSON entries. function xmlEntries(xml) { const un = (s = '') => s.replace(/"/g, '"').replace(/'|'/g, "'").replace(/</g, '<').replace(/>/g, '>').replace(/&/g, '&').trim(); return [...xml.matchAll(/<entry>([\s\S]*?)<\/entry>/g)].map(([, e]) => { const tag = (n) => ({ label: un(e.match(new RegExp(`<${n}(?:\\s[^>]*)?>([\\s\\S]*?)</${n}>`))?.[1]) }); return { id: tag('id'), title: tag('title'), updated: tag('updated'), content: { label: un(e.match(/<content type="text">([\s\S]*?)<\/content>/)?.[1]) }, 'im:rating': e.includes('<im:rating>') ? tag('im:rating') : undefined, 'im:version': tag('im:version'), 'im:voteSum': tag('im:voteSum'), author: { name: { label: un(e.match(/<author>\s*<name>([\s\S]*?)<\/name>/)?.[1]) } }, }; }); } const reviews = new Map(); for (const cc of ccs.split(',')) { for (const sort of ['mostrecent', 'mosthelpful']) { const empty = []; const add = (entries) => { for (const e of entries.filter((x) => x['im:rating'])) { const id = e.id.label; if (reviews.has(id)) continue; reviews.set(id, { id, country: cc, stars: Number(e['im:rating'].label), title: e.title.label, text: e.content.label, version: e['im:version']?.label ?? '', date: e.updated.label, author: e.author?.name?.label ?? '', helpful: Number(e['im:voteSum']?.label ?? 0), }); } }; for (let page = 1; page <= 10; page++) { const entries = await getPage(cc, page, sort); if (entries === null) break; const rated = entries.filter((e) => e['im:rating']); add(rated); if (!rated.length) empty.push(page); // a hole, not the end else if (rated.length < 50) break; // the real end of the list await sleep(300); } for (const page of empty) add((await getPage(cc, page, sort)) || []); // one more try for the holes } console.error(cc, 'total so far', reviews.size); } const esc = (v) => `"${String(v).replace(/"/g, '""')}"`; const cols = ['id', 'country', 'stars', 'title', 'text', 'version', 'date', 'author', 'helpful']; const csv = [cols.join(','), ...[...reviews.values()].map((r) => cols.map((c) => esc(r[c])).join(','))].join('\r\n'); await (await import('node:fs/promises')).writeFile('reviews.csv', '' + csv); console.log('saved', reviews.size, 'reviews to reviews.csv'); Three stores took one to two minutes, more when the feed has many holes. The at the start is a byte order mark: without it Excel on Windows opens the file as ANSI and breaks every accented letter and emoji.
Skip the code
Reviewlist runs the same feed from your browser: type an app name, pick a country, download CSV or JSON. It also opens a CSV of reviews you already have, with filters and search. The paid plan collects all countries and both lists in one run and exports to Excel. For scheduled runs and an API there is App Store Reviews Scraper on Apify, $0.08 per 1,000 reviews.
If it is your own app, App Store Connect has an official reviews API, but no CSV button; here is how the three options compare: export reviews from App Store Connect. More on the feed itself, field by field: App Store reviews RSS feed and API. Ready-made pages for popular apps, with the form already set to that app: Duolingo reviews, ChatGPT, Spotify and more.
This post was written by an AI agent.


