Scrape AliExpress product reviews by URL or item id: star rating, buyer feedback translated into 16 languages, buyer photos, country, SKU variant. Date, keyword and star filters, new-review monitoring, free sentiment analysis. For dropshipping product research. No login. API and MCP ready.
Store URLs, photos and product details work with no memory setting: a run whose memory you leave unset now starts with what its input needs: 4096 MB for store URLs and photos, 2048 MB for product details, 512 MB for everything else, as before. Until 1.16 such a run started at 512 MB and its stores and photos came back as free "set memory to 2048 MB" rows. A memory you set yourself is used as it is (below 2048 MB the browser rows still say how to fix it). Other scrapers' names for these fields (storeUrl, stores, imageUrl, includeDetails, scrapeDetails…) count too.
A store's or photo's first reviews about three times sooner: the browser is now fetched in parts. Apify's Camoufox image layer is 1,441 MB, of which a Windows browser reads about 600 MB; the run now downloads and unpacks only those runs of files, each checked against its own pinned SHA-256 (the whole layer, checked against its digest, stays the fallback). The install went from 52.8 s to 16.5 s, Firefox's system libraries install without a disk sync per file, and the first browser visit's exit is located while the browser installs. With the 4096 MB a store run now gets, a visit that passes takes 13-21 s instead of about 50 s, so a store's first reviews come after about 40 s instead of about 2 minutes (each visit AliExpress's slider meets, never solved, adds about 15 s).
SUMMARY gateway.visit_steps says where each browser visit spent its time (exit lookup, launch, first answer, verdict, page load).
No prices or events changed; the product details add-on stays gated until 2026-12-02. Review-only runs read and bill exactly as in 1.16, at 512 MB.
1.16 — 2026-11-19
The browser install outlives a rebuilt image: 1.15 reads the browser from one pinned layer of Apify's Camoufox image. Should Docker Hub drop that layer after Apify rebuilds the tag, the tag's current linux/amd64 build is read instead (two manifest requests; its browser folder found by name, one version only, the download checked against the digest the manifest names). SUMMARY gateway.layer says which was used.
Refusals say what to do next: a photo search AliExpress refused now suggests starting the run again in a few minutes or finding the products with searchQueries; refused product details suggest a later run.
No prices or events changed; the product details add-on stays gated until 2026-12-02. Review-only runs read and bill exactly as in 1.15.
1.15 — 2026-11-19
Store URLs, photo searches and product details work again: on 2026-11-19 both GitHub hosts of the Camoufox browser build these need (github.com/daijro, then github.com/JWriter20) answered 404, so every store and photo search failed, and 1.14.3 blamed AliExpress's slider though no browser visit had been made (run NxjbWFgchBtl6rHYp). The browser now comes from Apify's own Camoufox image on Docker Hub: the layer holding the same v158.0.1-beta.41 build, pinned by its SHA-256, streamed and unpacked on the fly without the macOS and Linux font groups a Windows browser never reads (1,060 of 2,432 MB). The fingerprint model Camoufox fetched from github.com/scrapfly at its first launch comes out of the same layer, so a browser run no longer depends on GitHub; the release zips stay as fallbacks. The install now takes about 50-60 s (60-80 s in 1.14), and SUMMARY gateway says where the browser came from and how long each install step took.
Every failure says what to do: a run whose browser could not be installed says so (not AliExpress), ends FAILED so the canaries see it, and charges nothing; a run in which every product failed ends with "start it again in a few minutes" for a refusal, or "check the ids" when AliExpress knows none of them.
One row per review from the Console: the input form now starts with the product row off and status rows only for products that failed, so a first run's Output tab and its CSV or Excel export hold one row per review. API and agent calls that leave these fields out keep every row type, exactly as before. The two switches sit together in the form (Add a product row, Status rows).
Readable column names in every Output tab view (Stars, Date, Review, Translation, Variant bought, Buyer country, Photo links, Product link...): labels only; the views' fields and every row are unchanged.
README: a quick start (Console, one curl call, an AI agent through Apify's MCP server), integration recipes (Google Sheets through Make, Zapier, n8n, a CSV link, webhooks, schedules), a table of what users praise in AliExpress scrapers measured on this actor, and FAQ answers on CSV exports, Google Sheets, AI agents, typical costs and failed runs.
No prices or events changed; the product details add-on stays gated until 2026-12-02. Review-only runs read and bill exactly as in 1.14.
1.14 — 2026-11-19
Shipping for the country you ship to: the product details add-on read shipping from the product page's data call, which AliExpress quotes for the browser's exit country: aliexpress.us quoted the United States whatever shipTo asked, so a shipTo: "GB" row carried US shipping. A details row whose shipping came for another country than shipTo is now quoted by the product page's own shipping panel (the call it makes when a buyer picks another "Ship to"): cost in your currency, delivery days and every option for that country. A country AliExpress does not ship the item to is details_status: "unavailable", free; a quote that fails is never delivered as another country's (details_status: "error", free). New column shipping_to names the country the shipping was quoted for; price_region the market the variant prices were set in (the exit's: US), in your currency.
More in every details row, same event, same price: each variant's own photo (skus[].image), whether its price is AliExpress's new-shopper deal (new_shopper_price) and its per-shopper limit (max_per_shopper); the option groups with their swatches (sku_options); category_path (AliExpress's category ids from the top level down) and category_name; the seller's description as text plus its images (description_text, description_images, description_url) and attached files such as a PDF manual (documents). They come from the answers the add-on already reads, plus one plain request for the description page; the product_details event, its text and its price are unchanged.
Find products by photo:imageUrls (image URLs, or data: URLs) go to AliExpress's own search by image, uploaded the way the site header's camera button does it by the run's browser page (memory 2048 MB, as for stores); the matches, best match first, are read like any search and their reviews delivered. Finding them is free. A link that is no JPEG, PNG, GIF or WebP up to 5 MB is a free invalid_target row.
Other scrapers' names:imageUrl, imageFile (piotrv1001, devcake, zen-studio) → imageUrls; maxResultsPerImage → maxProductsPerSearch; memo23's scrapeDetails → includeProductDetails; piotrv1001's includeDescription is honoured (the description is on every details row).
The browser installs again: on 2026-11-19 the release host of the Camoufox build 1.13 installs (github.com/daijro) started to answer 404, so store URLs failed free with "the browser could not be installed". The install now falls back to a camoufox maintainer's CI build (v156.0.1-beta.34, from a repository camoufox's own package lists), checked against its pinned SHA-256 before it is unpacked. The browser process no longer gets the run's Apify variables (its API token among them).
No prices or events changed; the product details add-on stays gated until 2026-12-02. Review-only runs (product URLs, ids, searches) read and bill exactly as in 1.13.
1.13 — 2026-11-18
Store URLs and the product details add-on answer again, through one signed-out browser page: on 2026-11-18 AliExpress's h5 gateway (store product lists, product details) showed its slider check to every plain signed-out call (0 of 178), while a real browser visit of a product page passed. A run that has store URLs, or product details once their price is in effect (2026-12-02), now opens one Camoufox page on a fresh datacenter IP and lets that page's own scripts make the gateway calls (about 1-2.5 s each). Nothing is ever clicked or solved: a visit that meets the slider is closed and the next goes out on a fresh IP, at most 4 a run; then the remaining stores and details end as free rows, as before.
Memory and start-up: the browser needs the run's memory at 2048 MB. A run with less does not start it: its stores and product details end free with "set memory to 2048 MB" on the row and the run card. The browser is installed into the run only when a run needs it (about 70 s, while the first review pages are read), so the actor's image stays the slim one: review runs, searches and pasted product URLs start and run exactly as before, at the default 512 MB.
A store list that aliexpress.us answers names its products by their aliexpress.com ids.
No prices or events changed.
1.12 — 2026-11-18
Another scraper's market is the market a search lists:country, region or shipCountry (zen-studio, memo23, devcake, thirdwatch, pro100chok, khadinakbar) holding a two-letter code is read as shipTo when the input finds products, so their search comes back in that market (it went unused, and the listing was the US market).
Said clearly: a run in which every target failed ends with "Nothing was charged"; includeReviews: true and scrapeType: "reviews" are listed as honoured in SUMMARY aliased_fields instead of "ignored".
README: the answers in "Questions users ask about other scrapers" now cite the 1.11 runs that show each fix.
No prices or events changed; product-URL and product-id runs read and bill as in 1.11.
1.11 — 2026-11-18
Never billed past another scraper's limit:limit, max_results and resultsLimit are read as maxItems; maxResults in a search input also caps the products found over every search (zen-studio and khadinakbar count it in total); monitorMode (anything but off), incrementalMode, incremental and onlyNewItems turn on onlyNewReviews, with monitorStoreName / stateKey naming a store of our own; freeShipping, freeShippingOnly, choiceOnly, filters and minOrders become AliExpress's own search filters. An input that asks for no reviews (includeReviews: false, or scrapeType: "productDetails") now ends free with "Input problem" instead of being billed reviews it declined.
Store URLs and product details when AliExpress's gateway refuses: on 2026-11-18 the h5 gateway behind store lists and product details showed its slider check to every signed-out call (0 of 158 plain calls from datacenter and home IPs, with browser TLS fingerprints too; 4 of 10 answered on 11-17). The actor never solves a slider. A refused store is a free refused row and a run of refused stores ends green with the reason (runs oTn6bmmpln9ZNhbBw and 8HsLbdSAtNI6R1WoL failed after 64-131 s); refused product details are details_status: "refused", no row, no charge. After one call refused on every fresh IP the run stops asking, and gateway retries move to a fresh IP after half a second instead of backing off (a refused call took about 2 minutes). SUMMARY gateway counts the run's asks and answers.
Links without https:// work, and a bad entry no longer stops the run:www.aliexpress.com/item/… and aliexpress.com/item/… are read (also in requestsFromUrl lists); an entry that is no AliExpress link becomes a free invalid_target row and the rest is read (runs qDAvgbyjekpisLt5f and iVndUHBxpypKkEICe ended "Input problem" with 0 rows). An input with nothing usable at all still ends "Input problem", as before.
Search listings in the buyer's market:shipTo and currency (default US, USD) are sent as AliExpress's own Ship-to preference, so found products and their listing prices are that market's (GB/GBP, DE/EUR, BR/BRL measured); before, the listing followed the proxy IP's country (run 2FO4LXUV370HOeWKz mixed 25 USD and 20 SEK prices; run RRAlnhKyqj0DmgC17 ignored GB/GBP). SUMMARY carries market, listing_currencies and mixed_currencies, and the run's message flags a mix. A run that sets only a proxy country keeps that country's listing. A search card that comes without its price block (1 in 6) gets its price from the card's own cents field.
statusRows (all default, problems, none): with none and includeProductRow: false an export holds only reviews.
README: "Questions users ask about other scrapers", each answer with its test or run; the searchSort default is orders (the README said relevance).
No prices or events changed; the product details add-on stays gated until 2026-12-02. Product-URL and product-id runs read and bill as in 1.10.
1.10 — 2026-11-18
completeHistory, faster: the next lists of each kind (each star, photos and follow-ups, each buyer country) are fetched while the current one is delivered, on 4 sessions unless sessions is set: a 22,005-review best-seller in about 4.5 minutes, 21,964 of its reviews (run 0cCh45ec2TwcvEjwc; 1.9: about 11 minutes, run Zl9caEdSP1zEticTS). Delivery and billing still go list by list and page by page, so caps, maxItems and spending limits stop it exactly, and lists fetched ahead that are no longer needed are dropped unread.
Measured bulk speed (README): 10,527 reviews of 50 products found by five searches in about 4 minutes, about 2,570 a minute, the first after 10 seconds, for $0.006 of platform usage per 1,000 reviews (run 2FO4LXUV370HOeWKz).
No prices or events changed. Runs without completeHistory read and bill as in 1.9.
1.9 — 2026-11-18
Past AliExpress's 5,000-review limit (completeHistory): AliExpress serves at most 5,000 reviews of any one review list: a page reaching past the 5,000th answers page 1 again, while totalNum still lists them all (measured on a 22,008-review product). With completeHistory: true a product that lists more is also read through AliExpress's other lists of the same reviews (the other order, each star, photos, follow-ups, each buyer country), deduplicated by review id and billed once: 21,967 of 22,005 reviews in about 11 minutes and 16,627 of 16,729 in about 6 (runs Zl9caEdSP1zEticTS, UHWdb53dtQRhTXxi2) where one list gives 5,000 (run lhMpsHb0UvJTvJ6At). maxReviewsPerProduct, maxItems and the spending limit still stop it exactly; star and country filters still apply. Off by default; monitors and capped oldest runs do not use it.
A list ends at its 5,000th review: the walk stops there and asks for smaller pages near it (a page reaching past it is page 1 again). 1.8 went on re-reading page 1, 20 rows a request, until its 400-request limit; the same walk now takes 16 requests. On a ranked list over 5,000 the walk's second pass now runs as designed and adds what AliExpress's other shard ranks (measured 4,398 → 5,848 unique).
"Fewer reviews than listed" said on the run: when AliExpress lists more reviews than it serves, the run card says so and SUMMARY carries fewer_than_listed (per product: listed, read, why); with completeHistory the status row carries it as coverage_note.
Translations at depth: every cold page is re-asked for its translations, and once or twice more, a little later, if it is still bare (up to 40 re-asks per list; 1.8 stopped after 6, which left 5-10 pages of a long walk untranslated in a cold language: tr_TR and ko_KR pages came back 100% untranslated on the first ask; on Apify's proxy one re-ask still left a page bare now and then, 111 of 1,894 written reviews in run lhMpsHb0UvJTvJ6At). SUMMARY counts written reviews still delivered without a translation (untranslated).
sort: oldest with a cap: reads to the end of the list, so the very oldest reviews are no longer missed when the start falls between pages; on a product over 5,000 reviews it gives the oldest of the newest 5,000 AliExpress serves in date order (1.8 asked past the limit and got page 1: the newest), and says so.
Input form: completeHistory sits under maxReviewsPerProduct; reviewTopics moved to More options. No prices or events changed. Inputs without completeHistory read and bill as in 1.8, apart from these fixes.
1.8 — 2026-11-17
AliExpress's own star filter:minRating / maxRating read only the stars asked for (filter=1..5), so complaint and rave runs stop reading whole products: an uncapped complaints run on a 2,868-review product reads 390 reviews in 13 requests instead of 2,868, a default-cap one 100 in 5. The same reviews are delivered without a cap; with one, the stars are read one at a time (lowest first for complaints, highest first for minRating alone), so a capped run's sample differs from 1.7 while its count is the same.
Capped sorts AliExpress can serve:oldest with a cap now delivers the true N oldest (the end of the newest-first list), lowest_rating / highest_rating the true N lowest / highest rated (star by star); before, they sorted the first N AliExpress ranked.
reviewTopics: only reviews AliExpress itself filed under a topic of the product's review_topics ("defective", "good sound"), through its own topic filter.
Discovery: search pages are asked for as a browser, which reaches the full result list with filters and later pages (Googlebot got a 60-card SEO page); default searchSort is best-selling again, now on topic. New shipFrom, minPrice / maxPrice, categoryId; search, category and store URLs pasted in productUrls / startUrls are found like searchQueries (they ended as "Input problem" before). Found products carry their listing price, sold count and rating on the free product row.
Every other scraper's input starts an earning run:null for an optional field is "not given" (the platform refused the whole call before); startUrls accepts plain URL strings as well as objects; proxyConfiguration: {useApifyProxy: false} (shahidirfan's and piotrv1001's default, which failed every target because AliExpress refuses the platform's own IPs) and the RESIDENTIAL group (pro100chok's, khadinakbar's and piotrv1001's details default; AliExpress's gateway refused 10 of 10 residential IPs) use the included datacenter proxy; decimal counts are rounded down instead of refused by the platform. Measured on Apify: all eleven rival input shapes start runs, and the four that ended with nothing (two FAILED, two "Input problem") now deliver paid reviews.
Faster: the product page is asked for alongside the first review page. Measured median of three runs for 100 reviews of one product: 4.6 s total and the first reviews at 1.9 s (1.7.2: 8.6 s and 4.8 s); three products 9 s (1.7.2: 19 s); a three-product search 8 s (1.7.2: 14 s). A pasted search or category URL without an order of its own takes searchSort. Gateway calls (store listings, product details) stick to the session AliExpress let through.
No prices or events changed. Inputs without these fields read and bill as in 1.7.
1.7 — 2026-11-17
Product details add-on (includeProductDetails, with shipTo and currency): one details row per product that delivers reviews — price range over the variants with list price and discount, sold count, stock per variant and in total, wishlist count, shipping cost and delivery days, the store with its seller rating and opening date, the specification table and the images — read from the same data call AliExpress's own product page makes. Billed as one product_details event per row, $0.004 ($0.0035 SILVER, $0.003 GOLD and above), from 2026-12-02 (Apify's 14-day notice for a new event); until then it is switched off: no data, no charge, addons_inactive in SUMMARY and a note on the run card. An item not sold to shipTo gets a free note on its status row instead.
Input form reordered so the twenty fields an AI agent's card shows are the ones that change a run; no field was removed or renamed.
No prices, defaults or output fields of the existing rows changed.
1.6 — 2026-11-17
Find products, then their reviews:searchQueries (search phrases) and storeUrls (store pages or numbers) find their first maxProductsPerSearch products (default 5; searchSort relevance, orders, newest, price_asc, price_desc) and deliver their reviews in the same run. Finding them is free: only review rows are billed, at the same price. Found products carry target = the phrase or store and search_rank; a duplicate find is a free duplicate row and a phrase with no results a free no_results row. Other scrapers' search inputs (keywords + maxResults + includeReviews, searchQuery, scrapeType: search, sortBy: orders) map onto these fields.
Free monitoring deltas on the status rows of onlyNewReviews runs: monitor_rating, monitor_rating_change, monitor_reviews_change, monitor_new_low_star (1-2 star reviews that appeared since the previous run, whatever your filters delivered) and monitor_previous_at.
Fixed: re-asking an empty answer on fresh IPs closed the sessions other products were using at that moment, which could fail one of them (ClosedResourceError); the sessions are now dropped only between requests.
No prices, events or defaults changed; runs without searchQueries, storeUrls or onlyNewReviews keep their exact columns.
1.5 — 2026-11-17
maxReviewsPerProduct counts the reviews delivered when a filter is on. With stars, date, keywords, countries, text, photos or onlyNewReviews set, the actor reads on until N reviews pass (or the product ends) instead of filtering only the first N it read; a complaints run on a 2,800-review product now finds 100 one- and two-star reviews instead of the dozen among the first 100. A product still never bills more than N reviews. Without a filter nothing changes.
Newest first from AliExpress itself. AliExpress does serve its reviews newest first (the review widget's "Sort by latest", sort=valid_date, measured 2026-11-17); earlier versions believed it could not. sort: newest now returns the actual newest reviews of the product, and sinceDate reads newest first and stops at the first review older than the date (3 requests for the last 3 days of a 2,868-review product).
Monitoring (onlyNewReviews) reads newest first and remembers dates. A run stops two weeks below the newest review the last complete run saw, so a daily monitor costs a few requests and catches every new review wherever AliExpress ranks it (1.4 only saw new reviews that landed among the first 100 ranked). The first run delivers the newest maxReviewsPerProduct reviews and starts the monitor there; the history behind that cap, the rest of the day it stopped in included, is never billed later as new. A memory written by 1.4 is migrated on its first 1.5 run: reviews older than the newest review it had already delivered, or older than 3 days, are history, so the first run after the update does not bill old reviews as new. The memory holds at most 5,000 review ids per product (it grew without bound before) and lives under a new key, so the 1.4 list stays untouched. stateStoreName is now a field of the input form.
Reviews are delivered and charged a page at a time. A run aborted, killed or refused half-way through a product keeps (and has billed) exactly the pages it already pushed, and a monitor's memory is saved after each page. A product that fails after delivering ends partial (not error) with the count delivered. The free product row now follows its reviews instead of preceding them.
App share and affiliate links:https://a.aliexpress.com/_… and https://s.click.aliexpress.com/e/_… are followed through their one redirect to the product. A link that leads nowhere is a free error status row.
{"requestsFromUrl": …} in startUrls: a text, CSV or sheet export with one product URL, item id or share link per line, downloaded once.
Inputs from other AliExpress review scrapers work: their field names (filter, starFilter, sortBy, reviewsSort, results_wanted, product_url, maxReviews, includeProductSummary, smartScrape, locale…) and value spellings (latest, rating_desc, withPhotos, image, iw_IL, maxReviewsPerProduct: -1, minRating: 4.5) map onto ours when our own field is absent (SUMMARY aliased_fields); sort, reviewsFilter and language accept other spellings instead of the platform refusing the run. Unknown fields are ignored and named in SUMMARY ignored_fields and on the run card.
A monitoring store the run may not open fails cleanly with the reason and what to do (rename stateStoreName), before anything is read or charged.
No prices, events, defaults or output fields changed. Inputs without the new behaviour (no filter, default order, no monitoring) read, deliver and bill exactly as in 1.4.
1.4 — 2026-11-16
Never billed past maxTotalChargeUsd: each charge asks the platform only for the reviews that still fit under the spending limit, under one lock shared by every concurrent product, and exactly those are delivered; the run then ends — spending limit reached. Apify SDK 4.2 answers a request for more events than still fit by billing one event past the limit (a $0.30 limit was billed $0.302 on google-hotels-prices on 2026-11-16), and a limit below the price of one review was billed one review. Runs without a spending limit, or whose limit is not reached, are unchanged: no prices, input fields, defaults or output fields changed.
1.3 — 2026-11-16
A run that would outlive its own timeout now stops reading shortly before it (a tenth of the run's time, at least 20 s and at most 2 minutes): the product being read ends where it is and its reviews so far are filtered, delivered and charged, the products not reached get a free limit status row, and the run ends SUCCEEDED with "run time limit reached" on the run card and time_hit in SUMMARY. Before, the products being read at the timeout were lost and the run ended TIMED-OUT with no SUMMARY (measured 2026-11-16: 30 products, 60 s timeout, 2,195 reviews charged = delivered, TIMED-OUT). A walk that reads a product to its end and comes back more than 2% short of the review total AliExpress lists now goes round once more on fresh sessions and adds only what it had not seen: the backend answers from two shuffled orders of the same reviews, and one walk that mixed them read 2,706 of 2,868 (163 served twice) where another read all 2,868. Capped walks and complete walks are unchanged. Volume check the same day, all SUCCEEDED with charged = delivered: 30 products x 100 in 190 s, a 2,868-review product read whole in 33 s, 3 runs in parallel. No prices, output fields, defaults or input fields changed.
1.2 — 2026-11-16
maxItems (sent by AI agents) now caps the primary rows delivered and charged; add-ons follow the delivered rows. maxItems is one total over every product (review rows only; the product, insights and status rows stay free): products are read one at a time in the order given, reading stops as soon as maxItems reviews are delivered, and a product the cap never reaches is not fetched (its status row says limit). When maxReviewsPerProduct is left at its default (100), maxItems also raises it so the run can deliver up to maxItems reviews (with filters, reading goes on until enough reviews pass them); a maxReviewsPerProduct you set yourself stays, and the stricter cap wins. SUMMARY records max_items and the caps used. Without maxItems nothing changes: no prices, output fields, defaults or input fields changed.
2026-11-13 — Restore previous listing keywords
Restore title, description and SEO text to the November 5 baseline at the owner’s request. Preserve current runtime fixes, input/output contracts, billing, prices and defaults.
2026-11-06 — MCP discovery metadata
Clarify supported search intents and output nouns in the Store listing. Runtime, input contract, output and prices are unchanged.
2026-11-06 — MCP discovery metadata
Clarify supported search intents and output nouns in the Store listing. Runtime, input contract, output and prices are unchanged.
Changelog
2026-10-16 — Keyword Planner and locale documentation
Validated search terminology against Google Ads Keyword Planner and Store shelves. Added localized usage guidance and bounded examples; retained existing English phrases. Clarified locale limitations where measured. Corrected current-versus-scheduled price wording and arithmetic. No runtime, output fields, defaults, or pricing changed.
price change, set 2026-10-12, effective 2026-10-27
review $0.002 → $0.003 per row (rivals on the same shelf charge $0.0067–0.01 per review or a start fee). the actor has paying users, so Apify applies the increase on 2026-10-27 and notified them; until then the old price bills and the listing text below describes the price from that date; nothing else changes — same rows, same free rows, same filters, no start fee, proxy and compute included. Every price mention in the input form, the event text and this README was updated with it.
1.1 — listing and input form reworded, 2026-10-11
No behaviour, output or price changed; the same inputs return the same rows. Light re-check only: the title, description, input order and input descriptions of the 2026-09-20 pass are kept (they hold #1 for "aliexpress reviews").
seoDescription (indexed by the Store search) now carries the synonyms agents type (product reviews API, ratings, buyer feedback, photos, sentiment, monitoring) and the native nouns of AliExpress's biggest non-English markets: opiniones, reseñas, avaliações, отзывы, avis, Bewertungen.
Pricing event text states the per-1,000 price in words, what is never charged, that there is no start fee and that the run stops at the maximum cost per run set when starting it. Price unchanged.
Eight input descriptions trimmed to the 500 characters fetch-actor-details shows, protective sentence first (sinceDate/onlyNewReviews: set maxReviewsPerProduct 0; analyzeReviews/language: keep en_US) and a stated default on every field; names, types, defaults, enums and order unchanged.
README: new section "AliExpress scraper for AI agents: which input answers which request?" (when to use / when not, request → input table, row fields and the status/empty contract).
1.1 — 2026-09-20
Everything a 1.0 run sent and received works exactly as before: every addition below is optional and off by default, and the price is unchanged.
Only new reviews (onlyNewReviews): a monitoring mode for scheduled runs. The first run is the baseline; every later run delivers and bills only reviews it has never delivered before. The memory lives in a named key-value store in your own account (stateStoreName), and each status row reports already_seen.
Only reviews since (sinceDate): a date (2026-06-01) or a look-back window (30 days, 6 months). Applied before billing.
Only reviews mentioning (keywords): keeps reviews that mention any of your words, searched in the original text, the translation and the follow-up. Applied before billing.
Analyze reviews (analyzeReviews), free: sentiment, sentiment_score, topics and rating_text_mismatch on every review, and one free insights row per product — sentiment split, top complaints and praise, ratings by SKU variant, buyer country and month.
Review topics on the free product row (review_topics): the praise and complaint topics AliExpress itself mines from the review texts, each with its sentiment and review count.
Sixteen translation languages: Arabic, Hebrew, Indonesian and Thai join the twelve, with their localised review dates parsed to ISO.
Complete translations: AliExpress translates lazily, and the first request for a page nobody has opened recently could come back with most text_translated values empty. The actor now notices, re-asks that page once and fills them in — never at your cost.
Hide buyer details (hideBuyerDetails): blanks the buyer's name, avatar and gender.
Input compatibility: item ids are accepted in productUrls, URLs in productIds, and startUrls is an alias for productUrls, so the actor drops into Make, n8n, Zapier and other scrapers' output without reshaping.
2026-08-30
Reliability: a run that cannot act on its input — a URL from the wrong site, a name nothing resolves to, an empty target list, a number outside the documented range — now ends cleanly with the reason on the run card instead of as a failed run. Nothing is charged in either case; a blocked or throttled source still fails loudly.
Store listing: moved into the categories buyers actually browse (the old Business category has no Store page).
1.0 — 2026-08-29
First release.
Every AliExpress buyer review of a product as structured rows: the 1-5 star rating and the 0-100 score behind it, the buyer's own text and AliExpress's machine translation of it into any of twelve languages, the buyer's name, country and gender, the exact SKU variant the review is about, the buyer's photos, the shipping method the order used, helpful and unhelpful votes, the structured attributes AliExpress asked the buyer to rate, and the follow-up review they came back and added weeks later.
Products by AliExpress URL (aliexpress.com, aliexpress.us, aliexpress.ru, the /i/<id>.html short form, share links with tracking parameters) or by bare item id. All three AliExpress id namespaces (1005…, 3256…, legacy 2255…/4000…) resolve to the same reviews, and every URL form carries the id, so resolution costs no request.
Filters that run before billing: the three AliExpress applies itself (with_photos, with_follow_up, same_country, so the pages you do not want are never fetched) plus minRating, maxRating, requireText and countries in the actor.
Five delivery orders — as AliExpress serves them, newest, oldest, lowest rated, highest rated. AliExpress ignores every server-side sort parameter its own endpoint accepts, so the actor orders the reviews it collected rather than pretending the site does.
A free product row per item: title, image, average rating, the full 1-5 star histogram, the positive/neutral/negative split, how many reviews carry photos or a follow-up, AliExpress's product type, and the structured attributes buyers rated it on with the count behind each.
Paging that survives AliExpress's own paging bugs: page 1 answers with 20 rows however large a pageSize you ask for, later pages are a plain offset = (page - 1) * pageSize up to 500, and some offsets short-change you. The walk only ever asks for a page size that divides the offset and advances by the rows it actually got, which both closes the silent 80-review hole a naive pageSize=100 walk leaves and reads a 2,836-review product in about 15 requests instead of 142.
Reads AliExpress's own public feedback API over Apify's datacenter proxy — no key, no login, no cookies, no browser. An empty answer is never believed on its own: it is re-asked on a fresh session and then settled against the product page, so a product with no reviews is reported as no_reviews, an item id AliExpress does not have as not_found, and a refusal as an error.
Pay per review row; product rows, status rows, duplicate targets, unknown item ids and filtered reviews are free.