Yandex Market Reviews Scraper

Turn the customer reviews on any Yandex Market product into a clean, analysable table. For every review you get the star rating, the free-text opinion, and — because this marketplace asks its shoppers for them separately — what the buyer liked and what they disliked as two distinct columns, ready for analysis without any text-splitting of your own. Add the author, the publication date, the helpful votes, the number of photos attached, and the exact product version that was reviewed. Point it at one product or at a list of hundreds, choose whether you want the newest reviews, the angriest ones or the most helpful, and run it on a schedule to watch what buyers are saying over time.

Built for product managers, brand owners, marketplace sellers and market researchers who need real customer language about a product rather than a star average.

What you can do with it

• Mine complaints at scale. Order by lowest rated first and pull every one-star and two-star review for your product and your competitors', with the "disliked" column already separated out.
• Track sentiment after a launch or a change. Order by newest first, run it weekly, and see how the rating and the wording move.
• Benchmark against competitors. Give it a list of rival products and get one table with every review side by side, ready for a pivot table.
• Feed a language model or a sentiment tool with structured opinion text instead of screenshots — each row already separates praise from criticism.
• Find your product's real selling points by counting what buyers put in the "liked" column most often, in their own words.
• Spot review-worthy quality problems early by watching which product version the low ratings cluster around.

What you get

One row per review. An abridged real record:

{
  "productId": "1912483624",
  "productTitle": "Смартфон Apple iPhone 15",
  "productUrl": "https://market.yandex.ru/product/1912483624",
  "purchasedVariant": "Смартфон Apple iPhone 15 128Gb 2SIM черный",
  "reviewId": "324998227",
  "rating": 5,
  "text": "мне понравился.",
  "pros": "Качество сборки и дизайн, производительность, качество камеры",
  "cons": "Высокая цена, ограниченные настройки",
  "authorName": "дмитрий андреев",
  "authorPublicId": "b08e6dmyw0m11d1zr41jrxhqy8",
  "publishedAt": "2024-07-21",
  "publishedAtText": "21 июля 2024",
  "usefulCount": 5,
  "notUsefulCount": 5,
  "commentCount": 0,
  "photoCount": 0,
  "photoUrls": [],
  "page": 1,
  "positionOnPage": 1,
  "sortOrder": "useful",
  "sortRequested": "useful",
  "market": "ru",
  "regionId": 213,
  "regionName": "Москва",
  "collectedAt": "2026-08-17T10:00:00.000Z"
}

Input reference

Field	Type	Default	What it does
productUrls	list of links	—	Product pages on Yandex Market whose reviews you want. Copy the address of a product page and paste it in.
productIds	list of text	—	The numbers that identify products, if you already have them. Use these, the links above, or both.
maxReviewsPerProduct	number	100	How many reviews to collect for each product, up to 510 — the most the marketplace will serve for any one product.
sortBy	choice	useful	Which reviews you want first: Most helpful, Newest, Highest rated or Lowest rated. This decides what you get when a product has more reviews than you asked for.
minRating	number	1	Keep only reviews of at least this many stars.
maxRating	number	5	Keep only reviews of at most this many stars. Set both to 1 for one-star reviews only.
market	choice	ru	The market the reviews belong to. This marketplace publishes one public pool of product reviews, so it is fixed and shown for clarity.
productsAtOnce	number	3	How many products the run works through side by side, up to 10. Higher finishes a long list sooner.

Output fields

Field	Type	Description
productId	text	The number that identifies the product on the marketplace.
productTitle	text	The product's name as the marketplace publishes it.
productUrl	text	Link to the product page.
purchasedVariant	text	The exact version the reviewer bought — colour, storage size and so on — where the marketplace attaches one.
reviewId	text	The number that identifies this review.
rating	number	Stars, 1 to 5.
text	text	The free-text body of the review.
pros	text	What the buyer said they liked, as a separate field.
cons	text	What the buyer said they disliked, as a separate field.
authorName	text	The display name the reviewer publishes.
authorPublicId	text	The public profile identifier the marketplace shows alongside the review.
publishedAt	text	The publication day as YYYY-MM-DD.
publishedAtText	text	The date exactly as the marketplace prints it, so you can always check the line above.
usefulCount	number	How many shoppers marked the review helpful.
notUsefulCount	number	How many marked it not helpful.
commentCount	number	Replies attached to the review.
photoCount	number	Photos attached to the review.
photoUrls	list of text	Links to those photos at full size.
page	number	Which batch of the product's review list this came from.
positionOnPage	number	Where it sat inside that batch.
sortOrder	text	The ordering the marketplace actually applied, read back from its own answer. Empty if it did not say.
sortRequested	text	The ordering you asked for, kept beside the line above so the two can never be confused.
market	text	The market the reviews belong to.
regionId	number	The market identifier the answer came back stamped with.
regionName	text	The market's name as the marketplace prints it.
collectedAt	text	When this run collected the row.

Any field the marketplace does not publish for a given review comes back empty rather than guessed, and the column is still there — so your spreadsheet keeps stable headings.

Pricing

You pay for what you actually receive.

You are charged	How much
For every review written to your table	$0.0015 (that is $1.50 per 1,000 reviews)
For every product the actor works through	$0.020

The per-product charge covers the work of opening and checking a product's review list, which is the same job whether that product has four reviews or four hundred. A product that turns out to have no reviews to return is still charged for, because it was still checked. Two things are not charged: a product the marketplace refuses to show, and a product number the marketplace does not recognise at all — a stale or mistyped ID costs you nothing.

Worked example. 50 products at 200 reviews each: 50 × $0.020 = $1.00, plus 10,000 × $0.0015 = $15.00 — $16.00 in total. A weekly watch over 10 products, 20 newest reviews each, costs $0.50 a run.

Limits & what this actor cannot do

• Results are a snapshot. Reviews arrive on the marketplace continuously and a product's rating moves; every row is stamped with the moment it was taken.
• The marketplace serves at most 510 reviews for any one product. That is a wall on its side, not a setting on this one: a product listing five thousand reviews still stops there, and no actor can go past it. A quiet product returns the handful it has, and that is the honest answer, not a failure. Set the per-product number to what you need up to 510; for wider coverage, run more products rather than deeper ones.
• A deep pull can come back with fewer than 510 even on a busy product. The marketplace repeats some reviews as you go down its list, and a repeat is not a second review: it is served once here, and you are charged once. On a heavily reviewed product measured at the full depth, 510 places on the list held 444 distinct reviews. Every one of them is a real review, and none of them is a duplicate row.
• Reviews are in the language shoppers write them in, which on this marketplace is overwhelmingly Russian. Nothing is translated.
• Only the day of publication is published, never a time. Where the marketplace writes "today" or "yesterday" instead of a date, that is resolved against the day the run happened, and the original wording is kept in publishedAtText so you can always check.
• One market. This marketplace publishes a single public pool of product reviews and does not accept a different market, so there is no city or country to choose. Every row is stamped with the market its answer came back for.
• Nothing behind a sign-in. The actor reports what the marketplace publishes to the public. It does not sign in, sees nothing private, and cannot reveal a detail the marketplace keeps back — including a reviewer's real identity, their contact details or their purchase history.
• Not every review carries every field. Plenty of shoppers leave a rating and a "liked" line and nothing else. Those fields come back empty rather than invented.
• Product links must be product links. A link to one seller's listing of an item does not carry the product number, and the actor tells you so instead of quietly returning an empty result. Copy the address of the product page.
• A seller, brand or category link is refused before the run starts. Each of those addresses carries a number that is not a product number, and the marketplace answers it with a page that reads exactly like a product with no reviews — so they are rejected by name rather than turned into a confident empty result you would be charged for.
• A narrow star range takes longer and can return fewer reviews than you asked for, simply because fewer of that product's reviews match it.
• A product the marketplace refuses to show is reported as a failed product, counted separately in the run summary — never as "no reviews found". That distinction is the whole point: an empty result you can trust is worth more than a fast one you cannot.
• A product number the marketplace does not recognise gets its own line too. A stale or mistyped number is served the same page as a real product with no reviews yet — same heading, same size, same everything a reader can see — so a weaker tool tells you a product you no longer stock has zero reviews. This one reads the marketplace's own internal key for the product it resolved, says "the source does not recognise this product ID", counts it under productsNotFound, and does not charge you for it.
• Speed depends on the size of the job and on how quickly the marketplace answers on the day. No fixed rate is promised.
• Reviews are written by private individuals. Treat the author name, public profile identifier and review text as personal data: collect only what you need, keep it no longer than you need it, and handle it in line with applicable privacy law, including GDPR and Russian personal data law where they apply. The marketplace's own terms govern automated access, and you are responsible for using the data lawfully and in line with those terms.

FAQ

Do I need an account on the marketplace? No. Nothing is needed but the product links you want reviews for.

Does it need my login or password? No. The actor never signs in anywhere and has no place to put a password.

Can I schedule it? Yes. Set it to run daily or weekly on Apify and each run writes a fresh table, which is the usual way people track how sentiment moves after a launch, a price change or a bad batch.

Why are "liked" and "disliked" separate columns? Because the marketplace asks shoppers for them separately, and they arrive that way. It is the single most useful thing about this source: you get structured praise and structured criticism without doing any text analysis yourself.

Can I get every review a product has ever had? You can go a long way down — far further than the marketplace shows on its own first screen — but no source publishes an unlimited archive, and this one is no exception. Set maxReviewsPerProduct to what you need, choose the ordering that matters to you, and the run reports exactly how many it collected.

What happens if a product has no reviews? You get no rows for it and the run summary counts it under "products without reviews", clearly separated from products that could not be read at all — and also separately from products that have reviews of which none fall inside the star range you asked for. Those get their own line, so a one-star-only run never reads as "most of this catalogue is unreviewed".

What happens if one of my product numbers is wrong or out of date? The run tells you so by name. It is counted under productsNotFound in the run summary and logged as "the source does not recognise this product ID" — never as "this product has no reviews" — and it is not charged. This matters most when you feed a long list from your own catalogue: the products that genuinely have no reviews and the numbers that no longer exist are two different answers, and you can see which is which.

Can I filter to one-star reviews only? Yes. Set both minRating and maxRating to 1, and choose "Lowest rated first" so the run finds them quickly.