# Yandex Market API Product Research: Read the Russian Complaints First

> We pulled 147 Russian kettle complaints with the yandex market api and store reviews. Claude sorted them into 8 types: 57% are after-sales. Cost: $0.057.

- Canonical page: https://everyinfra.com/en/blog/yandex-market-reviews-claude-product-research
- Published: 2026-11-16T01:45:39+08:00 · Fact-checked: 2026-11-16T01:00:00+08:00

Open the top-selling page for electric kettles on Yandex Market and every product looks safe. In our pull of 20 rows, every rating sits between 4.5 and 5.0. The page does not tell you which kettle leaks after three months or which brand makes you ship a return to another city. We tested a different route on November 16, 2026, from 00:17 to 00:32 Singapore time. We used the EveryInfra `yandex/market`, `yandex/search` and `yandex/reviews` capabilities, plus AliExpress and YouTube for comparison.

The run had 20 billed calls and cost $0.066892. We collected 147 Russian reviews with 1 to 3 stars from the Moscow and Saint Petersburg stores of two kettle brands, Polaris and Kitfort. Claude sorted them into 8 problem types. For 84 of the 147 complaints (57%), the main reason was after-sales: warranty, repair, delivery or refunds. Only 33 (22%) were mainly about a broken product.

> **Key points**
>
> - `yandex/market` returns price, rating, rating count, shop and stock. It returns no review text, and `review_count` was empty in 40 of 40 rows.
> - The Yandex Market Partner API returns reviews for your own goods only. The Content API has been closed to new users since December 26, 2018.
> - Yandex Maps reviews of brand stores and service centres hold Russian complaint text. Four addresses gave 336 reviews, and 147 had 1 to 3 stars.
> - Claude found 8 problem types. The top three main reasons: defects (33), slow repair or missing parts (28), and warranty refused (23).
> - A keyword-rule check agreed with Claude at kappa 0.24 to 0.72 per type. The price rule gave 43 false hits.
> - On the "top rated" page, 5 different kettle models share one rating count of 29,754.
> - One category costs $0.057279 in API fees for 2 shelf calls, 3 store searches and 6 review pulls.

![Illustration: the left half shows an online shelf with nine kettles in different colours, each with a row of five full stars. The right half shows a snowy street repair window. People in the queue hold a kettle with a cracked lid, a kettle with a loose handle, a steam mop and a cracked blender. A clerk behind the counter holds repair tickets, and the parts shelf behind him is empty. An orange dashed line links one shelf tile to the window. The image has no text.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/img-shelf-vs-store.webp)

Every kettle on the shelf has five stars. The complaints start after delivery: repair, parts, warranty and shipping. [illustration made for this article](https://everyinfra.com/en/blog/yandex-market-reviews-claude-product-research)

## Why read complaints before you pick a product for Russia

A seller who enters the Russian small-appliance market must answer three questions. What makes buyers in this category give one star? For brands at the same price, is the problem in the product or in the service? Can your own supply chain and service plan avoid these problems? Positive reviews rarely help with any of these.

They say "works well" or "fast delivery". Negative reviews name the part that failed, the month it failed, and the number of days the repair shop kept it.

The timing matters in late 2026. Russia's platform economy law, No. 289-FZ, has applied since October 1, 2026 ([Federal Law No. 289-FZ](https://www.consultant.ru/document/cons_doc_LAW_511088/)). Under it, a marketplace can no longer change commissions on its own, block sellers without a reason, or fund discounts with the seller's money ([Glavbukh explainer](https://www.glavbukh.ru/art/392176-izmeneniya-v-rabote-marketpleysov)).

The sales calendar is also close. Yandex Market planned its Black Friday for November 25 to 27, later than Ozon on November 13 and Wildberries on November 16. Before its Black Friday, Ozon ran an "11.11" sale from October 22, 2026, to November 12, 2026. More than 600,000 sellers joined the Ozon November sales, 16% more than one year before ([Komsomolskaya Pravda, citing RIA Novosti](https://www.kp.ru/online/news/7244033/)). Finish the product research and the complaint labels before November 25, so that your stock arrives in time.

We chose electric kettles for three reasons. First, it is a high-volume category on Yandex Market, and the top-selling page has kettles from 690 to 13,110 rubles. Second, kettle failures cluster in a few places: leaks, lids, handles and temperature control. That makes the results easy to compare with other categories.

Third, our earlier article already covers sourcing from 1688 and the margin on Yandex Market ([Claude Code for 1688 sourcing and Yandex Market margin](/en/blog/claude-code-1688-sourcing-yandex-market-margin)). This article does not repeat the cost and margin math. It answers a different question: which kettle to pick and what to avoid.

One condition applies before you start. Russia is not on Anthropic's list of supported countries and regions. Kazakhstan, Uzbekistan, Kyrgyzstan, Armenia and Tajikistan are on the list ([Anthropic supported countries](https://www.anthropic.com/supported-countries)). The workflow below is for teams in a supported region, for example a Russian-speaking operations team in Almaty or Tashkent. Anthropic's terms decide which company can sign up and use Claude. This article does not give legal advice on that point.

![Animated flow. Step 1, read the shelf: yandex/market returns price, rating and rating count, 20 rows top-selling and 20 rows top-rated. Step 2, find brand stores: yandex/search with the brand name plus service centre; check the business category and drop name clashes. Step 3, pull complaints: yandex/reviews with lowest_rated, up to 100 rows per store, keep 1 to 3 stars. Step 4, Claude labels 8 problem types with a main reason and all reasons, and keyword rules check the labels. Step 5, table out: count per type, a Russian quote and summaries in Chinese, English and Spanish. The bottom row shows yandex/market 2 calls $0.044778, yandex/search 3 calls $0.004167, yandex/reviews 6 calls $0.008334, total $0.057279.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/fig-01-review-flow.en.light.svg)

## What the yandex market api can give you: three different doors

Type "yandex market api" into a search box. The suggestions include documentation, orders, swagger, github and key. Two more suggestions are "yandex marketplace api" and "yandex market seller api." These words point to three different products. Each one gives a different answer to the main question: can you read the review text of a competitor's product?

The first door is the Partner API for sellers. Merchants who sell on Yandex Market use it to manage orders, stock, prices and reviews. Its review method, `POST v2/businesses/{businessId}/goods-feedback`, returns reviews of your own goods. A page holds up to 50 reviews, and the time window is up to 6 months.

A `paid` filter separates reviews that got Plus points as a reward ([getGoodsFeedbacks reference](https://yandex.ru/dev/market/partner-api/doc/ru/reference/goods-feedback/getGoodsFeedbacks)). The changelog has entries on October 20 and October 27, 2026 ([Partner API changelog](https://yandex.ru/dev/market/partner-api/doc/ru/changelog/)). The OpenAPI specification is on GitHub, and you can generate a client in most languages from it ([yandex-market-partner-api](https://github.com/yandex-market/yandex-market-partner-api)). For product research, the limit is in the words "your own goods". If you have no store yet, or you want to read a rival's reviews, this API cannot help.

The second door is the Content API. Yandex's terms say that since December 26, 2018, it is open only to existing users with extended access and to marketing partners. It needs registration and payment ([Content API terms](https://yandex.ru/legal/market_api_content/ru/)). The Yandex Market terms of use, in the version dated October 12, 2026, still mention it ([Yandex Market terms of use](https://yandex.ru/legal/market_termsofuse/)). A new team cannot get access.

The third door is to build your own yandex market parser. On the Russian Stack Overflow, one user requested a product page with Python and got HTTP 403. The same request with curl also got 403, but a browser got a 302 redirect ([ru.stackoverflow question 1153371](https://ru.stackoverflow.com/questions/1153371)). The server checks whether a request looks like a person. Our YouTube search for "парсер яндекс маркета" returned 10 videos, and 8 of them demo or sell a scraping tool. When the page layout or the bot rules change, a home-made parser needs new code.

The EveryInfra `yandex/market` capability is a fourth option. It takes a keyword, a city and a sort order and returns product rows. Each row has `title`, `price`, `currency`, `rating`, `rating_count`, `shop_name`, `stock`, `is_available` and a product URL. One call costs $0.022389 ([EveryInfra capability catalog](https://api.everyinfra.com/api/v1/social/catalog), [Yandex data API](/en/yandex-api), [Yandex Market capability](/en/yandex-api/market)). The catalog note says that `review_count` and `purchase_count` are usually empty. In our two calls of 20 rows each, `review_count` was empty in all 40 rows, and no row had review text.

| Door | What you get | Who can use it | Review text for rival products? |
| --- | --- | --- | --- |
| Partner API goods-feedback | Reviews of your own goods, 50 per page, up to 6 months | Sellers with a store | No, your own only |
| Content API | Product and review data | Existing users and marketing partners since 2018-12-26 | Not for new users |
| Your own parser | Anything the page shows | Anyone who writes it | Yes, but you handle 403 errors and layout changes |
| EveryInfra `yandex/market` | Price, rating, rating count, shop, stock | Anyone with an API key | No, rating and count only |
| EveryInfra `yandex/reviews` | Review text for Maps businesses (stores, service centres) | Anyone with an API key | Yes, but about the store, not a product card |

![Screenshot of a yandex/market JSON response from EveryInfra. The request is POST api.everyinfra.com/api/v1/social with platform yandex, action market, keyword электрический чайник, city Moscow, sort popular, limit 20. The response id is req_73b7d87b1f684677. Two rows are shown: Polaris PWK1725CGLD at 8990 rubles with rating 4.8, and Kitfort KT-8762 at 690 rubles with rating 4.7. Teal boxes mark rating_count 833 and 4982 and charged_usd 0.022389 in billing. The rows have no review_count and no review text.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/img-market-response.webp)

A live response from 00:17:05 SGT on November 16, 2026. Two of the 20 rows and the billing field are shown. [EveryInfra REST response screenshot](https://api.everyinfra.com/api/v1/social/catalog)

## yandex market parser or API: what each choice costs you

The Russian suggestions tell you who searches for these terms. "яндекс маркет парсер" leads to "парсинг." "парсинг отзывов яндекс маркет" leads to "python" and "маркета python." Some of these people look for a way to write their own code. A home-made parser has two costs. The first cost is blocking. The Stack Overflow case shows a 403 for scripts and a 302 for browsers on the same URL.

The second cost is upkeep. Yandex Market renders product pages in the browser. Review pages use paging, and reviews merge across product variants. When any of these rules change, your parser breaks.

An API removes both costs, but you get only the fields that the API returns. We split the job into two parts. Shelf data, which means price, rating and rating count, comes from `yandex/market`. Review text comes from `yandex/reviews`, which reads reviews of Maps businesses such as stores and service centres.

The catalog says that the `url` parameter must be a Maps business URL and not a Yandex Market product URL. At 00:18:01 SGT we sent a product URL on purpose to test the limit. The call returned HTTP 503 and cost nothing. This capability accepts only Maps business URLs.

This is the first shelf call, sorted by sales in Moscow:

```bash
curl -s https://api.everyinfra.com/api/v1/social \
  -H "Authorization: Bearer $EVERYINFRA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"platform":"yandex","action":"market",
       "params":{"keyword":"электрический чайник","city":"Moscow",
                 "sort":"popular","limit":20}}'
```

You can also call the same capabilities from inside Claude. Add the EveryInfra remote MCP server as a custom connector. Claude custom connectors use remote MCP. Claude connects from Anthropic's cloud, so the server must be reachable on public port 443.

In a chat, open "+" and then Connectors to turn it on ([custom connectors with remote MCP](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp), [use connectors](https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities)). Remote MCP uses the Streamable HTTP transport, and each JSON-RPC message is one HTTP POST ([MCP transports](https://modelcontextprotocol.io/specification/2025-06-18/basic/transports)). The tool list and tool call formats are in the [MCP tools specification](https://modelcontextprotocol.io/specification/2025-06-18/server/tools).

At 00:29:54 SGT we sent `tools/list` and got 12 tools back. The next four calls, for the Redmond and Kitfort reviews, went through the `everyinfra_call_api` tool. Developers who use the Messages API can attach the same server with the [MCP connector](https://platform.claude.com/docs/en/agents-and-tools/mcp-connector). Our endpoint and auth options are on the [MCP server page](/en/mcp-server).

## The shelf: 20 top-selling kettles in Moscow

The top-selling call at 00:17:05 SGT returned 20 rows with 19 distinct URLs, because one URL appeared twice. Prices ran from 690 to 13,110 rubles, with a median of 4,799 rubles. Ratings ran from 4.5 to 5.0, and rating counts from 43 to 10,154. We picked five kettles that cover the low and middle price bands and two brand-owned stores:

| Model | Price (RUB) | About USD | Rating | Ratings | Shop |
| --- | --- | --- | --- | --- | --- |
| Kitfort KT-8762 | 690 | $8.40 | 4.7 | 4,982 | Kitfort |
| Comfee CF-KT7022 | 1,990 | $24.22 | 4.8 | 4,001 | Midea official store |
| Tuvio TKG0117 | 4,473 | $54.45 | 4.9 | 4,767 | Yandex Fabrika |
| REDMOND KG228 | 4,799 | $58.42 | 4.7 | 627 | not in the response |
| Polaris PWK 1753CGL | 4,990 | $60.74 | 4.9 | 835 | Polaris brand store |

We converted rubles at the Bank of Russia official rate for November 14, 2026: 1 USD = 82.1516 RUB ([Bank of Russia daily rates](https://www.cbr.ru/scripts/XML_daily.asp)). The five picks differ by 0.2 stars at most. The chart puts all 20 rows on one plane, with rating count on the horizontal axis and rating on the vertical axis. The points form a thin band. Rating counts go from tens to ten thousand, but the ratings stay almost the same. These two numbers alone cannot separate the products.

![Scatter chart of the 20 top-selling rows. The horizontal axis is rating count on a log scale from 10 to 20,000. The vertical axis is rating from 4.4 to 5.1. All points sit between 4.5 and 5.0. Teal marks the 5 picks: Kitfort KT-8762 at 4,982 ratings and 4.7, Comfee CF-KT7022 at 4,001 and 4.8, Tuvio TKG0117 at 4,767 and 4.9, REDMOND KG228 at 627 and 4.7, Polaris PWK 1753CGL at 835 and 4.9.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/fig-03-shelf-scatter.en.light.svg)

The `stock` field gives one more signal. Tuvio TKG0117 showed a stock of 3, Polaris PWK 1753CGL showed 12, and Kitfort KT-8762 showed 35. Low stock can mean fast sales or slow restocking. Use it as a flag, and look for complaints about cancelled orders or empty shelves. The `shop_name` field was empty in 8 of the 20 rows, and REDMOND KG228 was one of them. To see who sells that kettle, you must open the product page.

## Rating counts belong to model groups: 5 models share 29,754

At 00:25:19 SGT we sorted by "top rated" instead. The 20 rows had only 12 distinct URLs. One URL appeared in 7 rows, and those rows were 7 different Polaris models. Next, look at `rating_count`.

Five models, including PWK 1716CGLD, PWK 1775CGLD and PWK 1755CAD, all returned 29,754. Four other Polaris models all returned 8,860. Carrera No. 580 and No. 581 both returned 140.

![Horizontal bar chart of the 20 top-rated rows grouped by rating_count. 29,754 ratings are shared by 5 models, 8,860 by 4 models and 140 by 2 models. The counts 497, 433, 242, 203, 160, 133, 47, 46 and 43 each belong to 1 model. The 20 rows have only 12 distinct URLs.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/fig-04-card-share-bar.en.light.svg)

Yandex seller help states a related rule in its section on reviews for points. When a product card has several variants, Market counts the sum of reviews across all variants. The same page says that a product card shows reviews from that seller's buyers only ([Yandex Market seller help: reviews and rating](https://yandex.ru/support/marketplace/ru/marketing/plus-reviews)). The shared counts in our data match this rule.

A rating of 4.8 from 29,754 ratings belongs to a group of kettles. The group includes a 16,699-ruble temperature-control model and a 26,699-ruble smart model. You cannot say that one of these kettles has 4.8 stars from almost thirty thousand buyers.

The same model name can also appear at different prices. PWK 1753CGL was 4,990 rubles on the top-selling page. On the top-rated page, it appeared once at 3,990 rubles and once at 7,699 rubles, under different URLs. In your research sheet, use "model plus seller" as the key. Treat the rating count as context only.

Read the "top rated" sort with care too. The first 11 rows were small cards at 4.9 or 5.0 with 43 to 497 ratings. The large Polaris cards with tens of thousands of ratings came after them. A 5.0 from a few dozen ratings tells you little. Prices also differ between the two sorts.

The median price on the top-selling page was 4,799 rubles (about $58.42). On the top-rated page it was 9,999 rubles (about $121.71), and the highest was 26,699 rubles. The top-rated page holds glass kettles, temperature-control kettles and Italian and German brands. If your price is under 2,000 rubles, compare yourself with the top-selling page.

## Where the review text comes from: brand stores on Yandex Maps

The product cards do not give us review text, so we went to a place that does: the brand's own stores and service centres. In Russia, a buyer with a broken appliance often takes it to the brand's service centre or back to the brand store. Many of them then rate that place on Yandex Maps. These reviews rate the store and the service. They also say what the buyer bought, what broke and who handled it.

At 00:18:26 SGT we searched `yandex/search` for "сервисный центр Polaris" in Moscow and got 8 businesses ([Yandex Maps search capability](/en/yandex-api/search)). Two were Polaris stores: Kommunarka, rated 4.5 with 142 reviews, and Khoroshevskoe shosse, rated 4.5 with 44 reviews. The rest were general repair shops and coffee-machine repair shops. We pulled reviews with `lowest_rated` at 00:19:32 and 00:22:00 SGT and got 100 and 44 rows ([Yandex reviews capability](/en/yandex-api/reviews)). Of these, 66 had 1 to 3 stars.

The REDMOND search went wrong in a useful way. At 00:28:28 SGT, "REDMOND фирменный магазин" returned 10 businesses, all named Redmond, all in Moscow or the Moscow region. We pulled the two with the most reviews, 29 and 19, at 00:30:01 and 00:30:26 SGT. The reviews talked about backpacks, suitcases and TSA locks. The category of these stores was "bags and luggage," and the website belonged to a different brand.

They only share a name with REDMOND, the kitchen-appliance maker. We left these 48 reviews out. Their newest entries were from 2026-10-24 and 2026-10-20. Before you pull reviews, read the `categories` and `website` fields in the search result, and drop any name clash.

The Kitfort search "Kitfort сервисный центр" in Moscow returned nothing, and the empty result cost nothing. At 00:31:52 SGT we searched for "Kitfort" alone. Both results were in Saint Petersburg. One was the service centre and store on Zanevsky prospekt, with the categories "appliance repair" and "appliance store," a rating of 4.4 and 92 reviews.

The other was the online-store office on Instrumentalnaya street, which showed 170 reviews and returned a `rating` of 0. We pulled 92 and 100 rows at 00:32:32 and 00:32:58 SGT. Of these, 81 had 1 to 3 stars. Kitfort is based in Saint Petersburg, and one review calls this service centre the only one in the city.

![Screenshot of a JSON response from the EveryInfra remote MCP server. The request is tools/call with the tool everyinfra_call_api, platform yandex, action reviews, the Maps URL of the Kitfort service centre in Saint Petersburg, sort lowest_rated and limit 100. The response id is req_f4cfb303dcf84ab4. One of 92 rows is shown: rating 1, posted 2025-05-27, language RU. The text says a kettle bought for 2,400 rubles worked for one month and the repair shop never replied. The author name is hidden. Teal boxes mark rating 1 and charged_usd 0.001389.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/img-reviews-mcp-response.webp)

A live call through the remote MCP server at 00:32:32 SGT on November 16, 2026. We expanded the JSON inside the tool's text field, show 1 of 92 rows, and hid the author name. [EveryInfra remote MCP response screenshot](https://api.everyinfra.com/mcp)

| Store | City | Rows | 1 star | 2 stars | 3 stars | 1 to 3 stars | Owner replies | Date range |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Polaris, Kommunarka | Moscow | 100 | 46 | 3 | 6 | 55 | 98 | 2021-06 to 2026-11-04 |
| Polaris, Khoroshevskoe shosse | Moscow | 44 | 6 | 2 | 3 | 11 | 44 | 2022-12 to 2026-10-08 |
| Kitfort, service centre | Saint Petersburg | 92 | 24 | 5 | 2 | 31 | 77 | 2023-11 to 2026-10-05 |
| Kitfort, online-store office | Saint Petersburg | 100 | 40 | 3 | 7 | 50 | 13 | 2021-09 to 2026-10-18 |

All four stores show the same shape: many 1-star and 5-star reviews, and few in the middle.

Across the four stores, 116 of 336 reviews have 1 star and 166 have 5 stars. The two ends hold 84%, and 2 to 4 stars hold only 53 reviews. Hu, Pavlou and Zhang found the same J-shaped curve in Amazon ratings of books, DVDs and videos from 2005. In those three categories, 78%, 73% and 72% of ratings had 4 stars or more. They showed that with this shape, the average rating is a poor measure of product quality ([Hu et al., 2009](https://doi.org/10.1145/1562764.1562800)).

Polaris replied to almost every review, 142 of 144. The Kitfort service centre replied to 77 of 92, and the Kitfort office replied to 13 of 100. The reply rate is one more signal of how much a brand invests in service.

![Stacked bar chart of reviews pulled with lowest_rated at four brand stores. Polaris Kommunarka: 46 one-star, 3 two-star, 6 three-star, 7 four-star, 38 five-star. Polaris Khoroshevskoe: 6, 2, 3, 4, 29. Kitfort service centre: 24, 5, 2, 3, 57. Kitfort online-store office: 40, 3, 7, 8, 42.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/fig-05-store-dist-bar.en.light.svg)

These reviews connect back to Yandex Market. At least 3 of the 81 Kitfort complaints say the buyer ordered on Yandex Market. One says "купил в яндекс-маркете кофемашину," and another says "Купил паровую швабру Kitfort KT-1015 через Яндекс Маркет." Two more mention Yandex delivery. The Maps reviews fill part of the gap that the product cards leave.

## Claude labels 147 Russian complaints into 8 types

The 147 low-star reviews hold 71,740 characters of text: 35,838 for Polaris and 35,902 for Kitfort. The current Claude models have a 1 million token context window and up to 128k output tokens ([context windows](https://platform.claude.com/docs/en/build-with-claude/context-windows)). This text fits in one request with a lot of room. You do not need to split it or summarise it first.

Russian reviews contain typos, slang and swearing, for example "полисос" for a vacuum cleaner and "горантии" for warranty. Keyword matching handles these badly. A language model reads them more reliably.

Our instruction to Claude was short. Give each review one main reason and a list of all reasons. Use only the 8 types, and use "other" when unsure. We set the types after reading the first 30 reviews. The English summaries in this table are what we would hand to a buying team.

| Code | Type | Main reason | Mentioned | Russian quote (short) | English summary |
| --- | --- | --- | --- | --- | --- |
| D | Defects | 33 | 81 | «Через три месяца чайник потёк» | Fails within months. In kettles: lid, handle, thermostat. |
| S | Slow repair, no parts | 28 | 54 | «Ремонт длился 40 дней» | Repairs take 40 to 45 days. Parts are discontinued or slow. |
| W | Warranty refused | 23 | 28 | «гарантия не работает» | The shop calls it user damage. Paid extended cover fails. |
| L | Delivery, orders, stock | 21 | 29 | «доставку переносили 3 раза» | Moved dates, lost parcels, missing parts, no stock at pickup. |
| T | Staff and hotline | 17 | 44 | «Операторы не берут трубку» | Hotline does not answer. Staff mislead or are rude. |
| R | Refunds and exchanges | 12 | 27 | «Вернуть новый товар практически невозможно» | Returns need an expert report or a parcel to another city. |
| V | Reviews for gifts | 7 | 10 | «обещали мне весы за отзыв» | A gift for a good review, often never delivered. |
| P | Price | 4 | 11 | «Цены выше среднего» | Store prices above the website or the market. |
| O | Other | 2 | 2 | — | One 5-star text posted with 1 star; one "hard to find." |

The main-reason column adds up to 147. The "mentioned" column counts a review once for each type it names. We quote short phrases only and leave out author names.

The 3-star reviews can change the shares, so we counted again by star level. Of the 147 complaints, 116 have 1 star, 13 have 2 stars and 18 have 3 stars. The four after-sales types are warranty, repair and parts, delivery, and refunds. They are the main reason in 68 of the 1-star reviews (59%), 8 of the 2-star reviews (62%) and 8 of the 3-star reviews (44%). The 3-star group has more complaints about price (2), staff and hotline (3) and defects (3).

Two of the four price complaints have 3 stars. If you remove the 3-star reviews, after-sales is the main reason in 76 of 129 reviews. The share goes from 57% to 59%, and the order of the types stays the same.

Stars and text also do not always agree. The RuReviews authors studied 821,000 Russian product reviews. One short review text that means "fine" had 5 stars in 42.45% of cases, 4 stars in 36% and 3 stars in 29.25% ([RuReviews](https://github.com/sismetanin/rureviews)). So we use the stars only to select the reviews to read. We count the shares by the main reason that Claude found.

![Stacked bar chart of 147 Russian complaints by main reason. Defects 33 (Polaris 12, Kitfort 21). Warranty refused 23 (15 and 8). Slow repair or no parts 28 (6 and 22). Delivery and stock 21 (11 and 10). Refunds and exchanges 12 (7 and 5). Staff and hotline 17 (8 and 9). Reviews for gifts 7 (2 and 5). Price 4 (4 and 0). Other 2 (1 and 1). The title says 57% are about after-sales.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/fig-06-problem-share-bar.en.light.svg)

The two brands fail in different places. For Polaris, warranty refused is the top main reason, with 15 of 66 complaints. Nine Polaris complaints mention an extra warranty, and four name the paid store plan "Мощный сервис". The story repeats: the sales clerk promised a free swap if the device broke, and the service centre later called the fault user damage.

For Kitfort, repair and parts lead with 22 of 81, and defects follow with 21. The Kitfort complaints focus on two things. The same fault comes back after a repair, and parts such as filters, HEPA filters and coffee-maker glass jugs take months or never arrive.

### The last two years: the problems moved

The reviews span 2021 to 2026, and a brand's service rules can change in that time. So we also cut the data at November 16, 2024. Polaris has 27 complaints in the last two years. Warranty refused still leads with 9, staff and hotline has 4, repair and parts has 4, and defects has only 1.

In the last 12 months, Polaris has 11 complaints, and warranty again leads with 3. Kitfort has 40 complaints in the last two years: defects 11, delivery 9, repair and parts 8, and reviews for gifts 3. In the last 12 months it has 10, and delivery leads with 3. These complaints are about orders from the brand website that moved dates or got lost.

These samples are small, so read them as direction only. Polaris problems stay with the warranty sold in its stores. Kitfort problems moved a little, from "cannot fix it" toward "cannot deliver it". For a reseller of these brands, the Polaris risk is the warranty promise, and the Kitfort risk is durability and logistics. For a private label that competes with them, three promises answer the three largest after-sales types: a clear warranty, parts in stock and on-time delivery. Put all three on the product page.

We wrote the instruction to Claude in four parts. Part one sets the reader: "You help a small-appliance seller who plans to enter Russia. The readers are buyers and product managers." This keeps the summaries on what to check before buying. Part two lists the types. Each type has a one-line definition, one example and one near miss. For warranty refused, the near miss is "fixed under warranty but slowly," which belongs to slow repair.

Part three sets the edges. Judge only from the text. If a review attacks both the product and the staff, the main reason is the topic with the most words. If the stars and the text clearly disagree, use "other" and add a note. Part four sets the output: follow the schema, quote one Russian sentence, and remove names, phone numbers and email addresses.

The "most words" rule came from the first 30 reviews. One Russian complaint often covers three things: the device broke, the service centre kept it a month, and the hotline was rude. Without a rule, two runs on the same review can return different main reasons. Then the totals per type move between runs.

With the rule, each review has one main reason and any number of mentioned types, and we count the two columns apart. The rule on names matters too. Some reviews give a clerk's full name, and one gives the author's own email. That data must stay out of the sheet and out of any report.

Structured outputs make Claude return JSON that follows a JSON Schema. Required fields stay present and the type codes stay valid ([structured outputs](https://platform.claude.com/docs/en/build-with-claude/structured-outputs)). This is the schema we used:

```text
{
  "type": "object",
  "properties": {
    "labels": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "review_id": {"type": "string"},
          "primary": {"enum": ["D","W","S","L","R","P","T","V","O"]},
          "all": {"type": "array", "items": {"enum": ["D","W","S","L","R","P","T","V","O"]}},
          "product_hint": {"type": "string"},
          "quote_ru": {"type": "string"}
        },
        "required": ["review_id","primary","all","quote_ru"],
        "additionalProperties": false
      }
    }
  },
  "required": ["labels"],
  "additionalProperties": false
}
```

The `product_hint` field asks Claude to name the product in each review, such as "чайник" or "паровая швабра KT-1015". We used it later to find the kettle complaints. The `quote_ru` field holds one short Russian sentence, so a person can check the label against the source.

### The 11 kettle complaints

Eleven of the 147 complaints are about electric kettles: 5 for Polaris and 6 for Kitfort. Their failure types go straight into a sourcing checklist:

| Brand | Date | Stars | What happened | Type |
| --- | --- | --- | --- | --- |
| Kitfort | 2025-05-27 | 1 | A 2,400-ruble kettle (about $29.21) worked one month. The repair shop never replied. | Repair |
| Kitfort | 2025-02-17 | 1 | Two kettles at 1,700 and 3,800 rubles. One lid broke (thin plastic). One temperature sensor burned out. Repair cost more than a new kettle. | Defect |
| Kitfort | 2023-07-27 | 1 | Leaked from the bottom. 40 days in repair. The display then failed, and the brand refused a refund. | Refund |
| Kitfort | 2022-11-19 | 2 | A temperature-select kettle bought on Ozon failed within a year. | Defect |
| Kitfort | 2022-10-10 | 1 | After 6 months, the heat-aged plastic handle broke while the owner lifted a full kettle of boiling water. | Defect |
| Polaris | 2022-09-02 | 1 | Arrived broken. An exchange needed an expert report and a paid parcel to Kazan. | Refund |
| Polaris | 2022-03-20 | 1 | Arrived with parts missing. | Refund |
| Polaris | 2021-08-27 | 1 | Started to leak after three months. The hotline said all staff were busy. | Defect |
| Polaris | 2021-08-26 | 1 | The lid latch pops open and the rim of the fill opening is dented. | Defect |
| Polaris | 2021-08-10 | 1 | The clerk said the model had Wi-Fi. At home, it did not. | Staff |

The table has 10 rows for 11 complaints. The Kitfort review from 2025-02-17 appears at both Kitfort addresses with almost the same text, so we count it once. Two other complaints praise a kettle in passing. One says the electric kettle "has worked for a year and I am happy." Another says the only decent Kitfort products are simple ones, "a kettle, for example."

For sourcing, the kettle has three weak points: the lid hinge and latch, the joint between handle and body, and the electronics in temperature-control models. Temperature-control models seem to fail more than simple ones. Eleven complaints are too few to prove this, so treat it as a test item for samples. Three complaints describe a kettle that arrived broken or with parts missing, so plan an outbound check.

One more point is not about kettles but matters for a seller. Polaris asked one buyer to ship a return to Kazan, in Tatarstan. Kitfort has one service centre in Saint Petersburg. In Voronezh, 2 of the 3 listed Kitfort service points were not at their addresses. Before you sell into a city, check the complaints for a place that can repair your product there.

### Other categories in the same complaints

Store reviews do not stay in one category. We counted Russian word stems to see which products the 147 complaints mention. The result is a shortlist for your next category:

| Product (Russian stem) | Complaints | Polaris | Kitfort | Repeated problem |
| --- | --- | --- | --- | --- |
| пылесос, vacuum cleaner and robot vacuum | 23 | 15 | 8 | Brush heads, batteries, auto shut-off; parts "not sold separately" |
| отпариватель, паровая швабра: steamer, steam mop | 15 | 1 | 14 | No steam; fails again after repair |
| кофемашина, кофеварка: coffee machine | 13 | 5 | 8 | 40-day repairs; jugs and frothers hard to buy |
| чайник: electric kettle | 13 | 5 | 8 | Leaks, lids, handles, thermostats; 2 praise the kettle |
| блендер, соковыжималка, миксер: blender, juicer | 9 | 1 | 8 | Burning smell, torn filter mesh, "user overload" verdicts |
| увлажнитель: humidifier | 5 | 3 | 2 | Fails within two years; filter cartridges out of stock |
| мультиварка, хлебопечка: multicooker, bread maker | 4 | 2 | 2 | Parts stop when the model stops |
| утюг: iron | 2 | 2 | 0 | Steam button fails; dies right after warranty ends |

Vacuum cleaners lead the Polaris complaints. Steamers and steam mops lead the Kitfort complaints. If your next category is one of these two, this data already shows the risks. For vacuums, check brush-head warranty and parts. For steamers, test the steam system for durability. For a closer look, run the shelf call again with the new keyword and add stores of other brands.

## Check Claude's labels with keyword rules

We checked Claude's labels with a separate set of Russian keyword rules with Python's `re` module ([re documentation](https://docs.python.org/3/library/re.html)). Each type has one regular expression that matches word stems. The warranty rule is `гаранти|мощный сервис`.

The delivery rule is `доставк|курьер|заказ|самовывоз|в наличии|комплект|адрес`. We wrote the rules before we saw any comparison. Then we compared them, review by review, with the types that Claude marked as mentioned. We computed raw agreement and Cohen's kappa ([Cohen, 1960](https://doi.org/10.1177/001316446002000104)):

```text
RULES = {
  "W": r"гаранти|мощный сервис",
  "S": r"сервисн|ремонт|запчаст|комплектующ|деталь|детали|фильтр",
  "P": r"цен|дорог|дешевле|скидк|стоимост",
  "V": r"отзыв\w* за|за отзыв|приз|весы|подар\w* за",
}
for code, rx in RULES.items():
    hit = [bool(re.search(rx, r["text"].lower())) for r in rows]
    # compare with Claude's "all" labels: tp, fp, fn, tn and kappa
```

![Horizontal bar chart of kappa between keyword rules and Claude's labels. Defects 0.50 with 6 false hits. Warranty refused 0.60 with 16. Slow repair or no parts 0.67 with 13. Delivery and stock 0.53 with 22. Refunds 0.53 with 24. Staff and hotline 0.52 with 19. Reviews for gifts 0.72 with 2. Price 0.24 with 43.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/fig-07-rule-kappa-bar.en.light.svg)

Reviews for gifts reached 0.72, repair and parts 0.67, and warranty 0.60. On the common Landis and Koch scale, that is moderate to substantial agreement ([Landis and Koch, 1977](https://doi.org/10.2307/2529310)). Price reached only 0.24, and the rule gave 43 false hits. The quotes show why. The stem "цен" also matches "оценка" (a rating) and "ценник" (a price tag).

"Стоимость ремонта" means the cost of a repair, not the product price. The defect rule missed the most. Claude marked 81 reviews as mentioning a defect, but the rule found only 49. The rule missed phrases with no word for "broken," such as "прослужил меньше полугода" (lasted less than six months) and "одноразовая техника" (disposable appliances).

Keyword rules are a cheap spot check. Send only the reviews where the rule and the model disagree to a person. For price and defects, Russian stem rules give many false hits and misses, so let the model decide those two types.

The closest published work is by Kolmogorova and co-authors in 2025. They used language models on more than 12,000 Russian reviews of 15 museums from 9 platforms. The models first found positive and negative keywords, and then put them into 14 fixed categories ([Kolmogorova et al., 2025](https://doi.org/10.14529/jsfi250309)). The paper says that the output was mostly realistic, with some invented items. It gives no accuracy per category, so you cannot compare it with our kappa values. Both studies set the categories first and then let the model sort the text.

Our own labels have limits. Claude (Opus 5.5) labelled all 147 reviews in one read, and no second person labelled them. One Kitfort review praises a toaster but gives 1 star, and we labelled it "other."

## Reviews for gifts: discount the high scores

Reviews for gifts is the main reason in only 7 complaints, but it affects every rating on the shelf. Seven Kitfort complaints mention reviews for gifts. Six of them mention a card in the box that offered a kitchen scale for a review. Some buyers wrote the review and never got the scale. One went to the pickup address and found a locked door and phone numbers that did not answer.

The seventh Kitfort complaint does not mention the scale. It says that people give five stars to get the gift. One Polaris complaint says the store's "приз за отзыв" (prize for a review) was still missing after six months. Another says the store gave extra warranty for positive reviews, "that is why there are so many."

Yandex Market has its own review reward. A seller can give Plus points to a buyer who writes a review with a photo or video. One point equals one ruble.

The offer applies only to products with fewer than 1,000 reviews, and brands and Market itself can also give points ([Yandex Market seller help](https://yandex.ru/support/marketplace/ru/marketing/plus-reviews)). The `paid` filter in the Partner API marks these reviews. The platform rule rewards a review of any tone. The brand practice in these complaints asked for a positive review.

US rules give a useful reference. The Federal Trade Commission has a rule on consumer reviews. It calls an incentive unfair or deceptive when the incentive depends on positive or negative sentiment ([16 CFR 465.4](https://www.ecfr.gov/current/title-16/chapter-I/subchapter-D/part-465/section-465.4)). A University of Florida and University of South Carolina study used reviews of more than 5,000 Amazon products. Reviewers who got a free product rated it almost 0.5 stars higher on average, even when they disclosed the gift ([University of Florida news release](https://news.ufl.edu/2023/08/incentivized-online-reviews-deceive-consumers/)). On our shelf, the gap between 4.7 and 4.9 is only 0.2 stars, which is smaller than that bias.

## After-sales in Russia: the 45-day line

The complaints mention 40 and 45 days again and again. A coffee machine took 40 days to repair, and 8 of those days were transit. A robot vacuum was still in the shop after 45 days. The Kitfort service centre told one buyer to wait 21 to 45 days.

Article 20 of the Russian consumer protection law says that a written repair period agreed with the buyer cannot exceed 45 days ([Consumer Protection Law, Article 20](https://www.consultant.ru/document/cons_doc_LAW_305/c771be122fddbc6e3087cbf43aa39a85a960a296/)). Seven of the 147 complaints mention 40 or 45 days. One says the buyer filed a complaint with the consumer protection regulator.

For a seller, this line sets the cost of service. If a 690-ruble kettle needs a trip to another city and a 45-day repair, a replacement is cheaper. One complaint shows the math. The service centre quoted 4,600 rubles (about $55.99) to repair a steam mop that costs 6,500 rubles new (about $79.12).

The buyer opened it, found a failed thermal fuse, and bought two fuses on Ozon for 200 rubles (about $2.43). For low-price appliances in Russia, a swap often costs less than a repair. Put a swap rate into your price.

## Can AliExpress or YouTube fill the gap?

On AliExpress, a search for "electric kettle" at 00:18:16 SGT returned 18 products. We pulled reviews for 4 of them between 00:22:20 and 00:23:38 SGT and got 157 reviews ([AliExpress reviews capability](/en/aliexpress-api/reviews), [AliExpress data API](/en/aliexpress-api)). By the `country` field, only 1 came from Russia. Israel had 21, and South Korea and the US had 14 each. We called the global AliExpress site, and Russian buyers mostly order on the local AliExpress Russia site. Global reviews can show product faults, but they do not show what Russian buyers care about.

![Horizontal bar chart of 157 AliExpress reviews for 4 kettles by country. Israel 21, South Korea 14, US 14, Ukraine 13, Spain 13, Germany 11, Brazil 11, Italy 7, Chile 6, France 6, Russia 1, with Russia in orange.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/fig-09-ali-country-bar.en.light.svg)

On YouTube, a search for "какой электрический чайник выбрать 2026" at 00:18:45 SGT returned 10 videos. We pulled comments for two of them at 00:25:00 and 00:25:10 SGT ([YouTube comments capability](/en/youtube-api/comments), [YouTube data API](/en/youtube-api)). A review of a quiet kettle had 53 comments, and most said "thanks" or "good video". One commenter suspected that fake accounts were posting in the thread. The other video had 5 comments.

The `posted_at` field came back as relative text such as "4 months ago". Video comments show what people argue about, such as noise or how easy a glass kettle is to clean. They are not good for fault counts.

## Summaries in three languages, each written on its own

The sheet goes to three groups: Chinese-speaking buyers, English-speaking investors or partners, and a Spanish-speaking team in Latin America. Anthropic's multilingual guide says to state the output language in the system prompt. It also lists how each language performs against English ([multilingual support](https://platform.claude.com/docs/en/build-with-claude/multilingual-support)).

The benchmark comes from Global-MMLU ([arXiv 2412.03304](https://arxiv.org/abs/2412.03304)). We send one request per language. The prompt says: "Write directly for readers of this language. Do not translate another summary line by line." The three summaries can differ in length and in examples. This is the delivery type in English and Spanish:

| Language | Summary for the type "delivery, orders and stock" |
| --- | --- |
| English | Orders from the brand website move dates two or three times, and the brand and the courier blame each other. Gifts miss their date. Promise a delivery window you control. |
| Spanish | Los pedidos de la web de la marca cambian de fecha dos o tres veces, y la marca y el transportista se culpan entre sí. Los regalos llegan tarde. Promete un plazo que controles tú. |

![Illustration of a wooden desk from above. On the left, a laptop shows a sheet with coloured cells. Next to it is a pile of small review cards with one or two stars and wavy lines instead of words. Eight teal trays in two rows each have a sticker: a cracked kettle, a shield with a cross, a wrench, a delivery truck, a return arrow, coins, a speech bubble and a gift box. A hand puts a card into the wrench tray.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/img-sort-trays.webp)

Each complaint goes into one main tray, and the sheet also records the other types it mentions. The 8 trays match the 8 problem types. [illustration made for this article](https://everyinfra.com/en/blog/yandex-market-reviews-claude-product-research)

## Put it in a sheet: Claude Sheets or .xlsx

The labels must end up in a sheet that people can edit. Claude Sheets opened in beta on November 10, 2026, for all paid plans, with export to .xlsx and CSV. It is on by default for Team plans and off by default for Enterprise, where an admin turns it on ([get started with Claude Sheets](https://support.claude.com/en/articles/18324300-get-started-with-claude-sheets), [Anthropic announcement](https://claude.com/resources/articles/sheets-whiteboards-data), [TechCrunch, November 10, 2026](https://techcrunch.com/2026/11/10/anthropic-claude-sheets-whiteboards/)). Without Sheets, Claude can create .xlsx, .docx, .pptx and PDF files, and an uploaded file can be up to 30 MB ([create and edit files](https://support.claude.com/en/articles/12111783-create-and-edit-files-with-claude)).

Our workbook has three tabs. The "shelf" tab has one row per model and seller. Its columns are price, price in USD, rating, rating count, a flag for a shared rating count, and stock. The "complaints" tab has one row per review. Its columns are store, stars, date, main type, all types, product and the Russian quote.

Author names stay out. The "summary" tab counts each type per brand and holds the three language summaries. The tabs link by brand. A buyer reads the summary and opens the complaints tab only when a number looks odd.

## What to check for each of the five picks

Match the shelf data with the complaints, and each pick gets its own pre-purchase checks. This table uses only our live data and the types above. It is not a product review.

| Pick | What the shelf shows | Matching complaint risk | Check before you buy |
| --- | --- | --- | --- |
| Kitfort KT-8762 (690 RUB) | 4.7, 4,982 ratings, stock 35 | Brand complaints lead with defects and repair; kettle lids, handles and thermostats failed | Lid hinge and handle joint; boil a sample a few hundred times |
| Comfee CF-KT7022 (1,990 RUB) | 4.8, 4,001 ratings, Midea official store | No brand store found on Maps, so no complaints to read | Authorised repair points in Russia and parts supply |
| Tuvio TKG0117 (4,473 RUB) | 4.9, 4,767 ratings, stock only 3 | No brand store found on Maps; low stock on the shelf | Restock time; whether orders get cancelled when stock runs out |
| REDMOND KG228 (4,799 RUB) | 4.7, 627 ratings, no shop name returned | Maps search returned a luggage brand with the same name | The seller; a new store search with "техника" added |
| Polaris PWK 1753CGL (4,990 RUB) | 4.9, 835 ratings; also listed at 3,990 and 7,699 RUB | Warranty refusals and paid cover disputes lead; kettles leaked and a latch popped | Which seller and price go with which URL; the warranty terms |

We found no own store on Maps for Comfee or Tuvio. For these brands, a Russian buyer with a broken kettle needs a repair point. Have that answer before you list the product. To get their complaints, search for their authorised service centres. Or wait until your store is live, and use the Partner API on your own goods.

## The cost of one category

These are the 20 billed calls in this run. Failed calls, empty results and rate-limited calls cost nothing ([pricing](/en/pricing)):

![Horizontal bar chart of 20 billed calls by capability: yandex/market 2 calls $0.044778, yandex/reviews 6 calls $0.008334, aliexpress/reviews 4 calls $0.005556, yandex/search 3 calls $0.004167, youtube/search 2 calls $0.001556, aliexpress/search 1 call $0.001389, youtube/comments 2 calls $0.001112. Total $0.0669, and yandex/market is 67%.](https://everyinfra.com/articles/yandex-market-reviews-claude-product-research/fig-08-cost-share-bar.en.light.svg)

| Capability | Unit price | Billed calls | Subtotal |
| --- | --- | --- | --- |
| `yandex/market` | $0.022389 | 2 | $0.044778 |
| `yandex/reviews` | $0.001389 | 6 | $0.008334 |
| `aliexpress/reviews` | $0.001389 | 4 | $0.005556 |
| `yandex/search` | $0.001389 | 3 | $0.004167 |
| `youtube/search` | $0.000778 | 2 | $0.001556 |
| `aliexpress/search` | $0.001389 | 1 | $0.001389 |
| `youtube/comments` | $0.000556 | 2 | $0.001112 |
| Total |  | 20 | $0.066892 |

The useful core is 2 shelf calls, 3 store searches and 6 review pulls. That is $0.057279 per category, and the shelf calls are 78% of it. The review price does not change with the row count, so a `limit` of 100 costs the same as 50. The run also hit the per-minute rate limit 8 times, because several jobs shared one key. It also had one 503 and one empty result. None of these 10 attempts cost anything.

The Claude cost depends on the model. Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens. Opus 5.5 costs $4 and $20. Haiku 5.5 costs $0.10 and $0.50 for prompts up to 100k tokens ([Sonnet 5.5 overview](https://platform.claude.com/docs/en/models/sonnet-5-5/overview), [Haiku 5.5 overview](https://platform.claude.com/docs/en/models/haiku-5-5/overview), [pricing](https://platform.claude.com/docs/en/about-claude/pricing)). The pricing page also says that the newer tokenizer produces about 30% more tokens for the same text.

The Batch API halves the price ([batch processing](https://platform.claude.com/docs/en/build-with-claude/batch-processing)). We did not measure tokens for the Russian text. Count them before a run with the [token counting](https://platform.claude.com/docs/en/build-with-claude/token-counting) endpoint. Assume one token per character, which is cautious. Then one read of 71,740 characters costs about $0.14 in input with Sonnet 5.5. With Haiku 5.5 it costs about $0.0072.

A monthly check of 3 categories costs 3 × $0.057279 = $0.171837 in API fees. Store reviews change slowly. The two brands added only 21 complaints in the last 12 months. A quarterly check is enough, and after the first run a `limit` of 50 covers the new reviews.

## Limits of this method

First, Maps reviews rate stores and service, not a product card. They show a brand's service record and common faults in Russia. They do not show what buyers wrote under a product card on Yandex Market. For your own product reviews, open a store and use the Partner API.

Second, the time span is long. The 66 Polaris complaints run from 2021 to 2026, and only 11 are from the last 12 months. For Kitfort, 10 of 81 are from the last 12 months. Service rules may have changed, so split trends by year.

Third, the sample is two brands at four addresses. The REDMOND search found a luggage brand with the same name, so we have no REDMOND appliance store reviews. Tuvio and Comfee have no brand stores on Maps. The findings apply to these two brands and their cities only.

Fourth, reviews carry bias. Complaint writers are people who had trouble, and some positive reviews were paid with gifts. Use the type shares to set priorities, not as failure rates.

Fifth, Russia is not a supported region for Claude. Your team must be in a region that Anthropic's terms allow.

## How this fits with the 1688 margin article

Our earlier article uses Claude Code to compare 1688 supplier prices with Yandex Market prices and to work out the margin ([1688 sourcing and Yandex Market margin](/en/blog/claude-code-1688-sourcing-yandex-market-margin)). Use the two together. Run this complaint workflow first, and write the weak points into your supplier brief: lid, handle, thermostat and parts. Then compare suppliers on 1688 and work out the margin with the earlier method.

Use the complaint data to estimate your swap rate. For turning reviews into a data product, read [review data productization](/en/blog/review-data-productization). For an open-weights model on review labels, read [Kimi K3.1 Pro review classification](/en/blog/kimi-k3-1-pro-open-weights-review-classification).

## FAQ

### Does the yandex market api have official documentation?

Yes. The Partner API for sellers has documentation on the Yandex developer site, with changelog entries on October 20 and October 27, 2026. The OpenAPI specification is in the yandex-market/yandex-market-partner-api repository on GitHub. Only sellers with a store can use it, and its review method returns reviews of your own goods only. Source of this question: the search suggestions "yandex market api documentation" and "yandex market api swagger."

### How do I get a yandex market api key?

You create a Partner API key in the API settings of your seller account, so you need a store first. The Content API has not accepted new users since December 26, 2018. To read shelf data and store reviews, an EveryInfra API key can call `yandex/market` and `yandex/reviews`. Make your first call with the [quickstart](/en/docs#quickstart). Source of this question: the search suggestion "yandex market api key."

### Is there a yandex market parser on GitHub?

Yes, several, and the suggestion "yandex market parser github" shows that people look for them. On the Russian Stack Overflow, a user reports that Python and curl requests to a product page get 403, while a browser gets a 302. The server detects scripts, and parsers need updates after page changes. Check the date of the last commit before you use one. Source of this question: the suggestion "yandex market parser github" and ru.stackoverflow question 1153371.

### Can a "яндекс маркет парсер" collect product reviews?

A home-made parser can collect the reviews that a page shows, if you handle bot checks and paging. The EveryInfra `yandex/market` capability returns price, rating and rating count, with no review text. `yandex/reviews` reads reviews of Maps businesses. A Yandex Market product URL fails there. Our test at 00:18:01 SGT returned 503 and cost nothing. Source of this question: the Russian suggestions "яндекс маркет парсер" and "яндекс маркет парсинг."

### What can I use instead of "парсинг отзывов яндекс маркет" in Python?

If you are a seller, use the Partner API goods-feedback method: 50 reviews per page, up to 6 months back. If you have no store yet, start from Maps reviews of brand stores and service centres, as in this article. One REST call in Python returns 100 reviews. Source of this question: the suggestion "парсинг отзывов яндекс маркет python."

### Why do several models on Yandex Market show the same rating count?

When a product card has several variants, Market counts reviews across all of them. On the top-rated page, 5 Polaris models shared 29,754 ratings and 4 shared 8,860. Key your research sheet by model and seller, and treat the rating count as context. Source of this question: the suggestions "yandex market reviews" and "market yandex ru reviews."

### Are yandex market seller api and yandex marketplace api the same thing?

Both terms point to the Partner API for sellers. Its official name on the Yandex developer site is "API Яндекс Маркета для продавцов," and the GitHub repository uses the same line. It manages your store's orders, stock, prices and reviews. Reading a rival's shelf or store reviews is a separate task. Source of this question: the suggestions "yandex market seller api" and "yandex marketplace api."

## Get started

1. Sign up for EveryInfra and get an API key. Follow the [quickstart](/en/docs#quickstart) to call `yandex/market` with your target city.
1. Run the shelf call twice, with `sort` set to `popular` and then to `top_rated`.
1. Key the shelf rows by model and seller. Flag rows that share a `rating_count`.
1. Call `yandex/search` with the brand name plus "сервисный центр," or with the brand name alone.
1. Check `categories` and `website` in each result. Drop businesses that only share the name.
1. Call `yandex/reviews` for each store with `sort` set to `lowest_rated` and `limit` set to 100. Keep 1 to 3 stars.
1. Add the EveryInfra remote MCP connector in Claude ([MCP server](/en/mcp-server)), or upload the reviews as a file.
1. Ask Claude to label the reviews with the schema above.
1. Spot-check the reviews where the keyword rules and Claude disagree.
1. Ask Claude for a separate summary in each language. Export the result to Claude Sheets or .xlsx.

## References

1. Federal Law No. 289-FZ on regulation of the platform economy. КонсультантПлюс, read Nov 16, 2026. https://www.consultant.ru/document/cons_doc_LAW_511088/
1. New marketplace law in Russia from October 1, 2026 (explainer). Главбух, read Nov 16, 2026. https://www.glavbukh.ru/art/392176-izmeneniya-v-rabote-marketpleysov
1. Russian marketplaces announce Black Friday dates. Комсомольская правда, read Nov 16, 2026. https://www.kp.ru/online/news/7244033/
1. Supported countries and regions. Anthropic, read Nov 16, 2026. https://www.anthropic.com/supported-countries
1. Partner API: getGoodsFeedbacks. Yandex, read Nov 16, 2026. https://yandex.ru/dev/market/partner-api/doc/ru/reference/goods-feedback/getGoodsFeedbacks
1. Yandex Market Partner API changelog. Yandex, read Nov 16, 2026. https://yandex.ru/dev/market/partner-api/doc/ru/changelog/
1. yandex-market-partner-api: OpenAPI specification. GitHub, read Nov 16, 2026. https://github.com/yandex-market/yandex-market-partner-api
1. Yandex Market Content API terms of use. Yandex, read Nov 16, 2026. https://yandex.ru/legal/market_api_content/ru/
1. Yandex Market terms of use. Yandex, read Nov 16, 2026. https://yandex.ru/legal/market_termsofuse/
1. Parsing Yandex Market in Python returns 403. Stack Overflow на русском, read Nov 16, 2026. https://ru.stackoverflow.com/questions/1153371
1. EveryInfra capability catalog (GET /api/v1/social/catalog). EveryInfra, read Nov 16, 2026. https://api.everyinfra.com/api/v1/social/catalog
1. Get started with custom connectors using remote MCP. Anthropic Help Center, read Nov 16, 2026. https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp
1. Use connectors to extend Claude's capabilities. Anthropic Help Center, read Nov 16, 2026. https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities
1. MCP specification 2025-06-18: Transports. Model Context Protocol, read Nov 16, 2026. https://modelcontextprotocol.io/specification/2025-06-18/basic/transports
1. MCP specification 2025-06-18: Tools. Model Context Protocol, read Nov 16, 2026. https://modelcontextprotocol.io/specification/2025-06-18/server/tools
1. MCP connector for the Messages API. Anthropic, read Nov 16, 2026. https://platform.claude.com/docs/en/agents-and-tools/mcp-connector
1. Bank of Russia daily official exchange rates (XML). Банк России, read Nov 16, 2026. https://www.cbr.ru/scripts/XML_daily.asp
1. Yandex Market seller help: reviews and rating. Yandex, read Nov 16, 2026. https://yandex.ru/support/marketplace/ru/marketing/plus-reviews
1. Overcoming the J-shaped distribution of product reviews (Hu, Pavlou and Zhang 2009). doi.org, read Nov 16, 2026. https://doi.org/10.1145/1562764.1562800
1. Context windows. Anthropic, read Nov 16, 2026. https://platform.claude.com/docs/en/build-with-claude/context-windows
1. RuReviews: a Russian product review corpus. GitHub, read Nov 16, 2026. https://github.com/sismetanin/rureviews
1. Structured outputs. Anthropic, read Nov 16, 2026. https://platform.claude.com/docs/en/build-with-claude/structured-outputs
1. re: Regular expression operations. Python Software Foundation, read Nov 16, 2026. https://docs.python.org/3/library/re.html
1. A Coefficient of Agreement for Nominal Scales (Cohen, 1960). doi.org, read Nov 16, 2026. https://doi.org/10.1177/001316446002000104
1. The Measurement of Observer Agreement for Categorical Data (Landis and Koch, 1977). doi.org, read Nov 16, 2026. https://doi.org/10.2307/2529310
1. LLM aspect analysis of Russian museum reviews. doi.org, read Nov 16, 2026. https://doi.org/10.14529/jsfi250309
1. 16 CFR 465.4: Buying positive or negative consumer reviews. eCFR, read Nov 16, 2026. https://www.ecfr.gov/current/title-16/chapter-I/subchapter-D/part-465/section-465.4
1. Incentivized online reviews inflate product ratings, sales, even when disclosed. University of Florida, read Nov 16, 2026. https://news.ufl.edu/2023/08/incentivized-online-reviews-deceive-consumers/
1. Russian Consumer Protection Law, Article 20: repair deadlines. КонсультантПлюс, read Nov 16, 2026. https://www.consultant.ru/document/cons_doc_LAW_305/c771be122fddbc6e3087cbf43aa39a85a960a296/
1. Multilingual support. Anthropic, read Nov 16, 2026. https://platform.claude.com/docs/en/build-with-claude/multilingual-support
1. Global-MMLU. arXiv, read Nov 16, 2026. https://arxiv.org/abs/2412.03304
1. Get started with Claude Sheets. Anthropic Help Center, read Nov 16, 2026. https://support.claude.com/en/articles/18324300-get-started-with-claude-sheets
1. Sheets and whiteboards in Claude. Anthropic, read Nov 16, 2026. https://claude.com/resources/articles/sheets-whiteboards-data
1. Anthropic adds sheets and whiteboards to Claude. TechCrunch, read Nov 16, 2026. https://techcrunch.com/2026/11/10/anthropic-claude-sheets-whiteboards/
1. Create and edit files with Claude. Anthropic Help Center, read Nov 16, 2026. https://support.claude.com/en/articles/12111783-create-and-edit-files-with-claude
1. Sonnet 5.5 overview. Anthropic, read Nov 16, 2026. https://platform.claude.com/docs/en/models/sonnet-5-5/overview
1. Haiku 5.5 overview. Anthropic, read Nov 16, 2026. https://platform.claude.com/docs/en/models/haiku-5-5/overview
1. Pricing. Anthropic, read Nov 16, 2026. https://platform.claude.com/docs/en/about-claude/pricing
1. Batch processing. Anthropic, read Nov 16, 2026. https://platform.claude.com/docs/en/build-with-claude/batch-processing
1. Token counting. Anthropic, read Nov 16, 2026. https://platform.claude.com/docs/en/build-with-claude/token-counting