Apify Scraper Data Normalizer

Combine datasets from different Apify scrapers without writing a separate reader for each one. Give this Actor datasets of posts, comments, videos, products, property listings, or places. Get back records in a consistent format, ready for your next script or analysis tool.

See the output

Here is a shortened example using synthetic Instagram data. The full record has additional fields for dates, links, and checks on the data.

{
  "schema_version": "2.1.0",
  "record_type": "social_post",
  "identity": {
    "namespace": "instagram",
    "source_id": "3000000000000000001"
  },
  "data": {
    "text": "New research mug https://shop.example/products/research-mug?currency=CAD"
  },
  "measurements": [
    {"name": "likes", "value": 10, "source_field": "likesCount", "state": "reported"}
  ],
  "source": {
    "actor": "apify/instagram-scraper",
    "dataset_id": "demo-posts",
    "record_index": 0
  }
}

Instagram calls the post text caption and its like count likesCount. TikTok uses text and diggCount. After normalization, your code finds the text in data.text and the like count in a measurements entry named likes for either source. source_field keeps the original field name visible.

Products and comments use the same places for IDs and source details, with their own fields inside data. For example, you can load posts, comments, and product listings in one run, then build your own analysis on top of the results.

$0.50 per 1,000 saved records. Upstream scraper costs are separate.

How to use it

1. Run one or more supported scrapers to collect your data.
2. Remove the prefilled inline items. In the JSON editor, omit the items key; items: [] still selects inline input and conflicts with dataset input.
3. Choose output version 2.2.0 for all released source modes and input options. Existing 2.1.0 workflows remain supported; Maps and custom mappings require 2.2.
4. For one dataset, select datasetId and declare its exact sourceActor or mode. For several datasets, use both the top-level datasetIds picker and sources.

For one dataset, replace the placeholder below. Leave datasetIds, sources, items and csvText unset. This example accepts ordinary video-section posts, including slideshows, from the exact TikTok Profile Actor:

{
  "outputVersion": "2.2.0",
  "datasetId": "REPLACE_WITH_YOUR_DATASET_ID",
  "sourceActor": "clockworks/tiktok-profile-scraper",
  "maxRecords": 1000,
  "includeRaw": true
}

This example combines Instagram posts and comments. Replace the placeholder IDs with your dataset IDs and select those same datasets in the picker:

{
  "outputVersion": "2.1.0",
  "datasetIds": ["POST_DATASET_ID", "COMMENT_DATASET_ID"],
  "sources": [
    {
      "datasetId": "POST_DATASET_ID",
      "sourceActor": "apify/instagram-scraper"
    },
    {
      "datasetId": "COMMENT_DATASET_ID",
      "sourceActor": "apify/instagram-comment-scraper"
    }
  ],
  "includeRaw": true,
  "maxRecords": 1000
}

The selected dataset IDs must match the IDs in sources. Selecting a dataset grants read access; naming it in sources tells the Actor how to read its fields. Declare each dataset once. For this source-list input, leave top-level datasetId, payload, items, csvText, customMapping and sourceActor unset, and keep top-level sourceType at auto. Put each source declaration inside sources.

For an after-run integration, map datasetId to {{resource.defaultDatasetId}}, declare the exact scraper, select a matching output version and trigger only on ACTOR.RUN.SUCCEEDED. The picker grants scoped READ; a dataset reference inside payload.resource alone is not a verified grant. Automatic handoffs process completed scraper runs. Upstream release monitoring, automatic mapping repairs and automatic deployments are not enabled.

V1 remains the default for the original Instagram and TikTok post format. Choose version 2.1.0 explicitly for the wider format shown here. Existing 2.0.0 inputs keep their earlier fields and behavior. You can also supply up to 100 inline items per source. Declare the source Actor or mode for V2 inline input and leave dataset fields unset. The Actor accepts at most 100 sources, reads them in the order supplied, and inspects at most 10,000 rows in total. maxRecords defaults to 1,000 and includes invalid rows and repeats.

Inspection limits up to 1,000 use 512 MB automatically; larger limits use 2,048 MB. The default timeout is 300 seconds. Let the Actor choose memory for large runs. A manual low-memory override can require a longer timeout to finish the reports.

Version 2.2.0 adds Google Maps places: addresses, coordinates, categories, hours, closure flags, contacts, ratings, and supplied review samples. Existing sources use their 2.1 fields in this format. Supply Maps and Instagram datasets together to give your next tool business details alongside social posts. Supplied social profile links remain evidence; the normalizer does not discover or verify them. Shared accounts do not merge branches or attribute posts to a particular venue.

What stays true to the source

• Missing values stay unknown. A missing count does not become zero.
• Counts keep their meaning. A view count and a play count remain separate.
• Prices keep their known currency and price type. An unclear dollar symbol does not become USD, and a product's starting price is not a specific variant's price.
• Links come from IDs or URLs in the supplied data. A comment can link to its parent post; similar wording alone does not establish a product connection.
• Changed records and records from separate datasets stay separate. V2 removes exact repeated source objects within the same declared source.

With includeRaw: true, each saved record also includes the complete original JSON object received by the normalizer. The standard fields cover a supported subset of that object. Version 2.1 includes supplied Instagram comment previews and source identifiers such as variant barcodes, with their original field paths. A preview is not a complete comment collection; an identifier match is not proof that two listings describe the same product. Nested data stays inside its parent record rather than creating extra saved records. Rejected and repeated rows are not copied to the output, so keep your source datasets or exports if you need every original row.

This Actor structures data already collected by other tools. Analysis, ranking, and decisions about what the data means are up to your next tool. It does not scrape websites or send your data to an LLM.

Other exports and sources

V2 includes four spreadsheet downloads: records, counts and ratings, prices, and explicit links. They share observation IDs, so your next tool can read them together. Counts keep their definitions, prices keep their currencies, and spreadsheet text formatting preserves exact IDs. JSON remains the main output. The export manifest reports any rows omitted by the spreadsheet size limit.

For an unsupported or private scraper, version 2.2.0 accepts an explicit customMapping. Declare column paths, identity, conversions and measurement definitions instead of relying on automatic guesses. The first release supports posts, videos, comments, products, property listings and media titles. It does not support custom place mappings or arbitrary nested evidence structures. Supplying a mapping does not mean that scraper has passed our built-in integration tests.

You can also paste CSV into csvText, alongside that mapping and a declared sourceActor. CSV cells remain text until a declared conversion applies. Empty cells stay empty strings, and large IDs keep their digits. CSV accepts comma, semicolon, tab or pipe delimiters, with up to 8 MiB of text across sources. Clear the prefilled items first. Remote CSV URLs and file uploads are not supported. With includeRaw, CSV records retain parsed string cells, not original file bytes.

Where to find the results

Open the run's output dataset for normalized records. The run's key-value store also contains reports and spreadsheet downloads:

• OUTPUT: how many rows were read, saved, skipped, or removed as repeats.
• PRESERVATION: what happened to each inspected row and which sources were visited.
• COMPARISON: available counts and links found in the supplied data.
• FIELD_COVERAGE: confirmed field destinations and source fields that need inspection in the original data. Untraced does not mean discarded.
• TABLES: spreadsheet filenames, coverage limits and write status. The files are TABLE_RECORDS.csv, TABLE_MEASUREMENTS.csv, TABLE_PRICES.csv and TABLE_RELATIONSHIPS.csv.
• REJECTIONS: skipped rows explained in plain language, with repeats, unsaved records and source-reading failures listed separately. REJECTED_ROWS.csv lists rejected rows with source references, hashes and suggested next steps.

These additional artifacts are available for V2 runs. They do not create extra billable records. Rejection reports contain references and reasons, rather than copies of the rejected content. Keep your sources if you need to revisit a row.

A run can save results before a later source fails. Check its status and reports before using the output or retrying. Failed or forcibly stopped runs may not have final reports, so also inspect any records already saved.

Supported scrapers

Support covers these result types, rather than every option each scraper offers. Declare the exact Actor name below, or its supported mode.

Source Actor	Supported data
apify/instagram-scraper	Posts and Reels
apify/instagram-reel-scraper	Reels
apify/instagram-post-scraper	Posts, carousels, and Reels
apify/instagram-hashtag-scraper	Post and Reel results from hashtag searches
apify/facebook-posts-scraper	Posts from public feeds, with separate reaction and video counters
apify/facebook-comments-scraper	Comments and separate reply rows with explicit source references
apidojo/tweet-scraper	Tweets, explicit reply references, and supplied quote/repost link evidence
trudax/reddit-scraper-lite	Posts and comments, with reported scores and explicit thread references
clockworks/tiktok-scraper	Video posts and slideshows
clockworks/tiktok-profile-scraper	Ordinary video-section posts, including slideshows
apify/instagram-comment-scraper	Comments, including supported nested replies
clockworks/tiktok-comments-scraper	Comments and separate reply rows
streamers/youtube-scraper	Videos, Shorts, and streams
streamers/youtube-comments-scraper	Comments and replies
harvestapi/linkedin-profile-posts	Posts
harvestapi/linkedin-post-search	Post results from searches
apify/e-commerce-scraping-tool	Supported product and offer records
junglee/amazon-crawler	Products with separate price types
autofacts/shopify	Products with nested variants
maxcopell/zillow-detail-scraper	Individual property details
compass/crawler-google-places	Ordinary place search and detail rows, with embedded review samples; requires 2.2

Standalone profiles, channel summaries, search metadata, category-only rows, standalone reviews, transcripts, external-service hotel entries, and unrelated custom formats are outside these mappings. IMDb support remains experimental and outside the released coverage promise. TikTok profile stories, repost sections, follower lists, and separate LinkedIn search reaction or comment rows are outside the new post mappings. Facebook comments keep source IDs and explicit direct parent references. Replies remain separate records; source context does not replace the reported author.

Counts keep their source definitions. Facebook reaction totals are separate from LIKE reactions and its video-view thresholds remain separate. Reddit scores are not likes. X repost counters have an explicit warning when their scope is not established. Instagram hidden likes remain unknown. Quote and repost links retain source evidence without claiming a new relationship type. For fields outside the standard mapping, use includeRaw: true to retain the complete source JSON.

Place ratings and review counts remain separate from social engagement. Opening hours keep their supplied text; missing schedules do not mean closed. Price brackets stay text, and open status describes the source observation. Supplied review samples stay inside their place, with dates, ratings, translations, and owner responses. They do not create extra saved records or establish completeness.

Pricing

$0.50 per 1,000 normalized records saved to the output dataset, with no Actor subscription or start fee. Skipped rows, removed repeats, records that were not saved, and reports have no normalization event charge. Results saved before a run fails can still be charged. Upstream scrapers are priced separately.

Set a spending limit of at least $0.0005, the price of one record. Apify rejects smaller limits before starting a run. Allow headroom when testing completion: 10,000 records succeeded with a $5.10 cap. An earlier exact $5 cap ended as platform ABORTED after outputs were saved. Check terminal status and reports rather than assuming a full dataset means success. Owner-account platform usage does not verify an actual paid-customer event charge.

Data handling

The Actor reads only the datasets you select, using limited permissions, and leaves them unchanged. Inputs, output, reports, and platform logs are stored in Apify under the running account's settings. Review sharing and retention settings before submitting personal data. Turning off includeRaw does not remove personal data from other output fields.

You must have the right to process the supplied data and use the results. Supporting a scraper does not grant rights to its content or imply affiliation with its platform. Record hashes help trace the input; they are not copies, anonymization, or proof that the scraped information is true. Supply large IDs as strings, since precision lost before the Actor receives them cannot be recovered.

Support

Operated by Joe Finnegan. For private support, data-access, or deletion requests, contact jaywfinny@gmail.com. Use is governed by Apify's Standard Actor Contract.

For general support, use the Issues tab with the build, run status, scraper name, and a small synthetic example. Sharing a run link can give the developer access to its input, logs, and output, so check the data first. Keep tokens, private dataset IDs, personal information, and raw customer records out of public issues.