FreshCrawl/australian-supermarket-prices

Australian Supermarket Prices, September 2026

Every product Coles and Woolworths sold online in Australia on 17 September 2026, with barcodes on both sides so the two chains can be joined, plus all 2,019 of their supermarkets with coordinates.

from datasets import load_dataset

products = load_dataset("FreshCrawl/australian-supermarket-prices", "products", split="train")
stores = load_dataset("FreshCrawl/australian-supermarket-prices", "stores", split="train")

The files are Parquet, so barcode, product_id and store_id arrive as strings and keep their zero padding. The same snapshot is mirrored as CSV on Kaggle; if you read that copy with pandas, pass dtype={"barcode": str, "product_id": str, "store_id": str} or the barcodes become floats and stop joining.

What is in it

Config	Rows	Contents
products	48,867	Coles 23,767, Woolworths 25,100. Price, was price, unit price, special flag and type, barcode, department, availability
stores	2,019	Woolworths 1,142, Coles 877. Address, suburb, state, postcode, latitude, longitude, phone

barcode is a zero padded GTIN-13 and is the only key that joins the two chains. It is present on 21,633 Coles and 25,089 Woolworths rows. store_id joins products to stores.

products

Column	Type	Notes
chain	string	Coles or Woolworths
product_id	string	Retailer's own id, unique within a chain only
barcode	string	GTIN-13, zero padded, empty when not published
name, brand, size	string	As the retailer writes them
department	string	Top-level department, 37 distinct values across both chains
price, was_price, save_percent	float	AUD. was_price and save_percent are empty unless discounted
unit_price, unit_measure	float, string	Comparable price and its unit, for example 2.48 per 100g
on_special, special_type, offer_description	bool, string, string	Discount flag, the retailer's own deal type, and multi-buy wording
in_stock	bool	Orderable at collection time
store_id	string	Woolworths 1101, Coles 7674
product_url, image_url	string	Links to the retailer's page and image. Images themselves are not included
collected_date	string	2026-09-17 for every row

stores

chain, store_id, store_name, address, suburb, state, postcode, latitude, longitude, phone, store_url, collected_date. Coordinates are present on all 2,019 rows; store_url is Woolworths only.

A measurable duopoly

Matching rule: a product counts as sold by both chains when the same barcode carries a price on both sides. 9,954 products qualify. Splitting them by whether either chain had the product on special:

Comparison	Products	Priced identically to the cent
Neither on special	4,179	3,794 (90.8%)
One or both on special	5,775	1,003 (17.4%)

Undiscounted shelf prices are effectively the same across the duopoly, and discounting is where the two chains actually diverge. On the 5,157 matched products whose prices differ, the median gap is $2.00; Coles is cheaper on 25.4% and Woolworths on 26.4%. Every number here was recomputed from the files in this repository.

Limitations and bias

• Single store per chain. Prices come from Woolworths store 1101 and Coles store 7674, the default online stores. Australian grocery prices vary by store and region, so these are a national online baseline, not local shelf prices. Any geographic conclusion drawn from the stores file plus these prices is an assumption, not a measurement.
• Point in time. One day, 17 September 2026. Specials in Australia run Wednesday to Tuesday, so the discount columns capture one week of one cycle and generalise poorly.
• Barcode coverage is uneven. 91% of Coles rows and 99.96% of Woolworths rows carry a barcode, so matched-product analysis is biased toward products Coles publishes a barcode for, which skews to national brands over own-label.
• Selection. Online ranges only. Products sold in store but not online are absent, as are age-restricted categories the retailers do not list to guests.
• No text or nutrition. Descriptions, ingredients, nutrition panels and customer reviews exist in the live sources and are deliberately excluded here, so this file does not support NLP on product copy.
• No personal data. Product and store attributes only. Nothing in this dataset identifies a person.

Source

Collected with two public Apify actors that return this data live, per store, including the fields left out here:

• Coles Scraper
• Woolworths Scraper

This is a snapshot. It is not updated.