Datasetpublic

FreshCrawl/capterra-b2b-software-reviews

Capterra B2B Software Reviews 56,606 B2B software reviews from Capterra, covering 66 products across 11 software categories. Most public review datasets are star rating + review text. This one carries five separate rating dimensions, pros and cons as distinct pre-split fields, reviewer firmographics, and, unusually, an incentive disclosure flag recording whether the reviewer was given a gift card, referred by the vendor, or wrote the review unprompted. Why this is… See the full description on the dataset page: https://huggingface.co/datasets/FreshCrawl/capterra-b2b-software-reviews.

sourceHugging Faceotherupdated 22d agoView on Hugging Face

0likes94downloads

Dataset Card

Capterra B2B Software Reviews

56,606 B2B software reviews from Capterra, covering 66 products across 11 software categories.

Most public review datasets are star rating + review text. This one carries five separate rating dimensions, pros and cons as distinct pre-split fields, reviewer firmographics, and, unusually, an incentive disclosure flag recording whether the reviewer was given a gift card, referred by the vendor, or wrote the review unprompted.

Why this is useful for ML

—Aspect-based sentiment. pros and cons arrive as separate fields, split by Capterra itself rather than extracted by a model. That is 56,606 naturally labelled positive/negative text pairs about the same product from the same author, which is unusually clean supervision for ABSA.
—Multi-target regression. Six numeric targets per review (overall, ease of use, customer support, functionality, value for money, and a 0-10 recommendation score) over shared text.
—Bias and fairness research. The incentive_type column is the rare part. It lets you study solicitation bias in review corpora directly rather than inferring it.
—Firmographic segmentation. Job title, company size band, industry, and product tenure on every row.

At a glance


Rows	56,606
Products	66
Software categories	11
Reviewer industries	148
Review date range	2010-01-15 to 2026-08-21
Snapshot collected	2026-08-22
Mean overall rating	4.56 / 5
Rows carrying an incentive	57.1%
Rows with a vendor response	3,369
Rows naming a product switched from	8,989

The incentive column

57.1% of these reviews were incentivised in some form, disclosed per review in four flavours:

`incentive_type`	Rows	Meaning
`NominalGift`	27,743	Invited, offered a small gift card
`NoIncentive`	24,257	Wrote the review unprompted
`VendorReferredIncentivized`	4,474	Referred by the vendor and incentivised
`VendorReferred`	125	Referred by the vendor, no incentive
`NonNominalGift`	7	Larger, non-nominal incentive

A first pass at the obvious question gives a counter-intuitive result. Comparing incentivised against organic reviews within the same product, which controls for the product simply being good:

—Gift-card reviews: +0.06 stars. Higher in only 37 of 63 products. Essentially noise.
—Vendor-referred reviews: +0.32 stars. Higher in 15 of 16 products with sufficient sample.

So the intuitive story, that gift cards buy good ratings, does not hold up here. Who gets asked looks like it matters more than what they are handed. Treat that as a starting point, not a conclusion: vendor referrals concentrate in a handful of products, and the above is a simple mean comparison with no controls for reviewer segment or recency.

python

from datasets import load_dataset

ds = load_dataset("FreshCrawl/capterra-b2b-software-reviews", split="train")
print(ds)

Fields

Column	Type	Description
`review_id`	string	Unique review identifier. Primary key, no duplicates
`product_name`	string	Product name as listed on Capterra
`product_category_group`	string	Coarse category. Our grouping, not Capterra's taxonomy
`product_capterra_id`	int	Capterra numeric product id
`review_date`	string	ISO 8601 date the review was written
`review_title`	string	Reviewer's headline
`overall_rating`	float	Overall stars, 1-5
`ease_of_use_rating`	float	Sub-rating, 1-5
`customer_support_rating`	float	Sub-rating, 1-5
`functionality_rating`	float	Sub-rating, 1-5
`value_for_money_rating`	float	Sub-rating, 1-5
`recommendation_score`	float	Likelihood to recommend, 0-10
`pros`	string	What the reviewer liked. Pre-split by Capterra
`cons`	string	What the reviewer disliked. Pre-split by Capterra
`general_comments`	string	Overall commentary. Empty for ~32%
`incentive_type`	string	One of the five values in the table above
`is_incentivized`	int	1 if `incentive_type` is not `NoIncentive`
`reviewer_job_title`	string	Self-reported title. 17,548 distinct values
`reviewer_company_size`	string	Employee band, 9 values
`reviewer_industry`	string	148 distinct values
`reviewer_time_used_product`	string	Tenure bucket, 5 values
`reviewer_is_validated`	int	1 if Capterra validated the reviewer
`reviewer_verified_linkedin`	int	1 if verified via LinkedIn
`review_is_anonymous`	int	1 if posted anonymously
`alternatives_considered`	string	Pipe-separated product names. Empty for ~91%
`switched_from`	string	Pipe-separated product names. Present on 8,989 rows
`reasons_for_choosing`	string	Free text. Empty for ~90%
`reasons_for_switching`	string	Free text. Empty for ~86%
`vendor_responded`	int	1 if the vendor replied publicly
`vendor_response_date`	string	ISO 8601, empty if no reply
`vendor_response_text`	string	Vendor's public reply
`review_url`	string	Source Capterra review page

Collection method

—Collected directly from Capterra product review pages on 2026-08-22.
—Curated selection of 66 well-known B2B products across 11 categories, chosen for variance in vendor size and review-solicitation practice. Not a random sample of Capterra, and not a census.
—Up to 1,000 reviews per product, sorted most recent. Sorting by "most helpful" was deliberately avoided: that ordering is a curated, engagement-biased selection and would contaminate any analysis of the rating distribution.
—Products with under 1,000 total reviews are present in full; larger ones are truncated to their 1,000 most recent.

Limitations and bias

—Selection bias by design. 66 hand-picked well-known products. Ratings here will skew away from the long tail of small Capterra listings.
—Truncation bias. Popular products are capped at 1,000 recent reviews, so they are under-represented relative to their true review volume and skewed recent.
—Platform bias. Capterra is Gartner-owned and vendors pay for placement on it. The 57% incentivisation rate is itself evidence that the review population is solicited, not organic.
—Rating ceiling. Mean overall rating is 4.56 of 5. Like most review corpora this is heavily left-skewed, so accuracy on a naive classifier is misleading.
—English only, and the category grouping is ours rather than a standard taxonomy.

Privacy

Reviewer identity columns present in the source were removed before publication: display name and profile picture URL. What remains is non-identifying segment data.

Free-text fields are reproduced as written and may occasionally contain a first name inside a vendor's reply. The review text itself was not rewritten.

Licence

Review text remains the intellectual property of its individual authors and of Capterra. This compilation is published for research and educational use. Cite the dataset and link back if you publish work based on it. Do not redistribute the raw file as a commercial product.

Fresher data

This is a static snapshot from 2026-08-22 and will not be updated. It ages from the day it was published, which is fine for research and useless for anything operational.

The scraper that produced it is public:

—Capterra Review Scraper: same fields, live data, any Capterra product
—Software Review Scraper: the same shape across G2, Capterra, TrustRadius and Gartner in one run

Both return the identical schema used here, so code written against this file works unchanged against fresh data.