Two or more review sites displayed matching totals and ratings for all 12 Capterra-listed pure-play assessment products

Review counts are often cited for assessment products, so we recorded the totals and ratings that five review sites displayed on one day for a sample fixed in advance. We saved every page and report only what the review sites displayed.

This study makes no finding about any vendor. Every sentence below that describes a match uses a review site as its subject because the review sites displayed the figures.

The numbers

For 12 of 12 comparable products among the 12 Capterra-listed pure-play assessment products, two or more review sites displayed at least one exact matching review signature on 2 September 2026; the median R_p across the 12 comparable products was 0.481.

Capterra and GetApp each displayed a profile in all 12 matching groups, Software Advice displayed a profile in 10, and G2 and Trustpilot displayed profiles in none.

Across those 12 matching groups, the review sites displayed sub-ratings that yielded 7 corroborated classifications, 0 discordant classifications, and 5 unassessable classifications under the prespecified check; none establishes review identity.

The three sentences are designed to be quoted, and each has a fixed denominator. The sample comprises the 12 products classified as pure-play in Capterra’s pre-employment testing category on 29 July 2026. This frozen roster, fixed before any review page was read, is the same one used by the assessment pricing transparency index. “Comparable” means a product with at least two profiles that displayed both an integer review total and an overall rating; all 12 qualified. Two or more review sites form a “matching group” for one product when they display the same integer total and the same one-decimal rating, and when the listing names they display are identical or nested after normalisation.

The secondary numbers, in the order the study prespecified: coverage, F = 12 of 12 products with profiles found on two or more sites and A = 12 of 12 comparable products; the pooled volume-weighted share of displayed review-count entries classified as redundant under the exact-signature rule, 0.317; the median number of distinct displayed signatures per comparable product, 2.0; and the sensitivity run that excludes profiles with fewer than ten displayed reviews, which gives a median R_p of 0.482 and a pooled share of 0.317, with coverage unchanged.

How to cite this page

Quote the first headline sentence with the date and a link, and it stays checkable:

For 12 of 12 comparable products among the 12 Capterra-listed pure-play assessment products, two or more review sites displayed at least one exact matching review signature on 2 September 2026; the median R_p across the 12 comparable products was 0.481. (SharpAssessment, Assessment review cross-listing study, https://sharpassessment.com/assessment-review-cross-listing/).

The numbers are recomputed from the row data by a script, so if the rows change on the quarterly recheck the sentence changes with them and the change log notes the date.

Method

Unit of observation. One product brand from the frozen roster, on one review site, is one profile. For each product, we attempted to locate and capture one profile on each of the five sites on the same date. SharpAssessment publishes this page and has no row.

Sites. We observed Capterra, GetApp, Software Advice, G2 and Trustpilot so that the figures displayed across five sites could be compared for each product.

Roster, locked before any profile was read. Cohort 1, the headline frame, consists of the 12 pure-play products in Capterra’s pre-employment testing category on 29 July 2026; the headline table lists them. The pure-play classification is ours, and the thirteen exclusions are listed on the pricing index. Two sensitivity samples are reported separately and never in a headline: cohort 2, the 17 products with their own cost-shaped search demand, and cohort 3, the 3 products carried over from this site’s DISC comparison. Four appendix products enter no denominator. No product is dropped because a profile is absent or unreadable.

Discovery, three steps, each recorded per product and site. Step 1 was a Google organic search for “product site-name reviews” and, for Trustpilot, “product trustpilot”, run through a search data API configured for the United States. We accepted the first result whose URL matched the review site’s profile pattern and whose slug or title contained the product name; for Trustpilot, the reviewed domain also had to match the vendor’s own domain. Step 2: a Google organic search for “<product> site:<domain>”, run through the same search data API for the United States on desktop. Step 3, added after the first two had been run: a deterministic probe of each site’s fixed profile URL pattern with the product slug variants listed in the published roster script, fetched through a US proxy. A page was accepted only when its canonical URL, or final URL when no canonical was present, matched the site’s profile pattern, was not a comparison page, and its title or structured-data name carried a roster token. Since 2 September 2026, after publication, a Capterra, GetApp or Software Advice listing is also accepted only when its category sits in the HR family: for Capterra, the breadcrumb category in the page’s structured data; for GetApp and Software Advice, the first path segment of the URL. A listing outside that family is shown in the table as rejected, with its category and a link, and counts as not found; the saved page and its hash are kept. Step 3 exists because step 2 could not be executed as designed. For site-restricted queries, the API returned no pages from the restricted domain; a second search engine returned no results for any such query; and the original search engine’s site presented the browser with a bot-verification page. We did not complete that verification. Every probe attempt is saved as its own file with status, byte size, SHA-256, UTC time, canonical URL, title and acceptance decision. Each trace for steps 1 and 2 records the query, every saved response and hash, the response identified as the latest by modification time, the first result that met the acceptance rule, and whether that result reproduces the recorded profile URL. Capterra has no deterministic profile pattern, because its URLs carry a numeric id, and that is recorded as such. All negative results use the wording “not found in steps 1 to 3”.

Capture and signature. For every profile found, the page was fetched and saved with its SHA-256 and UTC time, and the review count and overall rating were read from the page’s embedded structured data; where a page also shows those figures in its visible text, the two matched in every case we compared. A saved body that is an error response from the fetch layer rather than a review-site page is labelled “profile found in discovery; capture unresolved” with its HTTP status; it enters no eligibility count and is never treated as a profile observation. This rule was written on 2 September 2026 after the live check found two such bodies (see the change log). The study was designed to compare an integer total plus five integer star-bin counts. On 2 September 2026, Capterra, GetApp and Software Advice displayed an integer total, a one-decimal overall rating and sub-ratings; Capterra also displayed sentiment percentages. None of the three review sites displayed star-bin counts. G2 displayed a total and rating but no bin counts in the page text, while Trustpilot displayed percentage bins with exact counts embedded. Under the original rule, every Gartner Digital Markets profile would have been unclassifiable and the study would have had no eligible pair. We therefore amended the signature to use the two fields that determine eligibility in this report: the integer total and the one-decimal rating. We made that amendment after examining the first capture and before computing any statistic for publication. The original rule yields zero eligible pairs, while the tables use the amended rule. A total-plus-rating coincidence between independent review pools is possible in principle, which is why the two checks below were retained.

Eligibility and matching. A profile is eligible when it displays an integer total of at least one and an overall rating. Two or more review sites form a matching group for one product when they display the same integer total and the same one-decimal rating on eligible profiles, and when their displayed listing names are identical or nested after normalisation. Review sites never form matching groups across products. The listing-name condition keeps different listings of a multi-product vendor apart.

Two checks. First, a sensitivity run excludes profiles with fewer than ten displayed reviews because review sites are more likely to display a coincidental match when profiles are small. The run changes eligibility only, never the coverage count. Second, the corroboration check normalises, by case and whitespace only, the labels of sub-ratings that each review site in a matching group displays numerically. It requires at least two labels shared by at least two review sites. Two-decimal values are rounded half up to one decimal, except when the second decimal is exactly 5; this is treated as a rounding tie and makes the group unassessable. Otherwise, the group is corroborated when every shared value matches at one decimal, discordant when any shared value differs, and unassessable when fewer than two shared numeric labels exist. GetApp displays sub-ratings as star glyphs without numbers and so never contributes to this check. Capterra displays Ease of use and Customer Service on its reviews page and Features, Value for money and Customer Service on its product page, so both pages were captured; Software Advice displays two-decimal values on its product page. Corroboration is reported separately from the matching counts and is never used to infer review identity.

Metrics. For each comparable product, T_p is the sum of displayed counts across its eligible profiles. D_p is calculated by summing all displayed counts in each matching group, subtracting that group’s largest count, and then summing the results across groups. R_p is D_p divided by T_p. The headline reports the number of comparable products with at least one matching group and the median R_p across comparable products, counting zero where there is no match. The pooled share, the sum of D_p over the sum of T_p, is labelled as the volume-weighted share of displayed review-count entries classified as redundant under the exact-signature rule, never as a share of unique or real reviews.

Counting and reproducibility. Every count on this page is produced by a script from the frozen dataset: one row per profile with the canonical URL, the displayed count and rating, the listing name, the sub-ratings, the saved file and its hash. The saved search responses, pages, probe attempts and supplementary pages carry SHA-256 hashes that were verified before publication. The dataset, discovery manifests and traces, results files, and exact scripts are published with this page in the Dataset section below. Anyone can therefore rerun the scripts and reproduce the rows. Because of their size, the saved pages are available on request; their hashes are included in the dataset.

The tables

The main study table has 32 rows. Cohort 1 is the headline frame; cohorts 2 and 3 are sensitivity samples with no headline of their own. Each review-site cell shows the displayed review count and one-decimal rating and links to the canonical profile URL. The final column gives the UTC retrieval date. “Not found in steps 1 to 3” means that the three discovery steps produced no accepted page; “no deterministic pattern” marks Capterra, whose profile URLs carry a numeric id that cannot be probed.

The four products in the pricing index appendix, Equip, Adaface, Canditech and Thomas International, were read on the same day but belong to no cohort and enter no denominator. Their rows are listed separately:

What the rows show

Capterra displayed a count and a rating for 12 of the 12 products. GetApp displayed a count and a rating for 12 of the 12 products. Software Advice displayed a count and a rating for 10 of the 12 products, and we found no profile in steps 1 to 3 for 2. G2 displayed a count and a rating for 10 of the 12 products, a profile with no count or rating displayed for 2. Trustpilot displayed a count and a rating for 7 of the 12 products, a page with zero reviews for 3, and we found no profile in steps 1 to 3 for 2.

Capterra and GetApp each displayed a profile in all 12 matching groups. Software Advice displayed a profile in 10 matching groups; G2 and Trustpilot displayed profiles in none. For Discovered and HackerRank, Capterra and GetApp displayed the matching signature, while Software Advice was not found in steps 1 to 3.

For every product for which G2 displayed a count and rating, G2 displayed a total that differed from the matching group’s total. For every product for which Trustpilot displayed a count and rating, Trustpilot also displayed a total that differed from the matching group’s total.

The corroboration check classified 7 of the 12 matching groups as corroborated, 0 as discordant and 5 as unassessable. For Discovered and HackerRank, Capterra and GetApp displayed the matching signature, but GetApp displayed no numeric sub-ratings, leaving fewer than two shared numeric labels. For Criteria, HR Avatar and Vervoe, Software Advice displayed an Ease of use value with a rounding tie. No review site displayed a differing shared value in any group.

For HackerRank, G2 displayed most of the recorded review-count volume, resulting in an R_p of 0.123. For EmployTest, Capterra, GetApp and Software Advice displayed matching profiles whose counts made up most of the displayed total, resulting in an R_p of 0.664. The median R_p across the 12 products was 0.481; excluding profiles with fewer than ten reviews moved the median to 0.482 and left the pooled share at 0.317.

Sensitivity samples

The two supplementary cohorts were selected by different criteria and are reported in this section only; they appear in no headline, in the cite block or in the FAQ. Cohort 2 is the 17 products with their own cost-shaped search demand: 15 of 17 products had profiles found on two or more sites and 13 of 17 were comparable. For 10 of those 13 comparable products, two or more review sites displayed an exact matching signature; the median R_p was 0.083 and the pooled share 0.335. Of the matching groups, 7 were corroborated, 0 discordant and 3 unassessable under the sub-rating check. The review sites displayed different listing names for Mercer Mettl (Mercer Mettl Assessments, Mercer Mettl Coding Assessments, Mercer Mettl 360View and Mettl). The review sites therefore formed no matching group under the listing-name rule, and the product is counted as comparable with an R_p of zero.

Cohort 3 is the 3 products carried over from the DISC comparison: 3 of 3 products had profiles found on two or more sites and 1 of 3 was comparable. For the one comparable product, the median R_p was 0.0. G2 and Trustpilot displayed different signatures, so the two review sites formed no matching group for the one comparable product. For Everything DiSC and DISCinsights, Trustpilot displayed profiles with zero reviews; steps 1 and 2 found no Capterra profile, Capterra had no deterministic step-3 probe, and steps 1 to 3 found no GetApp or Software Advice profile. One acceptance rule was added for this cohort only: a page whose structured-data name equals the roster label exactly also qualifies, because G2 lists Crystal under the bare name “Crystal”, which the roster’s product tokens did not cover. Cohort 1 is unaffected.

Limitations

The study has one observation date. Review sites do not update synchronously, so “same date” does not mean they were read at the same second. The five review sites were chosen in advance. The discovery protocol is limited to pages reached through its three steps. Two Software Advice profiles and two Trustpilot pages were not found. Capterra has no deterministic pattern to probe. Step 3 was added after steps 1 and 2 had been run, for the reason stated in the method. The signature definition was amended after the first capture, which is disclosed above; the original rule yields zero eligible pairs. When two or more review sites display a matching total and rating, that observation is evidence only about what the sites displayed. It establishes neither review identity nor reviewer identity, and it says nothing about whether one site took figures from another or whether a vendor caused or controlled publication. Cross-listing may be an ordinary practice of the sites involved, and this page does not characterise it. Coincidence between independent pools is possible in principle, which is why the corroboration check and the small-profile sensitivity run are reported. Rounding ties left 3 groups unassessable, and GetApp’s glyph-only sub-ratings left 2 more. Counts are displayed entries, not reviews. The headline frame is Capterra’s category on one date filtered by our pure-play classification, not the market, and the sensitivity cohorts describe only themselves. Product names are row identifiers, not allegations or rankings, and no rating on this page is a quality judgement. Corrections: write to [email protected] with the row and the page you read; the observation date stays on the rows and the change log records what changed and when, separately from the page’s dateModified.

Dataset

The frozen dataset behind every number on this page is published in one folder with a SHA-256 manifest (dataset version 2026-09-02, manifest hash 0742dab7f14fc606, 45 files) and a README. The row datasets, one per cohort, are signatures-appendix-2026-09-02.json, signatures-cohort1-2026-09-02.json, signatures-cohort2-2026-09-02.json, signatures-cohort3-2026-09-02.json. The discovery traces, which record every search attempt and its outcome, are discovery-trace-appendix-2026-09-02.json, discovery-trace-cohort1-2026-09-02.json, discovery-trace-cohort2-2026-09-02.json, discovery-trace-cohort3-2026-09-02.json. The script outputs are s3-results-appendix-2026-09-02.md, s3-results-cohort1-2026-09-02.md, s3-results-cohort2-2026-09-02.md, s3-results-cohort3-2026-09-02.md. The probe attempts, captures, supplementary pages and the file inventory are in the same folder, listed in the manifest. The scripts, in the exact version used, are build-draft.py, build-tables.py, capture.py, discover.py, extract2.py, freeze.py, inventory.py, probe2.py, protocol-evidence.py, publish-dataset.py, roster.py, supplement.py, trace.py. The saved review-site pages (about 300 files) are not published because of their size; every row carries the SHA-256 of its saved page, and the pages are available on request from [email protected].

SharpAssessment’s role

SharpAssessment publishes this page and appears in no row. For a brand-level example, see our TestGorilla review analysis. The roster and its exclusions are documented on the assessment pricing transparency index. The same roster’s pages are read for numerical validity claims in the validity citation study.

Change log

• 2 September 2026: first observation, 12 products in the headline cohort. Headline: for 12 of 12 comparable products, two or more review sites displayed a matching signature; median R_p 0.481.
• 2 September 2026: two protocol amendments were recorded before publication: discovery step 3 and the signature definition; both are described in the method.
• 2 September 2026, after publication: the live check found two rows whose saved bodies were fetch-layer error responses (HTTP 502) rather than review-site pages. They were relabelled from “profile, no count or rating displayed” to “profile found in discovery; capture unresolved (502)”. The same reading found that the token rule had accepted listings whose category lies outside the HR family; a category rule was added to the method and four such listings are now shown as rejected. No headline number changed. The dataset was re-issued with a new manifest hash.
• 2 September 2026: a link to the validity citation study was added; no row or count changed.
• 2 September 2026, evening: a per-site reading of Testlify’s review listings, with the labels each site attaches to its reviews and the dates of the reviews it displays, is on our Testlify reviews page; this study’s Testlify row is unchanged.
• Next recheck due by 2 December 2026.