Mining ITreview and BOXIL reviews for competitor weaknesses (summary edition)

What the reviews look like

On one ITreview product page (Sansan, read 2026-10-07, chosen as a large representative product and not a Steve competitor), 500 of 737 reviews scored 4.0 or higher (67.8%), 41 scored below 3.0 (5.6%), and 589 (79.9%) were posted as 非公開, anonymous. Users wrote 87.5% of them; decision makers wrote 3.8%. One product on one day says nothing about the site as a whole. It does show that a sample sorted by rating is almost all praise, and that criticism is likely to sit in the "points to improve" field of mid-scoring reviews.

Both sites run disclosed incentives. A vendor campaign page on ITreview offered up to a 2,000 yen Amazon gift card for reviews posted between 2026-06-03 and 2026-06-14; BOXIL ran a review campaign starting 2019-11-25 and still labels rewarded reviews 投稿経路 キャンペーン (for example, on a review dated 2023-08-19). ITreview shows no per-review label that we could find, so a given ITreview review cannot be called paid or unpaid. ITreview's own 2022-11-08 article reports rejecting about 20% of submissions and verifying every reviewer's affiliation; those are the operator's own figures, and we found no independent audit.

The routine

1. Pick one competitor and one question

   "What do decision makers at mid-sized companies complain about in competitor X" is one question. Filter by company size and role on ITreview, then read the lowest-rated and the newest reviews. Write down how you chose the sample and how many reviews exist in total.

2. Record the bias before the analysis

   Three lines: the rating spread, the share of anonymous reviewers, the number of campaign-labelled reviews, and the date. A table without them cannot be judged later.

3. Paste with IDs, ask for extraction

   Number each review R01, R02 and so on, keep its URL and attributes beside the text, and ask Claude for themes in the "points to improve" fields only, each with two or three verbatim Japanese quotes, the review IDs, and a count of n of N reviews in the sample. Anything that fits no theme goes to "unclassified".

4. Verify every quote

   Search each quoted string in your pasted text. Delete any quote that does not match character for character. Report counts in the sample; never write "most users".

When it stops holding

Reading reviews for several competitors and keeping the table current is the part that lapses. Steve's site (checked 2026-10-07) says Steve researches competitor websites, news, reviews and supported social sources alongside the company information you provide, names G2 and Trustpilot as review sources, shows a status and source links on card rows, and holds updates for your PMM or CI owner to apply. It does not list ITreview, BOXIL or Japanese-language sources, so the Japanese review reading stays with the method above. If research across public sources and keeping claims sourced is the part you cannot sustain, see Steve competitive intelligence.