What happened?
DeNA operated ten curation sites, including WELQ, MERY and iemo. In November 2016, outside criticism focused on WELQ, which published health and medical information, including articles without expert supervision and content whose basis was unclear. DeNA made all WELQ articles unavailable. Questions then spread to article-production practices and the use of text and images on other sites, and by December 7 all articles across the ten sites had been taken offline.
DeNA established a third-party committee to investigate the facts, legal and ethical issues, causes and measures to prevent recurrence. Its report, submitted on March 11, 2017, records not only problematic content but also how the systems for producing large volumes of articles at MERY and WELQ were designed and operated.
Eleven passages that stood out
This article focuses less on the best-known medical-information controversy and more on the operational details buried in the report: MERY’s image checks, WELQ’s outsourcing model, what was sometimes called the “WELQ method,” writer manuals, and actual work-chat instructions.
The primary report is in Japanese. The English text here is From Reports’ translation and summary, not an official English translation. Page numbers refer to the printed pages of the investigation report inside the 306-page disclosure PDF.
MERY began with female university students writing in what the report describes as an unpaid, volunteer-like role
According to the committee, female university students were central to writing articles from the start of MERY.
According to the report, around 10 to 20 students who knew the founders initially wrote without pay in a role resembling volunteer work.
About six months after launch, in autumn 2013, the company began paying them and changed the name of the role to “intern.”
The group later expanded, and about 140 interns were enrolled shortly before the controversy.
Restricting available images would narrow the range of articles. That was one reason MERY did not use a whitelist earlier
According to the report, after DeNA acquired MERY’s operator, rules were introduced against using images that could infringe copyright or portrait rights, and a blacklist approach was adopted.
For articles written by interns and part-time writers, however, a whitelist limiting images to approved sources was not introduced until after a complaint in late August 2016.
The report gives two reasons for not adopting a whitelist earlier: limiting images was thought to narrow the range of articles, and there was trust that interns and part-time writers would not use problematic images.
MERY checked the first few images. If those looked fine, the rest were not checked
According to the committee, for MERY articles written by interns and part-time writers, reviewers checked the source websites for images to determine whether the images could be used.
MERY articles often contained many images.
In practice, reviewers checked the first few images in an article and, if they found no problem, did not check the remaining images.
By July 2016 at the latest, an engineer also performed a weekly system-assisted review of source sites and removed entire articles when images from blacklisted sites were found.
Inside the division, the production system was sometimes called the “WELQ method”
According to the report, WELQ sought to reduce the involvement of internal editors as far as possible and extensively documented how article outlines, writing and content review should be performed. Management of outside writers was largely delegated to outside directors, with internal editors stepping in mainly for exceptional cases.
The report says this approach was sometimes referred to within the curation planning division as the “WELQ method.”
The committee also made clear that outsourcing and manualization were not unique to WELQ. It said WELQ built and operated the model in the most drastic and systematic way.
They recognized medical supervision was needed for accuracy, yet came to accept unsupervised articles as a consequence of cost and mass production
WELQ initially focused on subjects such as dieting and exercise that the team did not consider to require medical supervision.
As article-volume and SEO traffic targets rose, the site expanded into medical topics that had initially been held back because supervision was considered appropriate.
The committee found that team members recognized medical supervision was necessary to ensure accuracy, but came to regard the production of unsupervised articles whose accuracy could be questionable as unavoidable in light of cost and mass-production demands.
Serious work on adding medical supervision began around September 2016, but the controversy broke before the planned process was introduced.
Article outlines had a “rival-site URL” field, and external directors were told to check that it was filled in
WELQ also outsourced the creation of article outlines. By around January 2016, outside outline writers created them and outside outline directors reviewed them.
The outlines supplied to writers contained a “rival-site URL” field, and the external-director checklist included confirming that the field had been completed.
WELQ editors said the links were intended as references for the breadth of information to include, not as encouragement to copy or reuse wording from those sites.
After monthly publication exceeded 1,500 articles, review was outsourced. Eventually even internal spot checks stopped
According to the committee, by around December 2015, WELQ was publishing more than 1,500 articles a month, making it extremely difficult for internal editors to review every article adequately.
The team hired just under 20 external editing directors to review all articles written by outside writers. From around January 2016, those directors could publish articles without prior review by WELQ’s internal editors.
Internal editors initially spot-checked a small random sample after publication, but by around September 2016 at the latest, even those checks had stopped entirely.
The manual said combining information from several sites could produce a more “original” article
WELQ’s writing manual included concrete guidance for avoiding text that looked too similar to reference sites.
One method was to consult several sites for a single subsection and combine the information or views drawn from them.
The manual explained that doing so would make it less immediately apparent which site had been used as a reference and could produce an article with greater originality.
It also instructed writers to use facts as references while expressing them in their own words and order.
The manual split text word by word into “facts” and “expression,” then demonstrated how to rewrite the expression
The same manual showed writers how to analyze reference text by separating necessary factual terms from wording that could be changed.
Technical terms needed to preserve meaning were treated as facts, while phrases that could be restated were treated as expression.
The report reproduces worked examples in which words are classified and the wording and sentence order are changed while the required informational terms are retained.
The method was also shown for medical and specialist articles.
An actual work chat told a writer to look at sites on Google’s second page and rewrite the wording
The committee reviewed not just manuals but actual chat exchanges between WELQ editors and outside workers.
In one exchange, an editor decided that wording was too similar to a reference site and asked for further rewriting.
The instruction suggested using sites appearing on the second page of Google search results as additional references when producing different wording.
The people who created and revised the manual said their aim was to explain how to use references, not to teach unauthorized use of other sites’ writing or to encourage copying. Footnote 202 reports that, in the committee’s call for opinions, eight of the 18 outside respondents who said they had written for WELQ felt that the manual encouraged copying, while four did not. The remaining six did not answer that question.
A search for “shoulder pain” suggested “spirit,” so the outline added a section asking whether a ghost could be the cause
According to the report, WELQ used search terms and Google’s suggested terms when selecting article topics and building outlines.
When one outline writer searched for “shoulder pain,” one of Google’s suggested terms referred to spirits.
The writer therefore added a subsection asking whether a ghost could be a cause, and an outside writer produced the article from that outline.
The committee noted that the passage about spirits was only a small part of the article, but said SEO considerations had been overemphasized at the expense of the reasonableness of the content, while the outside editing director did not examine that reasonableness deeply because the article followed the approved outline.
What the report says about the background
The third-party committee concluded that DeNA’s curation business placed strong weight on numerical targets such as article volume and SEO traffic, while management and checking systems failed to keep pace with rapid expansion.
At WELQ, reducing internal-editor labor and outsourcing and manualizing each stage of article production helped the site increase article volume and traffic in a short period. At the same time, the system did not adequately control risks involving expertise, accuracy and the use of text and images.
The committee did not treat outsourcing or manuals as inherently problematic. Its broader concern was that the business expanded rapidly without a sufficiently effective process for identifying risks and for independently checking and correcting the direction of the operation.
Corrections to this article
The numbered body ends at page 276, not 277; the PDF offset is +30 rather than +29. Printed references for all 11 notes were checked again. Public disclosure was March 13, rather than the March 11 receipt date. Committee accounts of operating practices are distinguished from manual and chat records. Note 10 adds the responses of outside respondents who said they had written for WELQ in footnote 202. Note 11 adds the Committee assessment on pp.246–247 and its qualification that the passage about spirits was only a small part of the article.