DMG Mediasupplementary written evidence (FEO0110)

 

House of Lords Communications and Digital Committee inquiry into Freedom of Expression Online

 

  1. This evidence is submitted on behalf of DMG Media, publishers of the MailOnline, metro.co.uk and inews websites, the Daily Mail, Mail on Sunday, Metro and i newspapers, and the New Scientist magazine.

 

  1. The Committee has asked for written evidence on how DMG Media moderates user-generated comments on its websites. User-generated comments are currently carried on MailOnline and metro.co.uk (though not on all stories). They are not currently carried on inews, though that may be reviewed, or the New Scientist website.

 

  1. We would also like to take this opportunity to submit fresh information about discrimination in algorithms, which is raising growing concerns about some platforms’ readiness to restrict freedom of expression.

 

 

Moderation of user-generated content

 

  1. All our user-generated content is regulated by the Independent Press Standards Organisation in the same way as news content, unless it has not been reviewed or moderated before publication. In that case it becomes subject to regulation if a complaint is made about it and it continues to be published without the complaint being dealt with. In practise this system seems to work well, and IPSO deals with very few complaints about user-generated content from our websites. The way the system operates is set out in Table 1 below:

 

Table 1: DMG Media content moderation

 

 

MailOnline

Metro.co.uk

 

High risk stories – e.g. criminal charges, ongoing court cases

Comments not permitted

Comments not permitted

Lower risk stories – e.g. stories that discuss issues connected to protected characteristics

Comments moderated and subject to IPSO regulation as soon as posted.

Users can flag stories they find offensive – five flags and the comment is automatically removed and only reinstated if moderators are happy it is not in breach of our standards

Currently, comments are generally not permitted

Very low risk stories

Not moderated. However one flag triggers automatic removal and review. Comment then becomes subject to IPSO regulation upon being reinstated

Not moderated. However one flag triggers a review by moderators and three flags triggers automatic removal and review. In either case comment then becomes subject to IPSO regulation

Registration and banning

All commenters have to register their username and email address and are banned if they repeatedly breach standards

Commenters may register their username and email address but can also comment as guests.  They are banned by username and IP address if they repeatedly breach standards

Moderating team

UK – 8 full-time and 12 freelance. 5/6 working during the day and 2/3 evening until 2am.

USA – 3 working weekdays, 1 weekday evening and 1 weekend days

US covers UK during the night.

One designated member of staff to review flagged and filtered comments. Comments are enabled by the bylined reporter but in some cases, decisions may be reviewed by a moderator, if they encounter problematic comments.

 

  1. House rules include a prohibition on comments which are defamatory, false or misleading; insulting, threatening or abusive; obscene or of a sexual nature; offensive, racist, sexist, homophobic or discriminatory against any religions or other groups. They also do not allow users to post private information such as addresses or telephone numbers.

 

Discrimination in algorithms

 

  1. MailOnline is the most visited news website in the UK (excluding broadcasters)[1] and the fifth most visited in the world[2]. It would be logical therefore to expect, when a member of the public uses Google to search for a news subject such as ‘Covid’, ‘Boris Johnson’ or ‘Harry and Meghan’, that MailOnline stories would generally appear high up in the first page of search results. That is not the case. Data from search analytics companies Sistrix and NewsDashboard UK shows that Google overwhelmingly favours the Guardian and BBC in search results and discriminates heavily against other news websites, particularly MailOnline. Indeed MailOnline’s search visibility[3] for many important news search terms is close to zero – and for the terms ‘Covid’ and ‘Coronavirus’ actually was zero for the month of March this year.

 

  1. DMG Media’s submissions to the Communication Committee’s inquiry into the Future of Journalism, and the Competition and Market Authority’s Market Study on Digital Advertising, charted how Google’s algorithms have consistently reduced MailOnline’s search visibility since 2013. Particularly striking evidence that MailOnline was being targeted by Google came in 2019, when an algorithm change in early June cut MailOnline’s search visibility by 50pc, while other news websites’ visibility improved. Three months later, following protests to Google at the highest level, MailOnline’s search visibility was equally suddenly restored. At neither point did we make any changes to the structure or presentation of the site which would explain its rejection or subsequent re-acceptance by Google’s algorithms.

 

  1. At the time we were only able to guess why Google had reduced MailOnline’s search visibility. However when reviewing evidence for the CMA, it became apparent to us that the June algorithm change coincided with the introduction by Google of its Unified Pricing rules in digital ad markets. These rules had the effect of limiting the use by publishers of header bidding, a means of setting price floors which enabled us to fill more of our ad inventory with better-paying non-Google demand. We have since learned that other major publishers which made use of header bidding, such as the News York Times and Conde Nast magazines, also saw search visibility drop in June 2019, though not as dramatically as MailOnline. The consequence of Unified Pricing was that by the end of the three-month period June-September 2019, Google had forced MailOnline to sell twice as much ad inventory through Google’s ad exchange, while Google paid half as much for each ad slot.

 

  1. For most of 2020 MailOnline overall UK search visibility index hovered at around 100 – only a quarter of the best figures recorded in 2012-15, but similar to the level before the dramatic drop in June 2019. However, from January this year we started to see another steady decline in MailOnline’s overall search visibility index, which now stands at just over 50, around half the level seen through most of 2020, and one-eighth of the 2012-2015 peak. (See Tables 2 and 3).

 


Table 2. MailOnline UK overall search visibility index – last 10 years (source: Sistrix)

 

 

 

Table 3. MailOnline UK overall search visibility index – last 12 months (source: Sistrix)

 

 

  1. When MailOnline’s search visibility is plotted against its main UK rivals a disturbing pattern emerges. Pre-2015 Google heavily favoured the BBC, with the Guardian, MailOnline and Telegraph broadly grouped together[4]. From 2015 onwards two distinct groups start emerging. The Guardian and BBC are consistently favoured with a visibility index currently standing at around 400, and the Mail, Telegraph and Sun are grouped together with consistently poor visibility, currently standing at around 50-75. (See Table 4 - this chart does not include the Express and Mirror, both of which score slightly below MailOnline).

 

Table 4. UK competitive search visibility last 10 years (source: Sistrix)

 

C:\Users\Peter.Wright\AppData\Local\Microsoft\Windows\INetCache\Content.Outlook\J0U3Q02P\2021-04-06_dailymail.co.uk_Visibility_Index (1) (002).png

 

  1. The implications for freedom of expression and democracy are obvious. How Google’s search algorithms work is the company’s most closely-guarded secret. Google tells the public: ‘To give you the most useful information, Search algorithms look at many factors, including the words of your query, relevance and usability of pages, expertise of sources and your location and settings. The public place great faith in Google, and imagine that when they search for news on politics, health, business, or any number of other topics, Google’s emphasis on relevance and expertise means the content they are shown has been picked because it gives the most reliable and useful information. Unless they are students of search visibility they have no idea that when they search for news Google invariably takes them to two left-leaning news sources, the Guardian and BBC.

 

  1. As this submission was being drafted, news broke of the Duke of Edinburgh’s death. An hour after the news broke, the first page of Google search results for the term ‘Prince Philip’ yielded five links to the BBC, three to the Guardian, two to the Telegraph, and one each to the Mail, Sky News, WalesOnline and the American magazine website People.com. Possibly Google maintains an old-fashioned view of the BBC as the state broadcaster, but the relevance and expertise of the Guardian, which is scornful of monarchy and only covers royal matters when obliged to, is hard to fathom. In fact NewsDashboard UK[5] search analytics for the term ‘Prince Philip’ across all UK mobile devices for the 48 hours following the announcement of his death shows Google preferred the Guardian even to the BBC – a result that would probably surprise the Guardian’s own journalists.

 

Table 5. ‘Prince Philip’ - UK: Mobile Overall Visibility Share, April 9 to 11 (source: NewsDashboard UK)

 

  1. This pattern is repeated across many search terms, as the following charts demonstrate. The most striking is Table 6, which shows UK mobile visibility for the search terms ‘Covid/Coronavirus’ during March this year, in which MailOnline scores zero, while the Guardian and BBC each score more than 11pc, ahead of the World Health Organisation and NHS websites.

Table 6.Covid/Coronavirus’ - UK: Mobile Overall Visibility, 30 days to March 29 (source: NewsDashboard UK)

 

  1. Tables 7, 8 and 9 show search visibility for the terms ‘Boris Johnson’, ‘Joe Biden’ and ‘Lockdown’, for all of which MailOnline scores less than 1pc. Google favours the Guardian for political coverage, with the left-leaning title scoring 24.5pc visibility for its Boris Johnson coverage and 22pc for its Joe Biden coverage. The most extraordinary result is for Lockdown, where Google gave 8pc visibility to BBC.com, the BBC’s American website (on top of 24pc to BBC.co.uk), and 3.5pc to France24, the news site of the French state broadcaster. For this search term the Guardian was eclipsed by the similarly left-leaning Independent, but why Google should rank an American and a French website respectively 70 and 30 times more relevant for a British audience than Britain’s largest news website (MailOnline search visibility: 0.12pc) is impossible to guess.

 

Table 7. ‘Boris Johnson - UK: Mobile Overall Visibility, 30 days to March 29 (Source: NewsDashboard UK)

 

Table 8. ‘Biden/Joe Biden’ - UK: Mobile Overall Visibility, 30 days to March 29 (Source: NewsDashboard UK)

 

Table 9. ‘Lockdown’ - UK: Mobile Overall Visibility, 30 days to March 29 (Source: NewsDashboard UK)

 

  1. Google might argue – though without any justification – that on ‘serious’ issues such as these a ‘tabloid’ news website should not expect as much traffic as a ‘serious’ website. But even on ‘tabloid’ subjects such as Meghan and Harry’s Oprah Winfrey interview, Piers Morgan’s subsequent walk-out from Good Morning Britain, and the Premier League, Google still favours the Guardian and BBC as Tables 10, 11 and 12 demonstrate, although the Independent also scored well for the Piers Morgan story.

 

Table 10. ‘Meghan and Harry’ UK Mobile: Overall Visibility Share - 1st Mar to 11th Mar (Source: NewsDashboard UK)

Table 11. ‘Piers Morgan’ UK Mobile: Overall Visibility Share - 1st Mar to 11th Mar (Source: NewsDashboard UK)

 

Table 12. ‘Premier League’ UK Mobile: Overall Visibility Share – 30 days to March 29 (Source: NewsDashboard UK)

  1. In the case of the Oprah Winfrey interview Google might also argue that it restricted access to MailOnline reporting because of Meghan’s previous legal action against The Mail on Sunday and MailOnline (which is still subject to appeal). If so that was an undeclared editorial decision made by Google – precisely the sort of censorship we fear could be validated by online harms codes of conduct. Certainly Facebook did not feel under any obligation to restrict access to MailOnline. Social media does not use search terms, so is not open to the same type of analysis, but as Table 13 shows, the day of the Oprah Winfrey interview (March 7) saw MailOnline achieve its highest-ever volume of traffic on Facebook.

 

Table 13. Facebook article views – 12 months to March 30. (source: Adobe)

 

 

  1. Google never explains how its algorithms work, so we have no idea why it discriminates so consistently against some publishers and in favour of others. But there can be no denying it does discriminate, and in doing so it not only damages businesses, but also threatens democracy by distorting the public’s perception of news.

 

  1. Our concerns about the commercial effects of discrimination in search led us to argue successfully that the Competition and Markets Authority should include search in its Digital Advertising Market Study. We maintained that it was impossible to plan our business without fair warning and explanation of algorithm changes, and without remedies when those changes cause commercial damage. The Market Study’s Final Report found:

 

It is clear that many publishers rely on Google and Facebook for a significant proportion of their traffic and that changes to key search algorithms by either of these can have a significant impact on publisher businesses. We would, therefore, consider it reasonable that publishers have sufficient explanation of how these algorithms work and sufficient notification of changes to them where they might notably impact upon their businesses. We consider that provision to publishers of sufficient explanation about how the key search algorithms work as well as explanation and notification of changes to these are areas that would appropriately be covered by the proposed code of conduct.’[6]

 

  1. The Digital Markets Unit has now started work, and is drawing up codes of conduct in advance of the forthcoming Digital Competition Bill, which will give it statutory powers. We very much hope its codes of conduct will address algorithms in a similar way to the original draft of the Australian News Media and Digital Platforms Mandatory Bargaining Code, published in July 2020[7], though with the addition of a means to secure remedy.

 

  1. However our current great concern is that, just as the Digital Markets Unit introduces measures which will prevent discrimination in algorithms under competition law, online harms legislation will give platforms a new justification for continuing to discriminate. The platforms have a long history of exploiting regulation to reinforce and extend their monopoly positions in markets they dominate – as Google has done with GDPR.

 

  1. That is why we believe news publishers must have a cast iron, crystal clear exemption from online harms legislation, not only for their own websites, but also when their content is distributed by platforms such as Google and Facebook. If platforms interfere with democracy by promoting some news publishers and discriminating against others they must face penalties just as heavy as for distributing content judged harmful under online harms codes of practice.

 

 

April 2021

12

 


[1]              https://pressgazette.co.uk/huffpost-cuts-brutal-uk/

[2]              https://pressgazette.co.uk/top-50-largest-news-websites-in-the-world-sputnik-drudge-and-fox-see-biggest-traffic-falls-in-february/

[3]              Sistrix publishes indices measuring search ranking across sets of representative keywords. https://www.sistrix.com/support/sistrix-visibility-index-explanation-background-and-calculation/

[4]              The Guardian had no visibility pre-2013 because at that point it changed to its current domain name, theguardian.com. The Sun had very low visibility 2013-15 because it was behind a paywall.

[5]              NewsDashboard UK analyses the percentage share of available ranking given by Google to each publisher for a wide range of search terms, taking into account the position and size of search results.

[6]              CMA Digital Advertising Market Study, Appendix S, p.10

[7]              https://www.accc.gov.au/focus-areas/digital-platforms/news-media-bargaining-code/draft-legislation section 52N