Last week the Swedish news agency TT reported that Azerbaijan had discovered what it claimed to be mass graves from prior conflicts in the disputed Nagorno-Karabakh region. At the time, TT was unaware that the social media campaign surrounding this news was well planned by the Azerbaijanis. Blankspot has analyzed how the Azerbaijani regime uses bots and troll factories to propagate sympathetic views towards the country.
On October 5, the Swedish TT news agency put out an article that Azerbaijan had found mass graves in Nagorno-Karabakh. The story got republished in about fifteen major Swedish media outlets, among some of the largest in Sweden.
The article’s primary source was Hikmet Hajiyev, personal adviser on foreign affairs to Azerbaijani President Ilham Aliyev, who wrote a tweet. The content of his tweet was shared over a short period on September 19 thousands of times on Twitter, both from Hajiyev’s account and others with similar content.
After criticism from many people, the Swedish-Armenian historian and peace and conflict researcher Vahagn Avedian about the factuality of the text and whether Hikmet Hajiyev is a credible source, concequnelty the appointed head of news for foreign affairs at TT, Pontus Ahlmqvist, regretted the mistake.
On Twitter, he wrote: ”We generally try to include context and caveats when we quote information from authoritarian regimes that we have not been able to verify ourselves. The Armenian information on war crimes could, for example, have been included here. We take the criticism to heart.”
The news came just days after top diplomats from several countries, including the US, France and Norway, condemned Azerbaijan for gross war crimes during the September 13-14 invasion of Armenia. There is no TT article about this.
The question of how a tweet from an authoritarian regime could go through TT’s fact-checking can say something about the Azerbaijani regime’s tactics on social media.
With the help of Twitter user @Bedig_A , who developed software that utilizes Twitter’s publicly available APIs, we provide insight into how the Azerbaijani regime’s propaganda operates on Twitter. The data Blankspot investigated the source of a tweet, the number of interactions, engagements and connections between different accounts. For example, whether one account interacts with another and at what times this happens.
In other words, the software can produce statistical analysis about different users’ behaviour on Twitter. Based on the data provided, we looked into different areas: 1) signs are showing how professional ”troll factories control the accounts”. 2) Signs that there are bots and 3) Indication of state-coordinated information campaigns.
This has been done by partly focusing on particular #hashtags with many apparent tweets in a short time intervals, accounts that tweet a lot without getting much interaction, and tweets from regime representatives that get rapid non-organic spread.
The analysis has been based on periods when there was war or conflict and just before and after. For example, when Azerbaijan attacked Armenia on September 13-14.
Chart shows top accounts that spread #NoPelosi (see text below)
Signs of troll factories
One purpose of troll factories is to make particular #hashtags and messages trend. One such example is #ArmenianVandalism, which was the main message on October 5 when TT picked up the news about mass graves.
Our analysis shows that the tag #ArmenianVandalism, which normally occurs about ten times a day, started to increase on September 27. This was the same day the EU condemned Azerbaijan for the war crimes. Activity then increased to approximately 300 occurrences per day and remained there until October 5, when it suddenly spiked to almost 2000 posts, only to sink to 300 again the following day.
The top accounts spreading this hashtag shared some characteristics: they had generic Azeri names, followed by numbers. They either had no profile picture or a profile picture of a young male person; some had the ruling party logo. Almost all had relatively few followers, but enough to not seem like a startup. Most were started between one and six months ago.
Between the hours of 16:50-17:37, #ArmenianVandalism increased through 261 individual accounts, with several of the accounts we looked at tweeting the message upwards of 20 times within the time span.
Chart shows how #ArmenianVandalism hashtag gained momentum on October 5th
Soon after, it was picked up by the news agency TT.
Another example is the user @LiyevAqsin, who has occasionally been banned from Twitter. Since its creation during the Nagorno-Karabakh war in October 2020, the account has tweeted no less than 58,200 times. As a rule, the tweets are brief and exclusively praise the Azerbaijani regime.
Almost all tweets are posted between 09:00 in the morning and 18:00 local time, i.e. a normal working day by Baku standards.
On average, the account is spent on Twitter about four hours per day spread over 7 days a week, but the figure is higher if you start from working days when it is actually active.
The third example is the campaign against US Democratic Senator Bob Menendez, who, on September 28, presented a proposal to ban US money supporting the Azerbaijani defence.
During #CorruptPolitician, 7,246 likes and 22,504 tweets were generated from 1,514 accounts, of which 1,452 were unique. The average time between each tweet was 2 seconds. Several regime representatives attended.
A fourth example is the hashtag #NoPelosi, which was established on September 19 when US Congress Speaker Nancy Pelosi was on a state visit to Armenia. In connection with that, she made unusually strong statements against Azerbaijan as a result of the invasion a couple of days earlier.
The analysis shows that on September 18, #NoPelosi began to be used. Then it was tweeted 205 times with seven minutes between each tweet – on average. On September 19, it increased to 2,370 tweets, with 27 seconds between each tweet. One account, @Darya111991 tweeted #NoPelosi 519 times, which equates to an average of three tweets per minute on September 19 in the 24 hours.
Examples of the above accounts, back in June 2021, that appear to have abnormal patterns on Twitter. Some accounts have occasionally been suspended. The the green bar is sometimes shorter than the other two.
Signs of bots
The case with @Darya111991 shows signs of the use of bots.
Azerbaijani bots typically have generic usernames which often include many numbers. They tweet a lot and for a long time and their tweets are typically not elaborate in content – in other words, they do not appear unique.
One example, the user @NarnialsFree tweeted 24 hours a day for 170 days. For 510 days this account tweeted mainly during working hours. A total of 24,498 tweets.
There are many more examples. However, the intensity of tweeting increases when special messages are to be communicated in times of crisis and conflict.
The above map shows how these accounts have shared the same tweet about Cawa Media. It all started from Zaur Ahmadov's account. This indicates that the tweet did not go viral organically but in a single trajectory through content Retweeting. A tweet is usually spread in several stages through shares.
Signs of coordinated information campaigns
In addition to the use of bots and troll factories, ‘real people’ with accounts are also needed to make a substantial impact. In the case of the TT article, it was Hikmet Hajiyev who was the source, but his is just one of many Azerbaijani accounts.
The information campaigns that have had the greatest impact are where bots, troll factories and accounts run by real people all cooperate, as was the case in #ArmenianVandalism, #CorruptPolitician and #NoPelosi.
On the other hand, there is coordinated cooperation without the other components as well. The dividing line between organic and artificial propagation in such cases is possibly not great.
One example of this is from December 2021 when the Azerbaijani Ambassador to Sweden, Zaur Ahmadov, (a known critic of Blankspot) tweeted a video by Cawa Media which – according to Azerbaijani rhetoric – showed Armenian war crimes in Nagorno-Karabakh. Only one day earlier, Blankspot had published an article about Cawa Media participating in an invitational trip to Azerbaijan which was followed up with an article about a seminar at an Azerbaijani propaganda conference in Brussels.
Zaur Ahmadov’s tweet received 130 shares and 227 likes. According to our analysis, 83 of these occurred within 58 seconds. Unlike the previous examples in the article, the vast majority of participating accounts here appear to be genuine, i.e. run by real people.
Example of an account supposedly run by a troll factory. It joined Twitter on 10-10-2020 and till today it tweeted 58,513 tweets.
Among those who shared the video in those 58 seconds are Fuad Muradov (Azerbaijan’s Minister of Diaspora Affairs), a number of ambassadors, the PR manager of the state oil company SOCAR and several government employees in Baku.
What is striking is that all shares originated from Zaur Ahmadov’s account. Thus there were no retweets from those who had shared it. Normally, tweets are shared by one person sharing something which leads to a new follower sharing it further and so on.
It is therefore likely that the sharing of the tweet was planned through internal channels.
Since then, Cawa Media has frequently published articles about Azerbaijan, and despite there being a war between Armenia and Azerbaijan, the organization has also managed to be event photographers for the Armenian embassy in Stockholm.
The fact indicates that through well-planned coordination on social media, Azerbaijan can get away with its tactics without any criticism from the Swedish news outlets due to the lack of attention and interest the Caucasus region gets.
Ambassador Zaur Ahmadov used the article from TT as an indication that Swedish media are covering what is happening in Azerbaijan at all.
One reason for TT’s lack of source criticism by other newsrooms is that the conflict in Nagorno-Karabakh is considered a peripheral part of Europe. It often falls outside the sphere of interest of the Western media.
Correction (2022-10-10, 21:17): One Twitter account was wrongly pointed out as an example of likely being a bot. The example is deleted.
(c) 2022, Blankspot