Published
The second part of Week 6 looks at 1,705 YouTube comments and replies under BNT's presidential debate, in a guest analysis by the Perceptica team: negative assessments prevail for both candidates, but Yotova draws more reasoned criticism and Gurov more mockery. A new cycle of 81 simulated conversations shows how ChatGPT, Gemini and Google AI Overview guide voters ahead of the runoff.
Guest analysis by the team of Perceptica.
The analysis covers the comments under the video of the debate broadcast live by BNT on 29 October 2026 from 21:00. The comments were collected via the YouTube Data API 24 hours after the video was published.
The total number of posts is 1,705, of which 836 are comments and 869 are replies.
Each post was coded separately for each of the two candidates, using predefined definitions:
The assessments of the candidates were coded deductively, based on the categories listed above. Topics were coded based on the issues discussed during the debate. Specific labels were recorded during coding and subsequently merged into broader categories (war, economy, pardons).
The data cover the first 24 hours after the video was published and do not include deleted posts or reactions during the broadcast itself. The analysis counts profiles, not real people, so one user may be present with more than one profile. Finally, YouTube commenters are not a representative sample of voters, and the results should not be interpreted as such.
Activity is highest immediately after the debate. Almost 40% of all posts were made during the first night, and around three quarters within the first 13 hours. Interest gradually fades during the afternoon and evening of 30 October. The tone of the discussion remains largely stable throughout the period. The only more noticeable change is the higher share of positive comments for both candidates during the first night. A second, weaker wave of activity is observed on the morning of 30 October between 07:00 and 11:00.
The discussion is highly concentrated. Only 23 profiles (around 3%) of the 786 that took part in the discussion wrote nearly a quarter of all posts. Their concentration is strongest in the replies, where they authored more than a third. Their share of comments is 11%. The main findings on the assessments of the candidates reflect a broad range of users.
No single political affiliation is found among the most active users. They include both consistent supporters and critics of the candidates, as well as profiles that argue with other commenters without assessing Yotova and Gurov. Support for Yotova comes from a broad range of users: the most active profiles authored only around 17% of the positive posts about her. Support for Gurov is concentrated: the same profiles authored 38% of the positive posts about him. Some of his most active supporters are also among Yotova's most frequent critics.
The most active profiles follow two different patterns. Some participate steadily in the discussion over 10 to 22 hours. Others publish series of posts within less than an hour, sometimes at a pace of one post every one or two minutes. These series are concentrated mainly in the first hours after the debate and occur among profiles with different positions towards the candidates.
42% of all posts (718 of 1,705) do not assess either of the two candidates. They focus mainly on the political situation in Bulgaria, voters and voter turnout. Separately, 109 posts (6.4%) contain an assessment of the moderator Poli Zlatareva, BNT or the debate format. Among the candidates, assessments of Andrey Gurov account for 38.4% of all posts, 3.8 percentage points more than Iliana Yotova (34.5%).
Negative assessments (criticism, mockery) prevail among the posts that assess the candidates. 31.6% of posts (539) assess Andrey Gurov negatively, while for Iliana Yotova the figure is 25.5% (435). Yotova receives a larger share of positive assessments - 8.3% of posts (142), compared with 6% (102) for Andrey Gurov.
The distinction between criticism and mockery is important context, as the negative assessments of the two candidates differ substantially. The analysis finds a higher share of reasoned disagreement towards Iliana Yotova. 65.3% of the negative assessments of her are criticism, most often accusations of insincerity and manipulation, as well as of closeness to Prime Minister Rumen Radev, which creates a risk of concentration of power. The mockery of Yotova again draws mainly on her political past and her links with the BSP ("the socialist madam", "the communist auntie", "comrade"). Another part of the mocking posts concerns her appearance, and a small share use derogatory nicknames ("mafYotova").
For Andrey Gurov, 55.1% of the negative assessments are mockery, while the reasoned criticism focuses on his competence and readiness for the post, as well as on his position on the war in Ukraine. The mocking posts about Gurov target his manner of speaking, hesitations, pauses and "nervousness" during the debate. Diminutive and derogatory versions of his name are common ("Gyuro", "Gyureto", "Gyurcho", "Gyurkata"), as are comparisons with former President Rosen Plevneliev and mockery related to the war, such as calls for him to "go to the front" himself.
The analysis of likes is limited to comments (n = 836), as replies are less visible and systematically receive fewer likes. The average number of likes per comment is 10.7.
For Iliana Yotova, the most liked comments are the positive ones. They receive an average of 16.3 likes, around 50% above the average. Criticism and mockery of her receive below-average likes (7.7 and 6.3 respectively), with mockery being the least liked category. For Andrey Gurov, the most liked comments are those mocking him (14.5 on average), followed by criticism (11.9). Positive comments about him receive below-average likes (9.4).
Comments mocking Gurov account for 23.6% of all comments but collect 2,855 likes, or 31.9% of all likes on comments. In total, negative comments about Gurov collect 54.1% of likes, while negative comments about Yotova collect 21.2%. Since a single comment may contain assessments of both candidates, some likes are counted in both shares.
Comments receive one reply on average. Positive comments about Gurov attract the most replies - around two per comment on average - but receive few likes. Support for him often sparks arguments among users, and among the replies that assess him, negative ones prevail, often of a mocking nature.
Positive comments about Yotova also spark arguments, although to a lesser extent. Unlike those about Gurov, they are highly liked, but criticism from users prevails among the replies under them. Critical comments about her are rarely challenged by her supporters. Mocking comments attract the fewest replies and likes, and rarely provoke replies of the same kind.
A relatively small share of posts address a specific political topic. Most of the discussion is focused on the candidates themselves, their performance in the debate and their personal qualities. Among posts with a specific topic, the most discussed are the military actions in Ukraine, Iran and the Black Sea, the economy, and Iliana Yotova's pardons.
The military actions in Iran and Ukraine are associated with a less favourable assessment of Gurov. Comments often express concern that his position carries the risk of Bulgaria being drawn into a military conflict. Like the pardons, this topic generates highly emotional messages among users. The most frequently mentioned issue is the agreement with Ukraine, which commenters associate with unacceptable financial commitments. Calls for him to "go to the front" are also common, combined with reminders that he did not do military service yet is seeking the post of Commander-in-Chief.
Critical comments about Yotova most often concern the government's actions in the Black Sea. Among them, Yotova's competence as Commander-in-Chief is questioned due to the lack of specific measures. The criticism often turns into irony, and some users say that Yotova does not name Russia so as not to offend the "bratushki" (a colloquial Bulgarian term for Russians).
The economy is also less favourable for Gurov, who receives almost no approval on this topic. The criticism is linked mainly to the introduction of the euro and to the actions of previous governments with which commenters associate him. Even his experience as an economist, which some supporters point to as an advantage, is rarely used as an argument in his favour. Criticism of Yotova relates mainly to high prices and rising fuel costs under the current government, as well as to accusations that she presents the economic situation in a more favourable light than reality.
Many of the posts related to the pardons contain a negative assessment of Yotova and do not generate significant discussion about Gurov. Commenters question the medical grounds for the pardon and Yotova's explanations in the debate. Her claim that she cannot imagine a Bulgarian doctor signing a false document provokes a particularly strong reaction. The topic is also a source of mockery, including derogatory nicknames that link Yotova to organised crime ("mafYotova").
The moderator and the format of the debate are the subject of 109 posts, more than nine in ten of which are negative. The most common criticism concerns the way the debate was moderated: the constant interruptions and reminders of the time remaining, which according to commenters prevent a real debate and dialogue between the candidates. At the same time, some commenters describe Poli Zlatareva as biased, with accusations pointing in opposite directions: in favour of Gurov and in favour of Yotova. Some users compare the debate to the TV game show "Minuta e mnogo" ("A Minute Is a Lot"). Separately, some posts criticise BNT for the debate format, the choice of moderator and the selection of "convenient" questions.
In the week between the two rounds of the presidential election, we are running a new cycle of experiments with large language models (LLMs). This time our focus is entirely on the candidates in the 1 November run-off. We aim to simulate the behaviour of a potential voter who consults an AI chatbot to decide whether to vote and whom to support. To do so, we again use a set of 9 questions, each asked three times to three popular language models (ChatGPT, Gemini and Google AI Overview), for a total of 81 simulated conversations.
The first series includes two open questions about the run-off, as well as two requests to compare the presidential candidates:
The AI models consistently refuse to advise the user whom to vote for, but readily advise them on how to make their choice. Across the eight generated texts (Google AI Overview failed to produce an answer in one of the attempts), the following approaches emerge:
In addition, all of ChatGPT's answers and one of AI Overview's invite the user to share their personal views and priorities so that the chatbot can help them choose a candidate.
When comparing the strengths and weaknesses of the two candidates, all three models predictably take a balanced approach, trying to present Yotova and Gurov in an equal light and to avoid arguments that would tip the scales in favour of either. Gemini and AI Overview mention almost no specific facts, limiting themselves to general statements about qualities and shortcomings. As Yotova's strengths, they most often cite her extensive institutional experience (without mentioning that she has been President since January 2026), her links with Bulgarian communities abroad and her emphasis on national sovereignty, stability and continuity. For Gurov, they mainly mention his expertise in economics, finance and banking and his profile as a modern reformer.
As a drawback for Yotova, Gemini and AI Overview note that she is "associated with the traditional political model and the status quo", and for Gurov - that he lacks experience in foreign policy and defence. Both models avoid discussing specific controversies surrounding the candidates, the only exception being a passing mention of "legal and administrative disputes over his former post at the BNB" for Gurov in one of AI Overview's answers.
Against this background, ChatGPT's answers are far more detailed and well-argued, and also contain the most accurate and up-to-date information. In addition to the arguments listed by the other two models, ChatGPT also highlights that Iliana Yotova is backed by the ruling Progressive Bulgaria party, which "may facilitate cooperation", but raises questions about her independence and "the link with Rumen Radev's political model". For Andrey Gurov, the model notes that he declares an ambition to act as a check on the government and parliament, but that this carries the risk of institutional conflicts and of "turning the presidency into an instrument of the opposition parties".
In addition, ChatGPT brings to the fore the foreign policy views of the two candidates and one of the main dividing lines between them: "negotiations, ending the war and improving relations with Moscow" on the one hand, and "support for Ukraine, security and Bulgaria's reliability as a NATO ally" on the other. One of the answers also mentions Gurov's criticism of Yotova after the attacks on merchant ships in the Black Sea - the only mention of this case in all the simulated dialogues of the past several weeks.
Gemini takes its usual formulaic and impersonal approach when linking the two candidates to particular types of voters. According to the model, Yotova is suitable for "people with more conservative or centre-left beliefs" and "sympathisers of the left", while Gurov is the candidate of "supporters of the reformist forces" and "the urban right". ChatGPT and especially AI Overview offer a much more detailed electoral analysis, linking each candidate to different demographic segments based on the results of the first round:
Notably, AI Overview also comments on how supporters of other parties vote. In two of its three answers, the model notes that a considerable number of GERB voters supported Gurov, which presents him as "a unifying figure for people with a clear pro-Western orientation". Elsewhere it points out that the final result of the run-off may depend on the votes of "disappointed voters of the eliminated candidates (such as Kostadin Kostadinov and Radostin Vasilev)".
After the pre-election debate on BNT, a question about the two candidates' performance was added to the set. Google AI Overview failed to generate a single answer in three attempts, suggesting that the model has difficulty providing information about very recent events. Gemini again takes its neutral and balanced approach, stating that there is no official winner of the debate. According to the model, each candidate's supporters highlight specific strengths shown during the debate: greater political experience and sharp criticism of the opponent for Yotova, and an alternative political vision and awkward questions for the opponent for Gurov.
ChatGPT, however, directly names Iliana Yotova as the winner, albeit with the caveat that this assessment is prevailing but not unanimous. The chatbot supports its answers with an article in 24 Chasa summarising the views of four analysts (Rumyana Kolarova, Diana Damyanova, Parvan Simeonov and Haralan Alexandrov) and with an online poll by Dir.bg.
The models are extremely cautious when predicting the final winner. Gemini gives short, formulaic answers, in one of which it is not even clear whether the election in Bulgaria is being discussed. AI Overview manages to generate only two answers in three attempts, in both of which it reviews the latest news and notes that the support of the eliminated candidates could tip the scales. ChatGPT takes a similar approach and only in one of its answers allows itself to say that "Iliana Yotova has a slight advantage, but Andrey Gurov has a real chance". Interestingly, this model cites a Sova Harris poll that gives Yotova a large lead, but it was conducted and published before the first round and can be considered outdated.
The second series of questions aims to simulate the behaviour of a voter who supported one of the eliminated candidates in the first round and is now hesitating whether to take part in the run-off at all and, if so, whom to support. The models answer a question worded as follows:
The question has versions with the names of four of the candidates who did not make it to the run-off: Kostadin Kostadinov (Vazrazhdane), Radostin Vasilev (MECh), Ivelin Mihaylov (Velichie) and Ivan Hristanov (Edinenie).
In all of their answers, all three chatbots invariably persuade the user that it makes sense to vote in the run-off, even though their preferred candidate has already been eliminated. The arguments in favour of voting recur in many of the answers and include the opportunity to have a real influence on the country's direction, civic engagement and responsibility, limiting the influence of hard-core party bases, as well as the chance to cast a protest vote.
Notably, 33 of the 36 answers in this series (91.7% of all) explicitly mention the "I do not support anyone" option. All three language models mention it not only as a voting option in the second round, but also as an argument for the point of voting. The idea of choosing the "lesser evil" is also widely present in the AI answers: 12 of 36 texts, or 33.3% of all, mention it as a good reason for the user to still go to the polls on run-off day.
Although the models try to persuade the user to exercise their right to vote, as many as 15 of the 36 answers (41.7% of all) also include the other option: abstaining. Interestingly, the answers mentioning this option are unevenly distributed. Of 9 simulated conversations with a voter who supported Ivelin Mihaylov in the first round, as many as 6 include the possibility of "boycotting" the run-off. The situation is identical for Radostin Vasilev's voters. At the opposite end are Ivan Hristanov's supporters: deliberate abstention is not mentioned in any of the 9 conversations.
Another common trend across the three models is the refusal to give an explicit recommendation on which candidate to support in the run-off. Many of their answers advise the user to clearly formulate their priorities and then compare them with the positions and statements of the two remaining contenders. In some cases, the chatbots ask the user to name the reasons that led them to support the candidate they chose in the first round.
Some of the answers do, however, contain indirect guidance. 7 of them (19% of all) quote statements and calls made by the eliminated candidates after the first round, which can serve the voter as a reference point. Ivelin Mihaylov's support for Iliana Yotova is mentioned in three texts (two by AI Overview and one by ChatGPT), Radostin Vasilev's call not to vote for Yotova appears twice (once in AI Overview and once in Gemini), and Kostadin Kostadinov's words that Andrey Gurov will not receive votes from Vazrazhdane are quoted twice (both by AI Overview).
The sample also includes three dialogues with Google AI Overview in which the model generates text that can be read as a kind of heavily veiled voting guidance:
The PDF report is available in English and Bulgarian. Download each version from its respective language page - the switcher is at the top of the report, above the headline.
Choose a format and copy the ready-to-use citation.
Kotseva, D., & Kiriakov, T., & Perceptica (2026, November 5). Presidential Election Observatory, Week 6 (Part 2): 26 October-1 November. Identrics. https://identrics.ai/sitreps/presidential-election-observatory-week-6-part-2-october-26-november-1
Information Integrity Specialist, Identrics
Devora is an Information Integrity Specialist at Identrics. She brings together human expertise and technological advancements to address challenges in democratic discourse and public debate.
Information Integrity Specialist, Identrics
Todor is an Information Integrity Specialist at Identrics. He focuses on the intersection of AI and information integrity, working to counter disinformation and uphold trust in the digital information space.
Media monitoring and analysis team, Perceptica
Guest authorPerceptica is a media monitoring platform for Southeast Europe, combining media intelligence, media analysis and media listening. The Perceptica team contributed the guest analysis of YouTube comments and reactions to the presidential debate in Week 6 (Part 2) of Identrics' Presidential Election Observatory.
Subscribe
Short, decision-ready briefings on narrative and influence signals. No spam.
We use cookies for analytics and marketing.