69 minutes of campaigning: what the fallacies in Lula and Flávio Bolsonaro’s speeches reveal
Across 69 minutes of campaign speech, Jiddu detected 53 possible fallacies in Lula’s address and 95 in Flávio Bolsonaro’s. The difference grew after adjusting for speech length and severity.
My vote is for Lula. I am stating that preference before the numbers because it affects what I notice and what I consider important.
On August 16, 2026, President Luiz Inácio Lula da Silva and Senator Flávio Bolsonaro opened their campaigns in political strongholds. Lula spoke at Vila Euclides Stadium in São Bernardo do Campo. Flávio addressed supporters on Copacabana Beach in Rio de Janeiro. The Associated Press described them as Brazil’s two main presidential candidates.
I captured both speeches with Jiddu Context and compared the complete transcripts. Together they contained 69 minutes and 15 seconds of speech, 8,084 words, and 597 sentences.
The fallacy analysis produced the clearest difference between the two addresses.
The central figures
In Lula’s 36-minute speech, Jiddu detected:
- 53 possible fallacies;
- 11.1 findings per thousand words;
- 4 high-severity findings;
- 8 instances of .
In Flávio Bolsonaro’s 32-minute speech, it detected:
- 95 possible fallacies;
- 28.6 findings per thousand words;
- 40 high-severity findings;
- 30 instances of .
Flávio’s speech was shorter and still produced 42 more findings. After adjusting for the number of words, its fallacy density was 2.6 times higher. The gap in high-severity findings was tenfold.
What the detector found
was the most frequent pattern in Flávio’s speech, with 30 occurrences. appeared 14 times, 10 times, 9 times, 6 times, and 5 times.
In Lula’s speech, led with 12 findings. appeared 8 times, 6 times, and 5 times. Most of Lula’s findings were classified as low severity; Flávio’s result was split much more heavily toward high severity.
A fallacy label does not mean that a sentence is false. It identifies a possible failure in the path from reasons to conclusion. A candidate can use a bad argument to defend a true proposition, just as a polished argument can begin with a false premise. Fact-checking and fallacy detection answer different questions.
Two ways of building a campaign speech
The difference also appears in the kinds of claims extracted from each transcript. Numbers and dates accounted for 33.2% of Lula’s claims and 17.4% of Flávio’s. Categorical and causal claims made up 74.2% of Flávio’s claims, compared with 55.9% of Lula’s.
Lula built much of his speech around personal history, the labor movement, education programs, and comparisons between governments. Flávio relied more on conflict, public safety, blame directed at the current government, economic anxiety, and promises of decisive action.
Both speeches were classified as explicitly partisan with 99% confidence. That is unsurprising at campaign rallies. The useful comparison lies in how each candidate tried to persuade an audience already inclined to applaud.
Why this weighs on my vote
I already intended to vote for Lula. These results reinforce one reason for that choice: in these two speeches, his argument depended less on reasoning shortcuts, especially the severe ones.
Political speech will always use emotion, repetition, symbols, and selective emphasis. Those devices are part of persuasion. I become concerned when fear, false choices, broad accusations, or an unsupported causal leap carry more weight than the reasons offered to the voter.
A president needs to defend decisions outside a friendly rally. Arguments eventually face Congress, the press, courts, public data, and people who do not share the speaker’s assumptions. A speech with fewer reasoning failures does not guarantee a better government. It does give the audience more material to examine.
What these figures do not prove
Jiddu’s analysis is automated. The detector compares passages with a closed catalog of fallacies, keeps the original excerpt, and explains why it applied a label. It can still misread irony, omit context, split a sentence badly, or treat campaign shorthand too literally. Jiddu has not published a human-labeled accuracy benchmark for the fallacy detector, so every finding remains open to human review.
I also chose not to turn the fact-check verdicts from these sessions into a truthfulness ranking. Some extracted claims still confuse verification of the underlying proposition with confirmation that the candidate pronounced the words. A political comparison should not hide that limitation merely because the current result favors my candidate.
The numbers describe these two speeches, not the candidates’ entire careers or every future campaign appearance. Repeating the analysis across debates, interviews, and new rallies will show whether the difference persists.
Examine the passages yourself
The complete sessions remain public. Each finding links to the part of the transcript that prompted it, so readers can agree, disagree, or report a classification that looks wrong.
I plan to repeat this comparison during the campaign. The next useful result will come from applying the same method to more speeches and publishing the passages behind every number.
Frequently asked questions
- Does a fallacy mean that a candidate lied?
- No. A fallacy concerns the reasoning that connects premises to a conclusion. Truthfulness requires a separate review of the underlying factual claim and its evidence.
- Why compare findings per thousand words?
- The speeches had different lengths. Normalizing by word count makes the density of detected reasoning problems more comparable than raw totals alone.
- Did Jiddu choose a candidate?
- No. Jiddu labeled passages under the same fallacy catalog. The author’s support for Lula is disclosed, and readers can inspect the transcripts and challenge individual findings.