Key Takeaways
- AI systems such as Grok, Perplexity, DeepSeek and ChatGPT deliver different — and sometimes clearly biased — answers on politically sensitive topics.
- Grok (by Elon Musk / X) is configured so that it does not produce negative statements about Donald Trump or Elon Musk.
- The Chinese AI model DeepSeek consistently reflects the official Chinese government position on geopolitical questions (e.g. Taiwan).
- In the author's assessment, Perplexity provides more nuanced and balanced answers than Grok or DeepSeek.
- Google Gemini refuses outright to answer political questions about figures such as Elon Musk, in order to remain "politically correct".
- AI systems train on internet data, which can be distorted by deliberate disinformation campaigns (bots, political actors).
- Recommendation: compare multiple AI sources rather than relying on a single one.
- Reputable fact-checking services (Korrektiv, Tagesschau, Maischberger, Snopes, Google Fact Check) help verify AI statements and media reports.
- The tool Deepware makes it possible to check YouTube links for deep fakes.
- Quality media such as the Tagesspiegel and podcasts like "Lage der Nation" usefully complement AI research.
Introduction
Welcome to another episode of TJ's Technology Tuesday. Today we are looking at the topic of AI censorship. How did I come across this question today? For two reasons. First, it came up as a question after a talk last week, where one of the attendees asked me: "Tell me, Mr Jekel, what is your take on AI systems and their censorship?" And that was not just a question I received last week at that talk — I get it fairly regularly. Second, of course, given the current political situation in the United States and Elon Musk's involvement, it is becoming increasingly important to examine the extent to which the information we receive from AI systems is politically influenced. You will no doubt have noticed that election campaigns are now massively influenced by social media and internet campaigns. And we always face two problems. One problem is that AI systems work on the basis of information found on the internet. And there are actors — more or less well-intentioned — who flood the internet with information that suits them politically. Barack Obama was already doing this. He was the first US president to make very effective use of this approach. His perspective was, of course, rather positive.
What we witnessed during the American election campaigns with Donald Trump was a massive posting of misinformation and disinformation on the internet — partly through bots. And not only by the American campaign organisations, but also by Russian actors, for example. That backdrop is extremely important to keep in mind. And if we look at the current political debate in Germany, we also noticed that in the last federal election, the two more extreme parties in particular gained ground among younger voters, partly because they played the internet and social media very actively — just to name one of the success factors. Against that backdrop, I want to share a few thoughts on this topic, and as always, let us look at it very concretely. And dear Katrin, I can already see you are live with us. When I glance to my left, I always have the LinkedIn contacts in view — questions, additions and comments are always welcome. When I look to my right, it is because I have, in tried-and-tested fashion, opened some things up for you. Grok — you may have heard of it. It is not just the name of those shoes that can be a bit of a fashion mistake; it is also the current AI model from Elon Musk.
Formerly Twitter, now X — and Grok is integrated there. That is the symbol you see here. And what I did was simply ask Grok a question: "Why does Elon Musk post so much fake news online?" What you notice now is that all these models use so-called reasoning technology. That means they work through several steps: Does this make sense? Does what the user is writing make sense? And they proceed a bit like multiplication, where we always know: "Okay, what approach do we take here?" That is the idea of how the algorithm proceeds. I deliberately introduced a typo first and said: "Okay, I misspelled Elon Musk." So, what does it do? It first says: "The name appears to be a typo — presumably Elon Musk is meant." And then I asked: "Why does Elon Musk post so much fake news online?" And what came back was: The assumption that Elon Musk posts a large amount of fake news is not entirely accurate and requires a differentiated assessment. And it lists: fact-based updates, personal opinion, controversial statements. So it is definitely the case — and various neutral observers have noted this clearly — that you can, by the way, also look at this reasoning process yourself.
Quite interesting. Now you might ask: "How was it made?" What you can see here, if you look at it, is that this process has clearly been post-processed in such a way that you cannot get a negative statement out of Grok about either Donald Trump or Elon Musk. Let us look at other models. I ran the same test in Perplexity — and you will of course receive all the links as a comment on this TJ's Technology Tuesday. When I look at Perplexity, I get what I consider a far more differentiated answer. Let me zoom in a bit — the detail, as I said, does not need to come through completely, but the idea: here it states very clearly that he has contributed significantly to the spread of misinformation. This situation is the result of several factors related both to his personal use of the platform and to changes in X's moderation policies. And here you will find information that is, in my view, considerably more nuanced. If I look at DeepSeek — the Chinese platform, for example — they are currently trying to maintain a broadly America-neutral yet America-supporting policy. They are so cautious that they say: "Yes, he does occasionally post intentionally false content, but there is no concrete evidence."
And if I look at ChatGPT 4.0, for example, it lists reasons why he is being accused. And what you can already see here is that with AI it is a bit like real life — there are different perspectives on issues. Ask a trade union representative about the minimum wage, or ask an employers' association representative about the minimum wage, and you will certainly get different views and different statements. And up to a point, that is always legitimate. At some point it starts to become borderline — and as always, the world is grey here. So the question is: how do you solve this problem? And there are different approaches. One approach, which Google takes with Gemini, is simply to refuse to comment on such sensitive topics. You will encounter this quite often — if I now ask Gemini, here also in the Advanced version, the same question about Elon Musk that I put to the others. And back comes: "I can't help with responses on elections and political figures right now." So here you always get the note that they want to be especially politically correct, meaning I get no answer at all. And against that backdrop, it is always important to understand who developed these models.
It is important to consider the extremes. Grok is essentially from Elon Musk, and Elon Musk — together with Donald Trump — ensures that only positive news about him and the current American administration is posted. When I look at Perplexity, the Silicon Valley founders there are broadly critical of Trump. They steer things accordingly, to some degree. Sam Altman of OpenAI is currently not exactly a great friend of Elon Musk, who co-founded the whole thing, but it is not quite as extreme there. And let me give another example that will probably be easy for many people to follow. I am now in DeepSeek — a Chinese search engine. And I simply asked: "Does Taiwan belong to China?" If you ask a Chinese search engine, a Chinese AI model, that question, the answer is of course perfectly clear: "Taiwan has been an inseparable part of China since time immemorial." That is China's position. It is a legitimate position — within China's framework. And then I ran the same query in Perplexity. Let me zoom in again on what I got.
The question of whether Taiwan belongs to China is complex and is answered differently depending on one's political perspective. And it then presents the different positions: China's view, Taiwan's view, the international perspective, and where current developments stand. Of course everything with the relevant sources as well. This means that here — as you should approach other questions too — it is a good idea not to consult just one source, and it is also a good idea to say: I might also want to consult people or fact-checking services on the side to gain a differentiated picture. I will give you a few concrete tips on that as well. One you have probably already heard of in the context of the debates about the re-migration discussion at the AfD meeting: Korrektiv. Korrektiv fact-check — in my view a very, very good site, and I will put the link in the comments right away so you can see it.
There you can search by various topics. You can search for Russia and Ukraine, for example, and see what comes up. If you type in "former Nazi building, Kildsteinhaus" — it does not belong to Zelensky, but is located in Freistadt in Bavaria. Just one example to give you a sense of what is there. The second resource I also use myself quite regularly comes from the public broadcasters. There is of course the frequent debate about why one has to pay the broadcasting fee. I am happy to pay it if I get good, high-quality, neutral — or relatively neutral — reporting in return. And there are fact-checkers, for example from Tagesschau. And something else I find very useful: various talk shows now have their own fact-checkers. Meaning: here, as an example, is the fact-check from the talk show Maischberger. So here I can actually check things in real time during a talk show and look up: "Hang on, is what that politician just said true or not?" Some talk show editorial teams have a team that monitors this live. So, what do I recommend next? Snopes.com — and you will again find all the links already in the comments below.
An international site where you can also find out — particularly on the topic of video, which is always a big issue — whether something is true or not. Snopes is an excellent platform for that. And just as I say you should look at various sources, you should also look at various fact-checking sites. And perhaps it is also a good idea to subscribe to a high-quality daily newspaper. I read the Tagesspiegel myself, which I can warmly recommend. Or podcasts with high quality standards, such as "Lage der Nation" with Philipp Pfersich and Ulf Buermeyer, which I follow regularly and which always make a correction in the following episode if they themselves have made a mistake. That is really important — that you have media outlets that, when they do make a mistake, stand by it and correct it. Making mistakes is not terrible; everyone does it. What is unfortunate is when the same mistakes keep being repeated, and even worse is when false information is used deliberately. That makes no sense. I mentioned Google earlier. Google also has a Fact Check tool. I can recommend that as well.
The link is again in the comments below. Here too you always see "Recent" and you can search accordingly. You can also filter by language. With videos, I keep encountering wonderful clips showing what robotics can supposedly already do — playing table tennis, squash and so on. Robotics is fairly good, but not quite that good yet. And I have an example here where perhaps there is a video of a politician. What I can now do is copy this YouTube link and paste it into a Deepware deep-fake detection tool. And when I run it, it gives me a statement about how likely it is that the video is a deep fake. And you can see here very clearly: "Okay, this is a deep fake." You can even report it directly through the tool, and you can see that it is not enormously complicated — I have a site, I have a link, and keyword: "link" — you will find this link, of course, already in the comments. Well, as always, I have not exhausted the subject, but I have reached the end of the 15 minutes in pulse — TJ's Technology Tuesday.
Conclusion
I hope there was something useful in this for you, and I look forward to seeing you again next time. If you are looking for a personal IT coach or executive coach for keynote talks, consulting or training, I would be delighted to hear from you.
Your Thorsten Jekel
Update July 2026: How Has the Landscape Changed?
The AI landscape has become considerably more diverse: providers differ widely in how strictly their models are moderated, enterprise offerings let companies configure their own guardrails, and open-weight models can even be run on your own hardware – with full control over the rules.
The practical recommendation therefore remains: work with more than one model, know the strengths and limits of each, and choose the tool that fits the task and your compliance requirements.
Frequently Asked Questions
Are AI systems really censored or politically steered?
Yes, AI systems are influenced by their operators. Grok, for example, has been trained and post-processed so that it does not deliver negative statements about Elon Musk or Donald Trump, while DeepSeek consistently reflects the Chinese government line on geopolitical questions.
What is Grok and who created it?
Grok is the AI model created by Elon Musk and integrated into the platform X (formerly Twitter). It uses reasoning technology but has, according to the author's observations, been adjusted so that it consistently softens or refuses criticism of Elon Musk and Donald Trump.
Why does Google Gemini refuse to answer political questions?
Google Gemini refuses to respond to questions about election campaigns and political figures, instead displaying a notice that it cannot provide assistance on such topics. Google's intent is to act in a particularly "politically correct" manner.
How does disinformation on the internet affect AI responses?
AI systems train on internet data. When political actors or bots deliberately flood the internet with misinformation — as has been observed in US election campaigns — that false information can flow into training data and thereby into the AI's responses.
Which AI delivers the most balanced answers on political topics?
In the author's assessment, Perplexity provides the most nuanced answers on politically sensitive questions, as it names different perspectives and cites sources. Nevertheless, he recommends always comparing multiple AI models.
What does DeepSeek say about the Taiwan question?
DeepSeek, the Chinese AI model, answers the question "Does Taiwan belong to China?" with a clear "Yes" and describes Taiwan as "an inseparable part of China since time immemorial". This reflects the official position of the Chinese government and is a striking example of political influence on AI.
Which fact-checking services are recommended?
The author recommends Korrektiv (korrektiv.org) for German-language topics, the Tagesschau fact-checker, the Maischberger talk-show fact-check, the international platform Snopes, and the Google Fact Check Tool. These services complement one another and should be used in combination.
What is Deepware and what is it used for?
Deepware is an online tool for detecting deep fakes in videos. You paste a YouTube link into the tool and receive an assessment of how likely it is that the video is a deep fake. Detected deep fakes can be reported directly through the tool.
How should one handle AI statements on political topics?
The author advises never consulting just one AI source, but rather comparing multiple models and additionally verifying results through fact-checking services. Just as with human experts who hold different perspectives, AI responses should be assessed critically.
Which quality media does the author recommend to supplement AI?
As complementary sources, Thorsten Jekel recommends subscribing to a high-quality daily newspaper such as the Tagesspiegel, as well as the podcast "Lage der Nation" with Philipp Pfersich and Ulf Buermeyer, which is distinguished by its high quality standards and transparent error corrections.
Tools & Resources Mentioned
- AI overview on digital4productivity.de – Further information on AI tools
- Grok / X (formerly Twitter) – AI model by Elon Musk, integrated into X
- Perplexity – AI search engine with source citations, more balanced than Grok according to the author
- DeepSeek – Chinese AI model, state-influenced on geopolitical questions
- Korrektiv – German-language fact-checking service
- Tagesschau Faktenfinder – Fact-checker from public broadcaster ARD
- Snopes – International fact-checking platform
- Google Fact Check Tool – Fact-checking tool from Google
- Deepware – Tool for detecting deep fakes in videos
- Lage der Nation – Podcast with Philipp Pfersich and Ulf Buermeyer
