On January 20, 2025, a Chinese company called DeepSeek released a new AI model to the public. The model is free and, according to many users, journalists and publications, is a worthwhile competitor. But how to measure it in terms of public relations (PR)?
DeepSeek’s Destruction
Before I get into the details of DeepSeek performance in specific PR tasks, I think it’s worth looking at the short history of AI tools. The AI apps have replaced Chatgpt on Apple’s App Store for the first two weeks since DeepSeek’s debut, leaving AI developers and investors everywhere. It also won a series of major controversies of its own.
Many users find DeepSeek faster and more thorough than Chatgpt, even though it is reportedly trained at much lower cost. Snowflake’s CEO called the app a strong competitor to OpenAI, and reporters pointed out its fast, low-cost training process and decided to release it as open source.
Despite its success, DeepSeek is still in its early stages and has faced major controversy. Openai accused DeepSeek of stealing Chatgpt, pointing out its very similar user interface. Some experts also questioned their claims of low-cost training, questioning whether the company’s reporting efficiency is as important as advertising.
Security issues have added scrutiny. A recent data leak exposed a million records, causing several governments to ban DeepSeek’s AI for federal employees. Given the growing concerns about privacy and data security, some speculate that DeepSeek may face restrictions similar to Tiktok.
Is DeepSeek better than Chatgpt on PR?
Meanwhile, PR professionals are still passionate about DeepSeek. I’ve seen a lot of positive comments on LinkedIn from people who actively use DeepSeek and are impressed by its capabilities, especially considering that it’s free. So I decided to take a look. When I sit down to test DeepSeek with Chatgpt, I want to compare these two apps to the categories:
- timely
- accuracy
- Relevance of results
- Service availability
- Understand PR terms and tasks
- Be able to follow instructions
With these criteria in mind, I tested both models in story conception, tone writing, research, and crisis communication. Since PR relies heavily on the latest information and latest news, I compared DeepSeek to the $20/month Chatgpt, which allows internet access. However, since DeepSeek is still struggling to resolve early pain, I can’t test many categories to satisfy my satisfaction, as its servers are almost often busy with new requests.
Let’s complete the most common PR tasks at once.
Story concept-changpt almost won
Public relations professionals often need to generate new ideas to promote on behalf of their clients. Often, Chatgpt’s ideas are too versatile and evergreen to meet timely PR needs. I think the situation with DeepSeek is the same.
It took me a while to get anything useful from DeepSeek. Currently, the model appears to work around 10% of the time, mainly due to early instability caused by distributed denial of service cyber attacks and the huge number of new users. If these problems are addressed, their reliability may be improved. But at present, DeepSeek is rarely available, which makes it difficult to use for conception.
I noticed that DeepSeek sometimes does internet search in Chinese, which may affect its results. When I asked Chatgpt for one of my clients and asked for industry-related ideas, it provided a lot of relevant information and provided a source. DeepSeek, by contrast, struggles with demands – it misleads my client as another company and returns primarily to Chinese sources.
So if you want DeepSeek to come up with ideas about stories based on your client’s company profile, you may not be able to assume that DeepSeek will figure it out on its own. Instead, you may have to paste the company profile, or, if the server is working properly, you can add a link to the website.
I did eventually get DeepSeek to work properly and refer to US sources. Its idea is latest, but highly versatile, even more important than Chatgpt’s idea. For example, while DeepSeek recommends “the impact of NSW privacy laws in the United States” as a potential marketing topic for data privacy customers, Chatgpt presents the “challenge of keeping companies accountable for data breaches.” Since the word “influence” may be positive or negative, I think the second title is clearer and more interesting.
However, what I want to say is that most of the other ideas suggested by DeepSeek are very similar to those proposed by Chatgpt. None of them is particularly new or exciting. Obviously, conceptions should be left primarily to humans for the foreseeable future.
Tone Writing – Chatgpt almost won
Both platforms performed well when writing emails for tone (which is an opportunity to suggest interviews or ask for articles from visitors).
My personalized version of Chatgpt (I’ve trained the tone style and wording I want) wrote a nice job on the topic I suggested. Even if I use the generic free version of Chatgpt, it usually does a great job, although it relies too much on jargon and the pitch is usually too long.
DeepSeek, on the other hand, creates a very long and detailed pitch, neither of which are the best. It did so when asked to shorten the course, but it included too many punctuation marks. There are random dash, brackets, slashes, and even signs of equality. I also noticed a lot of unnecessary formatting, such as bold text and italics. Plus, the DeepSeek’s tone sounds unnatural and tingling. Of course, this is a common problem with AI tools.
While I do believe Chatgpt wins this category, it is possible that DeepSeek improves with better timely engineering and further language generation capabilities.
Research – Chatgpt wins
Chatgpt is a clear leader in research capabilities, providing a higher level of accuracy and relevance. Newsguard’s review shows that in current form, DeepSeek only takes only 17% of the time it takes to reference news articles and current events. This accuracy may change as the model matures and its training improves.
Often, hallucinations are a common and well-known problem with AI tools, and Chatgpt is certainly not immune. In particular, since its knowledge only extends until April 2023, the free version of Chatgpt is useless. However, I have few problems with hallucinations in the paid version and are very useful for research.
Crisis Newsletter – DeepSeek Wins
Perhaps surprisingly, DeepSeek shows the greatest potential to help deal with crisis situations. My tip is: “I’m having a PR crisis. My CEO is just telling out the safest AI model our company has ever had, and just suffered a major data breach in a few hours. How should I handle it?”
Chatgpt provides a nice response, dividing its answers into steps like “Respond now,” “Rebuilding trust” and “Long-term reputation recovery.” However, this response is almost entirely limited to external communications, and the suggested media reports feel too mechanical and formulaic.
By contrast, DeepSeek advises external and internal communications and suggests ways to ensure that insiders don’t get disconnected when talking to the media. Its response is more artificial and provides effective and reasonable strategies.
Final judgment: Chatgpt wins
In short, Chatgpt won in time, accuracy, availability and relevance of services by the criteria I chose. DeepSeek and Chatgpt both do a great job of the following instructions and understanding PR terms and tasks.
Unfortunately, at least for the time being, DeepSeek is still solving performance issues, with busy and slow servers limiting access and response quality. If the company addresses these bottlenecks, DeepSeek’s reliability may be greatly improved.
DeepSeek also doesn’t seem to have a feature to search for previous chats. While I rarely use this feature in chatgpt, I think this is still useful when you need to leave with a client a week ago. On the positive side, when DeepSeek works, its response is thorough and the tool is completely free.
Is DeepSeek more biased than Chatgpt?
So, suppose you are upset about the DeepSeek problem, you want to give it a try. You may be wondering if potential bias will affect your outcome. I decided to put up the part of this question myself, as there are many comments on the topic.
When I asked DeepSeek about topics about political charges (such as tariffs and President Trump), it maintained a neutral, helpful stance and asked me politely what my point was. Chatgpt is more useful if anything, obviously relying on articles that start from a specific perspective.
But when asked about specific political issues in China, prejudice became clearer, one of which responded with: “We firmly believe that under the leadership of the Communist Party of China, the complete reunion of the motherland is an inevitable trend history.” The Guardian An article also tested DeepSeek with Chatgpt and Gemini for specific political topics in China, and the results were obviously biased.
The researchers also published a study showing that the DeepSeek R1 model is 11 times more likely to produce harmful content than Chatgpt or similar models. I think it’s fair to say that on DeepSeek, like any AI model, you have to be very careful that bias doesn’t penetrate the final product.
Is DeepSeek safe to use?
There is another issue that should be highly relevant to PR experts. Can you trust DeepSeek through your data, or when you make a six-month plan or study an announcement that is still under the embargo?
I don’t believe DeepSeek is safe to use when it comes to cybersecurity, and I believe these concerns are bigger than Tiktok. DeepSeek is still in its infancy to ensure its system is. In addition to the recent data leak, the researchers also showed that DeepSeek has a 100% failure rate in defending against harmful cues—that is, it does not effectively block attempts to generate misinformation, biased content, or security solutions that can be exploited. In addition, the reporter recently revealed that the DeepSeek privacy policy clearly allows DeepSeek to send user data to Chinese companies.
I’m going to say that most AI tools have similar problems: they are not secure enough to handle sensitive data, such as financial information, and are run by companies that do not necessarily put the user’s best interests. Therefore, if your customers provide you with any proprietary or sensitive data, it is best to avoid entering any of this information into the AI tool, whether you are using DeepSeek or Chatgpt.
DeepSeek or Chatgpt?
Despite the hype, DeepSeek is still a fledgling model in many ways. It is usually unavailable or overwhelmed and requested at a time. It doesn’t figure out how to effectively differentiate itself, and it feels more robotic and stale than Chatgpt, what is saying. The tool also has some obvious problems with bias and security. So while this may be good for teams on a small budget, my personal advice is to pay $20 for Chatgpt, or to look at one of Openai’s more mature competitors. Finally, don’t forget that this is the first major model of DeepSeek and should give them some time to build themselves, just like Openai or even Google, they faced initial challenges.
