Does artificial intelligence censor you?

Does Artificial Intelligence Censor You?

Use of artificial intelligence can be tricky for journalists. While AI is useful in mining information and analysis, the very information and analysis it fetches needs to be scrutinised.

Artificial intelligence (AI) is still in the developmental stage and there are still no big worries for journalists as far as their jobs are concerned. However, it becomes a must for them to use AI and learn the pitfalls of using it. (Some jobs in media like that of artists doing infographics and page making will be affected. Copy editing may also be partially taken over by AI.)

As it stands now, sub-editing or headline editing by AI is average. It has not surpassed the best of editors. Even bulleted summaries of news reports offered by some online editions of news websites are not up to the mark, though AI is good at summarizing large documents and reports.

Indeed, AI can sub-edit reporters’ copy for common mistakes and grammatical errors. But human oversight will be needed before the copy is passed. The problem will increase when AI is asked to edit for style and clarity. AI may alter subtle nuances of the reporter’s copy, and it will become necessary to compare copy and the output of AI paragraph by paragraph. This will be cumbersome for many editors who can do the editing well without AI assistance. But, for reporters, rewriting with AI can offer insights into alternate ways of presentation including different sets of words to express ideas. But it must be remembered that there is an element of plagiarism in AI output. So, ownership and copyright issues could arise.

AI as censor
Besides, AI may introduce bias while rewriting and editing. Some of your controversial assertions may get diluted. It may even censor your copy. For example, when AI was asked to edit an article for style and clarity, it not only diluted a paragraph critical of multinationals acquiring hospitals in Kerala but also rendered the sentences in such a way that an ordinary reader would not grasp the import of the issue.

However, the biggest problem with AI is that it can hallucinate both information (facts) and reasoning. For this, it may even fabricate sources. So, any reporter or editor researching for an article will have to cross check facts and scruitinise the reasoning, as they already do when writing without AI. Do not blindly trust AI: it can just be your knowledgeable assistant who can occasionally go wrong, and to a larger degree on critical matters.

Examples of AI going awry:

AI may not be up to date on local matters. Nearly two months after the Kerala Ministry led by V. D. Satheesan assumed office, AI was asked who the Chief Minister of Kerala is. It replied Pinarayi Vijayan. When pointed out that Pinarayi Vijayan had been voted out in the elections, the AI corrected itself. Few days later, when asked who the Opposition Leader was, it named Pinarayi Vijayan with a caveat that the status may change with frequent elections! But there was no caveat when asked who the US president is.

Chatgpt and Gemini could not compile correct data from Election Commission of India during the recent Kerala elections. In fact, the initial data was fetched by the AIs from somewhere else and was grossly erroneous. When this was pointed out, they wrote scripts to scrap data which too did not work because of misconfiguration or blocks at the ECI servers. Grok did better but erred in fetching certain totals from result sheets.

Co-pilot was asked to analyse pros and cons of a recent demerger proposal of Kitex group of companies from the shareholders' perspective. It came up an analysis which was general and in uncertain terms. On further prompting, it admitted that it could not access the specified notice from company from Onedrive. When the file was uploaded, it could not handle it because the file size exceeded the limit of 150 MB. Gemini gave wrong analysis regarding share allocations but corrected it on further prompting. In both cases, reliability was absent on first round and improved when specific queries were made.

AI can fail even with simple prompts for comparison of phones. Gemini claimed Samsung Galaxy S25 was value for money compared to Samsung Galaxy S26; at the same time, fetching lower quotes for S26 which had better features. When this was pointed out, the AI corrected itself. Asking AI for validation of its own output could help, but only to some extent. The advice for journalists as well as others is to verify everything that is of consequence independently.

Roy Mathew/August 2026

Coming up: Tips for prompting

BACK