Aman Bhargava Bangalore India Data Visualization Designer & Developer aman@diagramchasing.fun
Aman Bhargava Bangalore India Data Visualization Designer & Developer aman@diagramchasing.fun

AI-generated text in Indian government documents

December 10, 2025
AIgovernmentIndiaanalysis

Why would you lie about how much coal you have?
Why would you lie about something dumb like that?
Why would you lie about anything at all?

I came across this excellent Wikipedia article on signs of AI/LLM writing and decided to try searching for presence of such text artifacts in Indian government’s websites and documents. I’ve been reading tenders and RFPs for research on another project and they tend to be extremely verbose and lengthy documents.

Maybe there was a cottage industry in manufacturing filler text for these documents before 2023, but it can safely be assumed that like many smaller industries of artificial fluff, this one has also been made more efficient through LLMs. Here are some of my finds.

Note: In case the original links stop working, they have been archived on Wayback Machine, linked at the bottom of this page.

“As of my last knowledge update…”

A quick search for site:*.gov.in "as of my last training" OR "as of my last knowledge update" surfaced a “Fact Finding report” on farmer suicides from the Maharashtra State Human Rights Commission (MSHRC), which contains this disclaimer.

“As of my last knowledge update in January 2022, here are some key initiatives and regulations in India related to farmer suicides.”

A report on a sensitive topic like farmer suicides, with research outsourced to a chatbot.

And another one from the Kerala State Higher Education Council, in a handbook for master trainers.

“As of my last knowledge update in September 2021, the University Grants Commission (UGC) in India had proposed the concept of an Academic Bank of Credits (ABC)…”

The UGC's Academic Bank of Credits proposal.

“It wasn’t just…”

Another common AI writing pattern is the phrase “It wasn’t just [mundane thing], it was [another mundane thing]”. A search for site:*.gov.in OR site:*.nic.in "It wasn't just" filetype:pdf shows up in various official documents.

A search result page full of government PDFs using the same clichéd opening.

Removing the filetype:pdf filter reveals even more, especially from the Press Information Bureau (PIB), which seems to use this construction frequently in summaries of speeches.

More search results without the PDF filter.

utm_source=chatgpt

A utm_source tag is used for tracking where traffic comes from. Seeing utm_source=chatgpt.com in a URL tells us the link was likely copied directly from a ChatGPT session.

My search was site:.gov.in OR site:.nic.in "utm_source=chatgpt.com".

It led me to this PIB press release which has a link containing the ChatGPT tracker, which points back to another PIB press release.

A government press release finding itself via a link from ChatGPT.

Copy-pasted as received

Sometimes, the copy-paste job includes the chatbot’s sign-off. A document from a pharmacovigilance committee at Lakhimpur Medical College and Hospital, Assam, ends with “Let me know if you need any modifications”.

The committee is ready for your feedback.

Here is the presence of “I hope this helps! Let me know if you have any other questions :)” in a presentation on “LLMs for Defense” on the Indian Navy website. This was delivered by an external person (seems like), but frankly, the provenance of this doesn’t matter to me because it exists on a gov.in domain.

Slide with bullet points that are clearly AI-generated

Another common mistake is copying the entire text that an LLM returns, with the UI labels. A user adding stuff to myscheme.gov.in seems to have accidentally copied the button text from the AI interface right into the document. This description for a scheme ends with a stray “regenerate response”. The text itself is also suspiciously generic and repetitive.

Just click the button for more generic text, I guess.

And it’s not a one-off. Here is another one on the same portal.

The benefits listed are so vague they could be for any industrial scheme anywhere.

Rich Tapestry

“Tapestry” (especially those that are rich, vibrant and intricately woven) is a purple word that LLMs are particularly fond of. A search for (site:.gov.in OR site:.nic.in) ("intricate tapestry" OR "vibrant tapestry") after:2022-12-01 shows its presence everywhere.

Our culture, our heritage, our economy... everything is a rich and vibrant tapestry.

Although since LLMs are trained on public and non-public data, it is entirely possible that the government was in a habit of describing everything as a rich tapestry and we have affected the training data.

“It’s not just about…”

A slight variation on the earlier theme, “it’s not just about” is another go-to phrase for LLMs trying to add depth. It appears in this transcript of ‘Mann Ki Baat’ from the Ministry of Information and Broadcasting and multiple other places.

From a Mann Ki Baat transcript.

Bonus: AI-Generated Images

AI-generated images are also making their way into official documents. These ones, from a Delhi SCERT document on “Nurturing and Design Thinking,” are riddled with spelling mistakes like “Survvey Form” and illustrations depicting European people which feel out of place. This document is full of such weird, lifeless and waxy illustrations.

AI-generated image with synthetic, waxy features.

Unnecessary Emojis

LLMs have a tendency to sprinkle emojis everywhere. This page on cybersecurity from AIIMS Bhubaneswar is an example.

Stay safe online 💻🛡️💡

Markdown Leftovers

This PDF from the Government of Odisha has some stray markdown.

Render markdown in your brain.

Full documents

This article so far has been stuff I could find through dork searches, but the scale of the problem is unidentifiable in full because LLMs have gotten better at removing these kind of artifacts. But, the same garbage feel remains if you look at articles in full and not just through keyword searches. Walls of text of generic, absolute garbage. As I find more, I will keep adding to this section but for now, this page by the Comptroller and Auditor General (CAG) of India has an article about Generative AI written, obviously because of knowing satire, entirely with an LLM.

An article about AI, written by AI, on the CAG website.


Maybe this is just as bad as it was before.

The perverse incentive structure of length equaled to rigor that exists not just in the government but our schooling and education as well means that this was inevitable.

Permalink

Tap to give validation

52

Responses

6
Aatman Vaidya
Adithya Nair
Deepali Kank
Affective Old Truant
2
Aman Bhargava
1
thedivtagguy

Replies (10)

Pushkar Sinha
20 days ago 🦋 Bluesky
View

Even before AI generated content, Indian government bodies and the legislative branch (courts, lawyers) use horrendously complicated language. Complex enough for someone who knows English well, let alone those in towns and villages who don't.

Pushkar Sinha
20 days ago 🦋 Bluesky
View
I came across the Plain Language initiative of the US government last year, which has a wonderful guide on writing simply that a regular person can comprehend. digital.gov/guides/plain...
Aman Bhargava
20 days ago 🦋 Bluesky
View

Wow this is great, thanks for sharing!! A lot of talk on design these days, but stuff like this always gets missed out.

Aman Bhargava
20 days ago 🦋 Bluesky
View

Absolutely agree! A lot of this is/was/will be just content put together by think-tanks anyway, but I'd argue at least that took some _thought_ (even if it was useless/overly complicated by design). Now all that can be generated at scale.

Adithya Nair
20 days ago 🦋 Bluesky
View

It's depressing to see such clear signs like this in documents about these critical topics. Feels like we're accelerating to a future where everyone only engages with material through a language model as intermediary; to a future that's indirectly governed by AI.

Adithya Nair
20 days ago 🦋 Bluesky
View

To be fair, this is prevalent in every part of our society now. My larger problem with this is that it incentivizes people to be less accountable about their own work at a scale never really seen before. You could trust that at least someone cared to do the work pre-2022.

Aman Bhargava
20 days ago 🦋 Bluesky
View

Precisely! Even if it was stupid officialese, at least it took effort. utm_source=chatgpt is just depressing because do I really want press releases and reports researched/written by a random session in Perplexity/ChatGPT/Gemini? Probably not.

InsanityMonk
about 4 hours ago Reddit
View

Think of any worst scenarios. Our government will be there.

beard__hunter
about 3 hours ago Reddit
View

Good one. AI should be used but not blindly.

Mentions (1)