You can search the news. You cannot research it.
"What actually happened with this over the last year" should be the easiest question a news product answers. It is the one none of them can. Search returns the newest article and the most search-optimised article, and the piece from fourteen months ago that explained the whole thing is on page four, behind a wall, or gone.
The NEWX assistant exists for that question — not as a chatbot for its own sake. You reach it from the search icon and ask in plain language.
How it answers
- With cited sources. It shows what it used. The answer and the evidence arrive together, not as a claim you are asked to take on faith.
- Across time, not just today. It can reconstruct what happened over a long-running story, which is what the archive problem above actually asks for.
- Across both feeds. Local and global questions are both in scope, because both feeds live in the same app.
- In your language. The app runs in English, Hindi, Marathi, Gujarati, Kannada and Punjabi.
The important part: it says when it does not know
Every AI product most people use would rather produce something plausible than admit a gap. In a news product that instinct is not a quirk — it is a hazard. The NEWX assistant is built to say when it does not have enough to answer, and that behaviour is treated as a feature to protect rather than an embarrassment to hide.
It is the same principle that governs the rest of the app: the bias mix that names the reason it cannot be computed, the Blindspot label that tells you a story has no local confirmation, the provenance sheet that shows the exact paragraph. Showing the shape of your own ignorance is what makes the rest of the output worth believing.
Hard guardrails on health and money
Two categories get treatment stricter than a system prompt suggestion, because both are places where a confident wrong answer does real damage.
| Topic | What it draws on | What it is prohibited from doing |
|---|---|---|
| Health | PubMed, MedlinePlus, WHO, NIH, ICMR and MOHFW. | No diagnoses. No drug recommendations. No dosages. The "not medical advice" disclaimer is attached programmatically, not left to the model to remember. |
| Markets | News coverage, with sources shown. | No buy or sell recommendations, under an equivalent enforced guardrail. |
The distinction worth noticing is programmatically. A disclaimer that depends on a language model choosing to include it is a disclaimer that will eventually be missing. Attaching it in code means it cannot be.
What it is good at being asked
- "What happened with X over the last year?" — the reconstruction question ordinary search cannot answer.
- "Why does this matter?" — background that assumes you did not read the previous eleven articles.
- "Who is involved in this?" — the players across the coverage, not one article's cast list.
- "What is still unresolved?" — the open questions a filed article treats as closed.
- Local questions — about your city or state, where a general-purpose assistant has nothing to work with.
What it will not do
- It will not diagnose you, recommend a drug, or give a dosage. Ever, by design.
- It will not tell you what to buy or sell.
- It will not fill a gap with something plausible. If the sources are not there, it says so.
- It is not a general assistant. It is built for news questions, over news sources, with news citations.
Frequently asked questions
What is the NEWX AI assistant?
A news assistant reached from the search icon. You ask a news question in plain language and it answers using cited sources, including reconstructing what happened across a long-running story — a question ordinary news search cannot answer.
Does the NEWX assistant cite its sources?
Yes. Answers come with the sources shown, in keeping with the rest of the app, where every claim in a summary traces to a publisher and a specific paragraph.
Will the NEWX assistant give me medical advice?
No. For health questions it draws on PubMed, MedlinePlus, WHO, NIH, ICMR and MOHFW, and it is prohibited from providing diagnoses, drug recommendations or dosages. The not-medical-advice disclaimer is attached programmatically rather than left to the model to remember.
Will it tell me what stocks to buy?
No. An enforced guardrail forbids buy and sell recommendations on market questions.
What happens if it does not know the answer?
It says so. It is built to state when it does not have enough to answer rather than producing something plausible — the same principle behind the bias mix naming why it is unavailable, and the Blindspot label naming a story with no local confirmation.
Can I ask the assistant about local news?
Yes. Local and global questions are both in scope, because both feeds live in the same app. Local answers depend on coverage in your area, which is deepest in Maharashtra today.