Neutral News AI

We just got our Chrome extension for Neutral News AI published in the Chrome Web Store.

The extension brings our article analyzer directly into the browser, so instead of pasting links into a separate tool, users can open a news article and run the analysis on-page while they’re reading.

What it shows:

- article summary

- topic

- political orientation

- bias strength

- credibility level

- trustworthiness score

- sentiment

- readability

- emotional tone

- clickbait level

- subjectivity

Important limitation: this is not a fact-checker. It does not decide whether every claim is true or false. The goal is narrower: help people inspect how an article is framed, how it reads, and what credibility / quality signals stand out.

Why we built the extension:

The web analyzer worked, but it added friction. People had to copy a link, leave the article, and run it elsewhere. The extension makes the experience much more natural.

What I’m trying to learn now:

- whether the extension format is meaningfully better than the web analyzer

- which signals people actually care about most

- which article types or sites break the experience

- whether the value prop is clear enough to convert first-time users

Would love feedback from other founders on:

- positioning

- onboarding

- Chrome Store conversion

- where you’d promote something like this next

Chrome extension: https://chromewebstore.google.com/detail/neutral-news-ai/fbcjfaokmkiphodbikjkillklekeeddb

A few weeks ago I shared the story behind Neutral News AI, how it started with frustration at biased, fragmented news and turned into a project focused on transparent, evidence-based journalism.

Today I want to go deeper into what the product actually does and what’s coming next.

1. The Core Idea

Not “neutral news.”

Not ideology.

Just transparent, multi-source evidence.

Our system takes any news topic, then:

1. Collects coverage from multiple reputable outlets (left / center / right).

2. Extracts all atomic claims from the combined articles.

3. Retrieves evidence for each claim.

4. Runs MNLI fact-checking to classify each as: Supported / Contradicted / Inconclusive.

5. Generates a concise, source-bounded summary with all citations visible.

Everything is traceable.

Every sentence links back to its evidence.

If something is uncertain, we show it.

This is the part that early testers said “finally looks like journalism built on receipts.”

2. What we’re building now (the fun stuff)

A) Browser Extension (coming soon)

Paste nothing. Click nothing.

You read an article → click the NNAI icon → instantly see:

• Political Bias & Bias Strength

• Credibility & Trustworthiness

• Sentiment & Emotions

• Subjectivity

• Readability

• Clickbait Detector

• And a summary for every article

Basically an “X-ray mode” for any news article.

 This is the feature that most testers have asked for, and we’re already prototyping the UI.

B) B2B / White-Label API

Several journalists, analysts, and creators asked for programmatic access to our pipeline.

So we’re developing a clean API that lets anyone:

• Submit a URL → get structured claims back

• Retrieve multi-source evidence packs

• Run bias detection, sentiment, credibility signals

• Get MNLI fact-check results

• Pull neutral summaries with citations

• Retrieve comparison data across outlets

We want to support media outlets, researchers, fact-checkers, and even YouTube/podcast creators who want to provide more transparency in what they publish.

3. Why this matters

Most “fact-checking” tools stop at scores or flags.

Most “AI summary” tools hallucinate or remove context.

We built something different:

 A transparent pipeline that shows exactly why a conclusion was reached.

No black boxes.

No ideology.

Just structured evidence.

4. Ask for feedback

I’d love to hear from the community:

• Would you use a browser extension like this?

• What output format would you want from an API?

• Any specific features you’d like in a multi-source news analyzer?

Your feedback helps shape the roadmap more than you think.

Thanks for reading, and thanks to everyone who commented on the first post.

Hey Indie Hackers,

I’m Marcell, a PM and co-founder from Budapest working on Neutral News AI – a news site that:

• pulls from multiple left / center / right sources,

• runs bias and propaganda detection deep learning models,

• extracts claims and checks them with an MNLI fact-checker,

• and publishes short, neutral summaries with receipts (quotes, links, timestamps).

Why we are building this

Over the last few years we got fed up with:

• reading 5 articles on the same event and getting 5 different “truths”,

• opinion pieces disguised as reporting,

• doomscrolling Twitter to figure out what actually happened.

I have a background in data / AI and ended up building deep-learning models for political bias and propaganda detection. In tests on public datasets, my hybrid model (transformer + extra features) beat the best public baselines by roughly ~10 percentage points (93%)in accuracy on bias classification.

Neutral News AI is my attempt to turn that research into something normal people can use:

• Reader side: neutral, fact-checked summaries with receipts.

• Power user side: an Analyzer where you paste any article URL and get:

  • political bias,

  • sentiment, subjectivity, readability,

  • a credibility signal,

  • in the near future, we can expect comparisons from news outlets covering the same topic.

  How it works (high level)

For each topic:

1. We crawl multiple outlets across the spectrum for the same story.

2. A generative model drafts a source-bounded summary constrained to those articles.

3. A claim extraction module breaks the draft into atomic claims (events, numbers, quotes).

4. For every claim, an MNLI model checks whether the supporting articles entail, contradict, or are inconclusive about it.

5. We surface the claims + verdicts + citations directly on the article page, then do a final human pass before publishing.

Methodology is fully documented here:

https://neutralnewsai.com/methodology

Where we are at now

• ✅ MVP live: https://neutralnewsai.com

• ✅ Analyzer live: https://neutralnewsai.com/analyzer

• ✅ Baseline models trained and integrated into the pipeline

What we are trying to learn next

1. Positioning: For you as founders, does this feel more like:

   • a consumer product,

   • a media brand,

   • or an API/infra play (bias detection + fact-checking as a service)?

2. Monetization: Our current plan is:

   • keep the news free,

   • add a Pro tier on the Analyzer (more features, bulk analysis, saved reports),

   • plus B2B/API later for publishers or platforms.If you were me, what would you charge for first?

3. Growth: Right now traffic is mostly from SEO + a bit of Reddit. My next big move is a coordinated launch on Hacker News and Product Hunt. What worked for you when launching media/AI products?

Ask to the IH community

• Tear apart the idea and positioning.

• Tell me what would make you trust this kind of product more.

• If you run a newsletter, blog, or product that deals with news, politics, or media literacy, I’d love to chat.

I’ll answer every comment.