Overview
AR News & Fact Checker turns your phone or laptop camera into a checking companion for the things you read online. Point it at a headline, an article, or a screenshot of a forwarded message; the app runs Vision OCR to lift the text off the image, then asks an AI fact-checker to pull the individual factual claims apart and assess each one. Every claim gets a verdict — True, False, Misleading or Unverified — with a confidence figure, the evidence or reasoning behind it, and a note on what would settle the question. Alongside the claims, the app gives a read on the bias or framing of the piece and an emotional-language score that flags loaded, alarmist or manipulative wording. On the scan view, the overall verdict is drawn as a coloured augmented-reality badge over the camera frame.
You can run a check with no account at all. Everything personal — your saved checks and history — lives behind a free login, and premium removes the free-check limit for a flat monthly price.
A verdict here is an assessment of a claim or its language, produced by an AI model. It is not a legal finding, not a certification, and never a judgement of any person's character. Treat it as a prompt to look closer and to check the primary sources yourself.
Responsible AI — how we frame verdicts
Fact-checking is a sensitive category, so the product is built to under-state rather than over-state. Every result screen carries the same message and the app is designed around a few firm principles.
A media-literacy aid, not an authority
The tool exists to help you slow down and think, not to hand down a ruling. Every verdict screen shows the disclaimer the AI returns — “AI-assisted assessment, not a definitive ruling. Verify against primary sources” — prominently, never as fine print. Nothing in the interface presents a verdict as authoritative or final.
We assess the claim, not the person
A verdict is framed as an assessment of the specific claim or the language used, never a character judgement of whoever wrote or shared it. The app avoids naming real people or outlets in its own examples, and the model is asked to reason about the statement in front of it rather than to accuse anyone.
Unverified is used generously
When a claim can't be settled from what the model reliably knows — because it needs a live primary source, is a matter of opinion, or is genuinely contested — it is marked Unverified in neutral grey, not forced into True or False. Verdicts are hedged and a confidence number is always shown so you can see how sure the assessment is.
Every verdict is correctable
The verdict-detail screen carries a “disagree / report a wrong verdict” affordance so a reader can flag an assessment they think is wrong. AI models make mistakes; the product treats its own output as fallible and invites correction rather than presenting itself as the last word.
What it does
1. Scan & Vision OCR
Point the camera at a headline, a paragraph or a screenshot and capture it, or upload an image. A vision model performs OCR to extract the text, so you can check something you've seen on screen without retyping it. The extracted text is shown back to you so you can confirm the app read it correctly before trusting the assessment.
2. Claim-by-claim verdicts
Rather than a single thumbs-up or thumbs-down, the AI breaks the text into its separate factual claims and assesses each one on its own. Every claim card shows the verdict, a confidence figure, the evidence or reasoning, and what would be needed to confirm it. An overall verdict summarises the piece, but the detail is where the real value sits.
3. Bias & emotional-language read
Beyond whether the facts hold up, the app reads how the piece is written: a bias/framing note on the leaning it detects, and an emotional-language score (0–100) that highlights loaded, alarmist or manipulative wording with the actual phrases it picked up. High emotion with thin evidence is a classic signal to slow down.
4. AR verdict badge
On the scan view, the overall verdict appears as a large coloured badge drawn over the camera or the frozen frame — emerald for True, red for False, amber for Misleading, grey for Unverified — with the confidence and the OCR'd text underneath. A quick, glanceable read, with the full breakdown one tap away.
Plus: history & analytics
Signed-in readers get a history of every check they've run, each with its overall verdict, bias read and date. The analytics view goes deeper: how many checks you've run, how many came back false or misleading, your average emotion score, a checks-over-30-days line, and a verdict-distribution doughnut so you can see the mix of True, False, Misleading and Unverified across everything you've checked.
Reading a verdict
Each claim lands in one of four buckets. The colour is a signal, not a sentence — read the reasoning and the confidence alongside it.
| Verdict | What it means | What to do |
|---|---|---|
| TRUE | The claim is well supported by the model's knowledge and reasoning. | Still worth a glance at a primary source for anything important. |
| MISLEADING | Technically part-true but framed, cropped or exaggerated so it gives the wrong impression. | Read the reasoning — the missing context is the point. |
| FALSE | The claim contradicts what is reliably known. | Don't share it; confirm with a primary source before correcting others. |
| UNVERIFIED | Can't be settled from what the model reliably knows — needs a live source, is opinion, or is contested. | Go to the primary source. The tool is telling you it doesn't know. |
The confidence figure (0–100) shows how sure the assessment is; a True at 55 is a much weaker signal than a True at 95. When in doubt, the app leans towards Unverified rather than guessing.
How it works
Point the camera at what you're reading and capture a frame, upload an image, or paste the text directly on the paste view.
Vision OCR lifts the text off the image and shows it back to you so you can confirm it was read correctly.
The AI splits the text into claims and assesses each — verdict, confidence, evidence — plus a bias read and an emotional-language score.
Read the breakdown, note the confidence, and check the primary sources yourself before you act or share. Save it to your history if you like.
Trending fakes
A curated feed of common misleading or debunked claim patterns — the kinds of forwards that go around a lot — each with a verdict, a category and a short explanation of why it doesn't hold up. It's there to build pattern recognition, so you spot the shape of a bad claim faster.
Every example is generic and non-defamatory. The feed deliberately contains no real named people and no real outlets — only illustrative, educational patterns such as “a viral post claims X cures Y” or “a forwarded message claims Z”.
Plans & pricing
- Scan, OCR and check — no account needed
- A set number of checks to start with
- Claim-by-claim verdicts, bias & emotion read
- Trending-fakes feed and the AR verdict badge
- Checks without the free-tier limit
- Saved fact-check history
- Full personal analytics
- All scan, OCR, bias and emotion features
Billed monthly. Cancel any time.
Privacy & data
Anonymous to start. You can scan, paste and check without an account; anonymous checks are counted locally against a free limit and aren't tied to a personal profile.
Scanned images. When you capture or upload an image to check, the frame is sent to the AI vision model to read the text and assess it. Only checks you explicitly save are stored against your account.
What we store. For saved checks we keep the input snippet, the extracted text, the verdicts and the language read — so you can revisit your history — not the raw camera feed.
Accounts & deletion. Registered readers can delete their account from within the app and via a public deletion request page.
Payments. Premium payments are processed by the payment gateway; card details are never stored by the app.
Not a definitive ruling. Verdicts, bias reads and emotion scores are AI-assisted assessments of claims and language, not certifications or legal findings. Verify against primary sources.
Data model
Core tables (prefix AFC_). Every table carries soft-delete and audit columns.
| Table | Holds |
|---|---|
AFC_Users | Registered reader accounts, with premium status and expiry. |
AFC_FactChecks | One row per check: input type, input snippet, OCR'd text, overall verdict, bias leaning, emotion score and the full result (JSON). The user field is empty for anonymous checks. |
AFC_Verdicts | The individual claim verdicts within a check: the claim text, verdict, confidence, evidence and reasoning. |
AFC_TrendingClaims | The curated trending-fakes feed: generic claim pattern, verdict, category, blurb and a source note. No real people or outlets. |
AFC_Subscriptions | Premium subscriptions: plan, amount, dates and payment status. |
Identity uses the shared account kit scoped to product AFC. Every activity table stores the kit user id in its user field, or leaves it empty for anonymous checks. The analytics view reads AFC_usp_GetAnalytics, which returns headline counts, checks per day for the last 30 days, and the verdict distribution — all scoped to the signed-in reader.
Future roadmap
Live source lookups
Let a check reach out to a set of primary reference sources at assessment time, so more claims can move out of Unverified with a citation attached.
More languages
OCR and claim assessment for more Indian languages, so a forwarded message can be checked in the language it arrived in.
Share-sheet check
A share target so you can send a link or screenshot straight from another app into a check, without switching over manually.
Community reports
Turn the “disagree / report” signals into a moderated queue so wrong verdicts can be reviewed and the trending feed can be improved.
Media-literacy lessons
Short, optional lessons that turn common manipulation patterns from the trending feed into bite-sized skills for spotting them yourself.