FAXTR is an aggregator, not an arbiter. We do not declare a claim true or false on our own authority. Instead, we collect verdicts from established fact-checking organizations — most of them signatories of the IFCN Code of Principles — and present them in one place, in your language, with a direct link back to the original review.
This page documents the full pipeline: the organizations we cite, how we standardize their verdict labels, how our AI features are constrained, and how we correct mistakes. If you find something here that looks wrong, the corrections section explains exactly how to flag it.
Five principles we operate by
Non-partisanship and fairness
We do not pick claims for political effect. Topic selection is driven by what viral or trending claims have available primary-source verification — not by ideology, party, or country of origin.
Transparency of sources
Every verdict shown on FAXTR links back to the originating fact-check organization's full review. Readers can — and should — read the source before sharing.
Transparency of funding
FAXTR's funding and ownership are disclosed on /about. We do not accept payment to elevate or suppress particular verdicts.
Transparency of methodology
This document is the methodology. It is updated when our pipeline changes; previous versions stay in the git history of our public sitemap generator.
Open and honest corrections
When we make an error — a mislabeled verdict, a broken source link, a translation mistake — we fix it, mark the page as corrected, and acknowledge the change. The corrections email is on /contact.
The fact-checking organizations we aggregate
FAXTR's verdict pipeline draws from the Google Fact Check Tools index, which surfaces ClaimReview structured data published by more than 100 accredited organizations. The table below lists the most frequently surfaced sources; the full set is broader and rotates as new IFCN signatories publish.
| Organization | Region | Focus |
|---|---|---|
| IFCN (Poynter) | Global | Code of Principles, signatory verification |
| AFP Fact Check | Global / FR | Multi-language verifications across 80+ countries |
| Reuters Fact Check | Global / UK | Wire-service verification of viral claims |
| Snopes | USA | Long-form urban-legend and political fact-checks |
| PolitiFact | USA | Statement-level Truth-O-Meter ratings |
| FactCheck.org | USA | Annenberg Public Policy Center, political claims |
| Lead Stories | USA | Viral hoax and meme verification |
| Logically Facts | UK / India | AI-assisted misinformation tracking |
| Full Fact | UK | UK political and statistical claims |
| Maldita.es | Spain | Spanish-language disinformation, WhatsApp tip line |
| Newtral | Spain | Political verifications and methodology research |
| Correctiv | Germany | Investigative non-profit, EU funded research |
| dpa-Faktencheck | Germany | German wire-service fact-checking unit |
| Les Décodeurs (Le Monde) | France | Le Monde's verification desk |
| CheckNews (Libération) | France | User-question-driven French fact-checking |
| Aos Fatos | Brazil | Brazilian political and viral content checks |
| Boom Live | India | Multilingual WhatsApp-era fact-checking |
| AAP FactCheck | Australia | Australian Associated Press wire verifications |
How we standardize verdicts
Each organization writes its own verdict text — 'False', 'Pants on Fire', 'Mostly False', 'Sin contexto', 'Faux', 'Falsch', and so on. FAXTR maps that text to a small set of standardized labels so a user searching in any language can compare verdicts across organizations.
The original organization rated the claim as accurate. Source language often: true, correct, accurate, verdadero, vrai, richtig.
Partially true or missing critical context. Source language: half, partly, mostly, mixed, misleading, out of context.
The original organization rated the claim as inaccurate. Source language: false, fake, fabricated, pants on fire, falso, faux, falsch.
Organizations disagree, or the rating text does not map cleanly. We never invent a verdict — DISPUTED means 'read the original sources and decide'.
No accredited organization has yet published a ClaimReview for this claim. We never fill the gap with an AI guess.
Where AI is used — and where it is not
FAXTR uses large language models for narrow, supervised tasks. They never write the verdict. They never invent a fact. Guardrails are enforced at the prompt and post-processing layer.
- Claim normalization. We use an LLM to rephrase a long, messy user query into a shorter searchable claim. The user always sees both the original and the normalized version.
- Translation of verdicts. When a verdict is published in a language different from the user's, an LLM produces a translation. The original-language text is always shown alongside.
- Verdict mapping safety net. If the source verdict text does not match our keyword rules, an LLM is allowed to suggest the closest label, but only DISPUTED is permitted as a fallback when confidence is low.
- Hard rules: no AI-generated original verdicts, no AI-invented sources, no AI-invented quotes. If the pipeline cannot find a real ClaimReview, the result is UNVERIFIED. Full stop.
- Logged and auditable. Every AI step in the pipeline is logged with input, output, and model version. Errors caught downstream feed back into the prompt library.
Corrections policy
If you spot a verdict that contradicts the linked source, a broken link, a mistranslation, or a verdict that should be re-evaluated because the underlying claim has been updated, write to corrections at faxtr.com or use the /contact page. We acknowledge corrections requests within 72 hours and post a visible note on any page that is changed. We never silently rewrite history.
Frequently asked
Is FAXTR an IFCN signatory?+
FAXTR commits publicly to the IFCN Code of Principles in this methodology. As an aggregator that does not produce original fact-checks, we are not a Code signatory ourselves; the organizations we cite are. The Code's 5 principles still bind how we behave.
Do you fact-check claims yourselves?+
No. FAXTR's role is to find, translate, and present what existing accredited fact-checkers have already published. When no accredited fact-check exists, we label the result UNVERIFIED rather than guess.
Why do verdicts sometimes disagree?+
Organizations evaluate claims using slightly different evidentiary standards and at different times. When two reputable checkers split, FAXTR shows both and labels the result DISPUTED. Reading the disagreement is part of the verification work.
How often is the index updated?+
We refresh the ClaimReview index three times a day. Real-time fact-checking is impossible — the average lag between a viral claim and the first published verdict is roughly 8 to 36 hours.
Can I use FAXTR results in my own reporting?+
Yes — please cite the original fact-checker, not FAXTR. We are a navigation layer; the authoritative source is the organization that published the ClaimReview.
What about claims in low-coverage languages?+
When local coverage is thin, FAXTR will surface adjacent-language verdicts (e.g., an AFP English verdict for a Portuguese claim). The original language and our translation are always shown side by side.
Try it: check a claim in seconds
FAXTR searches 100+ fact-checking organizations in a single query — free, no login required.