Methodology
How we
weigh the words.
A good fact check shows its work. Here's exactly how Groundtruth turns a wall of text into verdicts, scores and next steps, and where its limits are.
The verdict scale
Every checkable claim gets one of seven ratings. The article as a whole gets an overall verdict based on its most consequential claims, not on a simple average.
The claim matches well-established evidence, and nothing significant is left out.
Substantially correct, but needs clarification or extra context, or has minor errors in the details.
Partly right and partly wrong, or it mixes solid facts with unsupported conclusions.
Contains a grain of truth but leaves a false impression through cherry-picking, omission or distortion.
Contradicted by the available evidence, or fabricated.
Can't be confirmed or refuted with reliable knowledge, often because it's too recent, too local or too vague.
A value judgment or prediction. We label it rather than rate it true or false.
The credibility score
The 0–100 score sums up how far a reader can trust the piece as written. It weighs the accuracy of the key claims, the quality of the sourcing, and how much the framing distorts the facts.
Solid claims, credible sourcing.
Minor gaps or framing issues.
Verify before you share.
Significant distortion.
Largely untrue or fabricated.
Signals beyond the facts
A story can be technically accurate and still mislead. Alongside the claims, every report reads how the story is told.
Measured or alarmist?
Is the writing neutral, emotional, sensational or promotional? Heated tone is often a sign of persuasion rather than reporting.
Who says so?
Named experts, documents and data score higher. "Insiders," "many believe" and unnamed researchers score lower.
Words that do the arguing
We highlight emotionally charged phrases right in the text, so you can see where wording replaces evidence.
Applied evenly
For political coverage only, we estimate which way the framing tilts. The same standard applies whatever the direction.
Know the limits
Groundtruth is a powerful first pass, not an oracle. Please keep these in mind:
- No live internet access. Analysis draws on established knowledge up to a training cutoff, so breaking news is often rated Unverifiable, and that's by design.
- AI can be wrong. Models can misread context or be confidently mistaken. That's why each claim carries a confidence level and a verification step.
- We analyze text only. Images, video and audio aren't examined. Paste the article's words rather than a link.
- Not professional advice. Reports aren't legal, medical or financial guidance. For decisions that matter, go to the primary source.
Questions
Can I check a link instead of pasting text?
Not yet. Websites block outside tools from reading their pages directly, so copy the article text (headline and body) and paste it into the analyzer. It takes a few seconds and gives the most accurate results.
What goes on the public Ledger?
Only checks you choose to share. The "Add result to the public Ledger" box is ticked by default and you can untick it before running a check. Public entries store the headline, outlet, verdict, score, summary, claim ratings and a short excerpt.
Is Groundtruth politically biased?
The analyzer applies the same rules to every piece, whatever its political direction, and labels opinions instead of judging them. If you think a report got something wrong, check the primary sources it lists. That's what they're there for.
Why did a claim come back "Unverifiable"?
Usually because it's about very recent events, is too vague to test, or depends on information that isn't publicly established. We'd rather say "we can't confirm this" than guess.
How long does a check take?
Most finish in 10 to 30 seconds, depending on article length. Very long pieces are trimmed to about 40,000 characters.