About AI Truth Bench
AI Truth Bench answers one question about each AI tool it covers: can you trust what it tells you? People use ChatGPT, Gemini, Perplexity and other assistants for work, study and research, and a confident wrong answer can cost a grade, a deadline or a reputation. This site gathers the published evidence on how often each tool gets things right, puts a date and a source on every figure, and gives a plain verdict.
What we do
We cover four areas:
- Accuracy by tool. How often each major assistant gets facts right, where it fails, and how to check its answers. Start at is AI accurate.
- Hallucinations. What they are, why they happen, real cases, and published rates by model. Start at AI hallucinations, explained.
- AI detectors. How accurate detectors such as Turnitin, GPTZero, ZeroGPT and Grammarly are, and what a score can prove. Start at are AI detectors accurate.
- Fact-checking. A short routine for checking any AI answer. See how to fact-check an AI answer.
Each tool page opens with a one-word verdict (Reliable, Mostly reliable, Mixed or Unreliable) and one sentence explaining it, then shows the results behind that verdict.
Who it is for
The site is written for people who rely on AI tools and want to know when to double-check them: students wondering whether a detector score or an AI citation means anything, teachers weighing a flagged essay, and analysts, writers and researchers deciding which assistant to trust with factual work. Pages lead with the verdict and keep the language plain, so you can get the answer mid-task and read the evidence when you need it.
Who writes it
Pages are researched and written by the AI Truth Bench editors, a small team that reads the studies, benchmarks and vendor pages behind each figure. Every page carries the byline "The AI Truth Bench editors" and the date it was last updated.
How we source figures
We collect published results rather than running tests of our own. Every figure on the site comes from a dated, linked source: a peer-reviewed or preprint study, a public benchmark or leaderboard, a news investigation, a court record, or the vendor's own stated figure. Vendor figures are always labelled as vendor claims. Figures from different tests are never mixed into one ranking, because each test measures something different. The full rules are on our editorial standards page, and the method behind each verdict is on how we score AI accuracy.
How to reach us
Email the editors at editors@aitruthbench.com. Send corrections, new test results, or questions about a figure. Please include the page address and a link to your source. The contact page has more detail on what helps us check a report quickly.
Helpful to include
- The address of the page
- The figure or sentence in question
- A link to the source that shows the right figure