AI accuracy rankings

Claude tops our AI accuracy rankings: it repeated false news claims 10% of the time in NewsGuard's August 2025 audit, the lowest of 10 chatbots. Gemini ranks second and Perplexity third. For checking AI-written text, the Originality.ai Fact and AI Hallucination Checker is our top fact checker.

The rankings

Each ranking below compares tools within one published test at a time, gives every figure with its date and source, and names a top pick.

The most accurate AI assistants

Top pick: Claude. Nine assistants ranked by published accuracy and hallucination results: Claude, Gemini, Perplexity, Microsoft Copilot, ChatGPT, Google AI Overviews, Meta AI, DeepSeek and Grok.

Rank	Assistant	Key result (test)	Verdict
1	Claude	10% false news claims repeated, lowest of 10 (NewsGuard, 2025-09-04)	Mostly reliable
2	Gemini	16.67% false news claims repeated, second of 10 (NewsGuard, 2025-09-04)	Mixed
3	Perplexity	37% wrong source lookups, best of eight (Tow Center, 2025-03-06)	Mixed

At the bottom, Grok 3 named the wrong source for 94% of news excerpts in the Tow Center test, the worst of eight AI search tools.

The best AI fact checkers

Top pick: Originality.ai Fact and AI Hallucination Checker. Seven tools ranked on what they check, what they cover and what they cost, with prices checked on 2026-10-29.

Rank	Tool	Best for	Price (checked 2026-10-29)
1	Originality.ai Fact and AI Hallucination Checker	Checking AI-written text claim by claim	100 free one-time credits; Pro $14.95/month
2	Facticity.AI	Claims in TikTok, X or YouTube videos	Free to start
3	Google Fact Check Explorer	Finding fact checks already published	No price listed; open web tool

AI hallucination rates by model

The data behind the rankings: how often leading models add unsupported content or give wrong answers on public leaderboards and studies. On the Vectara leaderboard, even the best model added unsupported content to 1.8% of summaries as of 2026-10-22.

Are AI detectors accurate?

AI text detectors and their published accuracy and false positive rates. One result sets the tone: seven detectors flagged 61.22% of non-native English essays as AI on average in Liang et al. (2023).

How the rankings work

Every ranking on AI Truth Bench follows the same rules.

Published results only. Each figure comes from an independent study, a news audit, a public benchmark or a vendor's own page, and carries its test name, date and source link.
One test at a time. Tests measure different things. The EBU/BBC study judged news answers, the Tow Center test checked source lookups, NewsGuard checked false claims, and Vectara checked summaries. A 10% in one test is never set against a 37% in another.
Vendor claims are labelled. OpenAI, xAI, Perplexity and others publish their own figures. These appear on tool pages marked as the vendor's claim, and they do not move a tool in a ranking.
Models are named. Leaderboards such as Vectara and AA-Omniscience test the models behind the apps, such as GPT-5.5 or Claude Opus 5.5. Pages say "the model behind" when they cite them.
Every pick has a reason. Each ranked tool gets the same fields: key result, what it does best, where it slips, and a "Choose it if" line.

The full method, with the verdict words Reliable, Mostly reliable, Mixed and Unreliable, is on how we score AI accuracy. Each assistant's own evidence is on its page in Is AI accurate?

A ranking tells you which tool is wrong least often; it does not tell you which answer is right.

Questions

Short answers, with links to the dated evidence.

Which AI is the most accurate?

Claude is the most accurate AI assistant in our ranking. It repeated false news claims 10% of the time in NewsGuard's August 2025 audit, the lowest of 10 chatbots, and Claude Opus 5.5, a model behind Claude, scores 48 on the AA-Omniscience index, second highest when checked on 2026-10-29. Gemini ranks second and Perplexity third.

Which AI is the least accurate?

Grok ranks last. In the Tow Center's March 2025 test, Grok 3 named the wrong source for 94% of news excerpts, and 154 of its 200 citations led to error pages. DeepSeek ranks just above it, after failing 83% of news prompts in NewsGuard's January 2025 audit. Both carry an Unreliable verdict.

How are the rankings decided?

Tools are compared within one published test at a time, never by mixing numbers from different tests. Tests of the consumer apps count most, and repeating false claims weighs heavily. Model leaderboards break ties. Vendor figures are shown as claims and do not change a tool's place. Every figure carries its date and source.

What is the best AI fact checker?

The Originality.ai Fact and AI Hallucination Checker ranks first for most people. It flags inaccurate statements and likely hallucinations in pasted text and offers 100 free one-time credits, with Pro at $14.95 a month as checked on 2026-10-29. Facticity.AI ranks second and suits claims made in social video.

How often do the rankings change?

A ranking changes when a new published result changes a tool's place within a test. Live leaderboards such as Vectara and AA-Omniscience move often, while large studies like the EBU/BBC report appear less often. Each figure shows its test date, so you can judge how recent the evidence is for every tool.

Where to go next on the bench.