Is Meta AI accurate?
Meta AI repeated false news claims 40% of the time in NewsGuard's August 2025 audit, tied with ChatGPT, while the Llama models behind it score well on summaries.
Short answer
Meta AI is about as accurate as ChatGPT on news, which means it is wrong often enough to need checking. In NewsGuard's August 2025 audit, Meta AI repeated false news claims 40% of the time. Meta itself has admitted hallucinations about breaking news.
Results
- Benchmark
- News audit
- Vendor claim
| Test | What was measured | Result | Date | Source |
|---|---|---|---|---|
| NewsGuard AI False Claims Monitor | Rate of repeating provably false news claims | 40% | Source 1News audit | |
| Incident statementVendor's own figure, not independently checked | Meta's account of hallucinations about a breaking news event | Some answers said the July 2024 Trump assassination attempt did not happen | Source 2Vendor claim | |
| Vectara Hallucination Leaderboard | Summaries by Llama 4 Maverick, a model behind Meta AI, that added unsupported content | 8.2% | Source 3Benchmark |
What these results mean
Meta AI is the assistant built into WhatsApp, Instagram and Facebook, so many people use it without opening a separate app. That makes its news answers matter.
NewsGuard's August 2025 audit fed 10 leading chatbots prompts about provably false news stories. Meta AI repeated the false claim 40% of the time, tied with ChatGPT and above the 35% average (NewsGuard, 2025-09-04).
Meta has described the problem itself. In July 2024, some Meta AI answers asserted that the attempted assassination of Donald Trump had not happened. Meta called these hallucinations, said they are an industry-wide issue, and said it was fixing them (Meta, 2024-07-30). Meta gave no error rate.
The Llama models behind Meta AI do better on a narrower task. On the Vectara leaderboard, Llama 4 Maverick added unsupported content to 8.2% of summaries, Llama 4 Scout to 7.7% and Llama 3.3 70B to 4.1%. Those results are for the underlying models, not the Meta AI app.
Breaking news is where Meta AI has failed most visibly, by Meta's own account.
Where Meta AI goes wrong
False news stories
It repeated provably false news claims in 40% of NewsGuard's prompts.
Breaking news
Meta acknowledged answers that denied a major news event had happened. Fast-moving stories are where a model's training data is most out of date.
Answers inside chats
Because Meta AI answers inside messaging and social apps, its replies are easy to forward without a source attached.
Summaries
Llama 4 Maverick still added unsupported content to 8.2% of summaries on Vectara.
How it compares
Each table uses one test only.
NewsGuard AI False Claims Monitor (2025-09-04): rate of repeating false news claims
| Chatbot | Result |
|---|---|
 Claude | 10% |
 Gemini | 16.67% |
 Microsoft Copilot | 36.67% |
 ChatGPT | 40% |
| Meta AIThis page | 40% |
 Perplexity | 46.67% |
Source: NewsGuard. Meta AI tied with ChatGPT, well behind Claude and Gemini.
Vectara Hallucination Leaderboard (2026-10-22): summaries with unsupported content
| Model | Result |
|---|---|
| Llama 3.3 70B (model behind Meta AI) | 4.1% |
| Llama 4 Scout (model behind Meta AI) | 7.7% |
| Llama 4 Maverick (model behind Meta AI) | 8.2% |
| GPT-5.5 (model behind ChatGPT) | 9.3% |
| Claude Opus 5.5 (model behind Claude) | 9.9% |
Source: Vectara. The Llama models summarize more faithfully than the GPT-5.5 and Claude Opus 5.5 models on this test. The full table is on AI hallucination rates by model, and the overall order is on the most accurate AI assistants.
How to check Meta AI's answers
Ask for a source
If Meta AI gives none, treat the answer as unconfirmed.
Check breaking news with a named publisher
Wait until a reliable outlet or official source reports it.
Do not forward an answer you have not checked
In a group chat, an AI answer can spread as fact.
Search the claim plus "fact check"
Many false stories already have a published debunk.
Confirm names, dates and numbers in the original record
before you rely on them.
The full routine is on how to fact-check an AI answer. How sources are weighed is on how we score AI accuracy.
Sources
- News audit
- Vendor claim
- Benchmark
Questions
Is Meta AI always right?
No. Meta AI repeated false news claims 40% of the time in NewsGuard's August 2025 audit, tied with ChatGPT and above the 35% average of 10 chatbots. Meta also admitted in July 2024 that some answers wrongly said the Trump assassination attempt had not happened. Check any fact before you rely on it or share it.
Is Meta AI reliable for research?
Meta AI is fine for quick questions, not for research you will cite. Its news record matches ChatGPT's in NewsGuard's audit, at 40% false claims repeated, and Meta has acknowledged hallucinations about breaking news. Use it to get started, then find and cite the original sources yourself.
Does Meta AI make up facts?
Yes, it can. Meta itself called some July 2024 answers hallucinations after they denied a real news event. On the Vectara leaderboard of 2026-10-22, Llama 4 Maverick, a model behind Meta AI, added unsupported content to 8.2% of summaries. Ask for a source and check it before you trust a claim.
Is Meta AI in WhatsApp accurate?
Treat it with the same care as any Meta AI answer. NewsGuard's audit found Meta AI repeated false news claims 40% of the time, and the same checks apply in WhatsApp, Instagram and Facebook. The chat setting does add one risk: answers are easy to forward without a source. Treat a Meta AI reply in a group chat as a starting point, and check it before you pass it on.
How do I check a Meta AI answer?
Ask Meta AI for its source and open it. For breaking news, wait for a named publisher or official statement. Search the claim with the words "fact check" to find existing debunks. Confirm names, dates and numbers in the original record. Do not forward an answer to a group until you have checked it.
