Is AI accurate for journalism?
Not on its own. The largest study of its kind found AI assistants misrepresent news content in 45 percent of responses, and the newsrooms that trusted AI drafts published fabricated books, invented authors, and errors in more than half their articles. AI can help you draft, summarize, and research, but nothing it produces is safe to publish under your byline until a human has checked every quote, source, name, and number against the original.
Accurate enough to draft, nowhere near accurate enough to publish
In the largest study of its kind, professional journalists at 22 public-service broadcasters evaluated more than 3,000 answers from ChatGPT, Copilot, Gemini, and Perplexity about the news. 45 percent contained at least one significant issue, 31 percent had serious sourcing problems such as missing or misleading attribution, and 20 percent had major accuracy problems including hallucinated details and outdated information. The result held across 14 languages and 18 countries.
The problem is not that AI writes badly. It writes fluently, in a confident news register, which is exactly what makes the errors hard to catch. A fabricated quote reads like a real one. An invented book title sits comfortably in a reading list. A misattributed source looks like sound reporting. Journalism is one of the worst possible tasks for a model that predicts plausible text, because the whole job is getting specific, checkable facts right, and the model has no way to know which of its plausible outputs are true.
of AI-assistant answers about the news contained at least one significant issue, across ChatGPT, Copilot, Gemini, and Perplexity, in 14 languages and 18 countries.
Why AI fails at journalism specifically
A large language model does not report. It predicts the next most plausible token, and the surface features of journalism are highly patterned: a name, a title, a date, a direct quote in quotation marks, an attribution to a named outlet. The model generates all of these because they fit the pattern, not because it looked them up. That is why AI invents quotes that were never said, attributes real reporting to the wrong publication, and confidently states that a politician still holds an office they left months ago.
The subtler failure is the one that ends careers. The model often starts from a real story and then distorts it: it alters a quote so it says something slightly different, adds a detail that was not in the source, or drops the qualifier that made the original accurate. When the BBC had journalists check AI summaries of its own articles, 13 percent of the quotes attributed to the BBC had been altered from the original or did not appear in the cited piece at all. Verifying that a source exists is not enough, because the source can be real and the claim still fabricated.
When newsrooms trusted it anyway
CNET (2023) — errors in more than half
CNET quietly published dozens of finance explainers written by an AI tool under a house byline. After Futurism found basic mistakes, including an article that badly misstated how compound interest works, CNET audited the set and issued corrections on 41 of 77 articles, more than half, some of them substantial.
Sports Illustrated (2023) — authors who did not exist
Sports Illustrated ran product reviews under writers like Drew Ortiz, whose headshot was on sale on a site selling AI-generated faces and who had no life outside the page. When Futurism asked about them, the fake authors quietly disappeared. The fallout contributed to the publisher losing the Sports Illustrated license.
Chicago Sun-Times (2025) — a reading list of books that do not exist
A syndicated summer section printed by the Chicago Sun-Times and the Philadelphia Inquirer recommended 15 books; 10 of them were invented by AI, including titles falsely attributed to real, award-winning novelists. The writer had used AI for research and never checked the output. The paper apologized and pulled the section.
The numbers behind it
of the time, AI search tools cited the news sources behind their answers incorrectly, across 1,600 queries on eight products, ranging from 37 percent wrong at best to 94 percent at worst.
Jazwinska and Chandrasekar, Tow Center for Digital Journalism, Columbia Journalism Review, 2025
of AI answers about BBC news stories had significant issues, and 19 percent that cited BBC content introduced factual errors in statements, numbers, or dates.
The pattern in every one of these cases is the same: a fluent draft, a quote or a source that looked real, and no independent check before it ran. That is exactly the gap TrueStandard is built to close. Paste the passage, and four to five models check every quote, name, and attribution against each other in about a minute, so the fabrications surface before your byline is on the story, not after the correction.
How to verify AI journalism output
Treat anything an AI gives you as an unverified draft, no matter how clean it reads. Before you publish, cite, or attribute it, run this check.
-
01
Verify every direct quote against the original recording, transcript, or document. Models routinely alter quotes or invent them outright, and an altered quote is a defamation risk, not a typo.
-
02
Confirm every named source, study, or article actually exists and says what the draft claims. Pull the primary source yourself rather than trusting the AI summary of it.
-
03
Check every proper noun, title, and affiliation. AI misstates who holds which office, who wrote what, and which outlet published a story.
-
04
Re-check all numbers, dates, and statistics against the source, including figures the model presents with total confidence.
-
05
Confirm any book, paper, case, or report it recommends or references is real. Fabricated titles attributed to real people are a signature AI failure.
-
06
Watch for distortion, not just invention. A real story summarized with an added detail, a dropped qualifier, or a shifted adjective is still wrong.
-
07
Never publish AI-assisted copy that a human has not fact-checked end to end. Your masthead's standards and your own name are on it regardless of the tool.
How to make AI output reliable: check it across models
The fix is not to hunt for a single more accurate model. Every large language model predicts fluent, plausible text, so each one can be confidently wrong on its own. What changes the odds is agreement. When several independent models are asked the same thing and all land on the same answer, the chance they share the exact same hallucination drops sharply. When they disagree, you have found the precise claim to check by hand before it ships.
That is what TrueStandard does: it runs your draft through four to five frontier models at once and surfaces every disagreement in about a minute, with sources. See the AI fact checker for how the method works, or read why AI cites studies that do not exist for the mechanism behind the failures on this page.
Common questions
Can I use AI for reporting and news writing at all?
For drafting, summarizing your own notes, generating angles, and speeding up structure, yes. As a source of facts, quotes, or attributions, no. Journalists evaluating AI answers about the news found significant issues in 45 percent of them, so every checkable claim has to be verified against the original before it runs.
Is AI reliable for summarizing news articles?
Not reliably enough to publish unchecked. When the BBC had journalists review AI summaries of its own stories, 51 percent had significant issues and 19 percent that cited BBC content introduced factual errors, including altered quotes and wrong dates. Summarization looks safe because the source is right there, which is exactly why the distortions slip through.
Why does AI invent quotes, sources, and even authors?
Because it learned the shape of journalism, not the facts of any particular story. It predicts a plausible quote, a plausible byline, and a plausible attribution the same way it predicts any other text, so a fabricated one looks identical to a real one on the page.
What is the safest way to use AI in a newsroom?
Use it to draft and organize, then verify every fact, quote, name, and source independently before anything publishes. Running the draft through several independent models and checking where they disagree surfaces the fabrications that any single model, however fluent, will state with total confidence.
Do not publish AI output on trust
Paste your draft. Four to five models check every claim in about 60 seconds, and you see exactly where they disagree before your name is on it.