Key Highlights
- AI is Commonly Used in Editorial Workflows: From January to June 2025, five major newsrooms published almost 80,000 articles, and 48% of those articles were generated or assisted by AI tools.
- CoinDesk Achieves the Fastest AI Adoption: CoinDesk recorded the fastest rise in AI-generated reporting in this period of time, cutting its human-to-AI article ratio from 244:1 in January to 2:1 by June.
- Transparency Improves Reader Trust: All five newsrooms clearly label AI-written stories, and a University of Zurich study shows this transparency raises reader engagement by up to 28%.
- Over Half of Investing.com’s Content Comes from AI: The outlet used AI for more than 50% of its articles in the first half of 2025, peaking at 65.36% in March.
- The Defiant’s AI Cutback: The Defiant reached nearly 68% AI-written content in May but dropped to 32.7% by June, likely aiming for a better balance between AI and human content.
Artificial intelligence no longer sits at the edge of experimentation. It has entered the core of modern infrastructure. In industries where speed, volume, and consistency define success, teams now treat AI as a foundational tool.
Technology leaders such as Google and Microsoft have already integrated AI deep into their workflows. In software development alone, ~50% of Google’s code and ~30% of Microsoft’s is now machine-generated.
The shift in journalism follows the same logic. Newsrooms operate under constant pressure to move fast, publish often and cover stories that matter. Timeliness matters as much as accuracy, and AI offers a way to meet those demands at scale. Generating first drafts, rewriting press releases, and formatting daily coverage no longer requires manual labor alone. As large language models improve, they’ve become a practical extension of editorial workflows.
Financial journalism, particularly in crypto, places even greater demands on speed and scale. Assets are global and trade 24/7, without breaks between time zones or market sessions. Critical developments such as price swings, protocol upgrades, and policy changes can unfold at any moment. Traders and investors rely on real-time reporting to make sense of volatility, regulation, and sentiment. To keep pace with that non-stop cycle, some crypto newsrooms have begun using AI to support content production and maintain consistency across high-volume coverage.
The Purpose of This Report
AI disclaimers are now appearing with increasing frequency at the bottom of crypto news stories. Phrases such as:
- “This article was generated with the support of AI and reviewed by an editor. For more information, see our T&C.”
- “Parts of this article were generated with the assistance of AI tools and reviewed by our editorial team to ensure accuracy and adherence to our standards.”
- “This is an AI-generated article powered by xx, curated by…”
This has become a common marker of how automation is entering editorial workflows.
As a crypto PR agency that delivers company announcements, news stories, and expert commentary to reporters and editorial teams every day, the shift of delegating part of what was once an entirely editorial responsibility to AI has a direct impact on our work.
We set out to identify which crypto news outlets use AI and disclose it publicly, along with the frequency of AI-generated reporting, output volumes, and other related trends.
Our Methodology
This report examines ~80,000 articles published by five newsrooms: Investing.com, The Defiant, Benzinga, CoinDesk, and Bitcoin News. The selection of these newsrooms was based exclusively on their explicit acknowledgment of AI involvement through clearly identifiable disclosures, such as bylines, author fields, or disclaimers included in articles. Newsrooms without explicit disclosures were excluded from this analysis to maintain strict transparency standards.
The analysis covers the period from January 1 to June 30, 2025. Articles classified as “AI-generated” were those explicitly marked by newsrooms, regardless of the extent or specific nature of AI involvement. This means an article was counted as AI-generated even if AI played a minimal role, such as data collection, initial draft generation, summarization, or other auxiliary editorial tasks. Conversely, any article without clear and explicit AI attribution was classified as fully human-written.
We recognize that this approach may group articles that use AI in significantly different ways. At the same time it’s impossible to determine how much or how little AI contributed to each labeled article. One newsroom may label articles where AI contributed only a paragraph or brief summary, while another may reserve labels strictly for fully AI-generated content. So, comparing these outlets side by side may exaggerate or understate actual AI usage. The data reflects only what is publicly disclosed, without insight into the scale or nature of AI input behind each label.
To address these variations, we contacted each newsroom mentioned in the report, as well as others not included in the dataset, to better understand how they approach AI attribution and editorial integration. Their responses appear later in this report and offer additional context around disclosure practices, internal policies, and attitudes toward AI in journalism.
Additionally, to support transparency and independent verification, we have made the full (raw) dataset and analysis available in this Google Sheets file.
The report does not speculate about internal tools or behind-the-scenes automation. It focuses solely on what readers can verify directly. The goal is to measure visible AI adoption and track how often each outlet relied on machine-generated content across the first half of the year.
Disclaimer
This report examines publicly available information on the use of AI in crypto media. Our aim is to understand the scale of this trend and its influence on editorial practices.
The media outlets featured in this report produce valuable work, and we encourage readers to follow their coverage. Our goal is not to discredit or criticize them; we respect their role in shaping the crypto narrative and value their contributions to the industry.
What Researchers Have Found About AI’s Role in Editorial Workflows
Surveys show an overwhelming majority of journalists and news organizations are now using AI for various editorial tasks. For instance, a Reuters Institute study in late 2024 found that 87% of newsrooms reported being fully or at least somewhat transformed by generative AI, up from 73% the year before.
In practice, 96% of publishers are prioritizing AI for back-end production tasks such as transcribing interviews, tagging content, or copyediting, with the goal of streamlining output. Around 80% are using AI for personalization and content recommendations. Another 77% are applying it to content creation, including the generation of simple news briefs or headlines. And roughly 73% of newsrooms also rely on AI for newsgathering support, particularly for tasks like verification, data mining, or investigative research.
The reasons behind that adoption are mostly practical.
News organizations turn to AI to manage pressure around speed, scale, and limited staff. Many see it as a tool for keeping up with non-stop publishing cycles.
The Tow Center’s report shows that AI helps reduce friction in production. Editors use it to automate mundane tasks that slow down reporting but add little editorial value.
Journalists still lead the process. AI handles the background work so reporters can focus on writing, verifying, and interviewing. The goal isn’t to replace the newsroom but rather to make it faster and more efficient.
However, AI adoption comes with problems. Some models still invent facts and present them with confidence. Without a second layer of review, those errors can slip into final copy and damage credibility.
Newsrooms can’t afford that risk. Editors need to check every line, verify each claim, and stay alert to hallucinated content.
Out of the top 15 models ranked by hallucination, error rates range from 0.7% to 1.7%, meaning factual accuracy is above 98%. But that doesn’t change the fact that even one confidently stated falsehood can derail a story and erode trust. For journalism, 98% isn’t good enough.
Labeling matters just as much. Readers should know when a story involves machine input. A 2024 study from the University of Zurich showed that labeling increased reader engagement. When articles carried an AI-generated or AI-assisted tag, willingness to keep reading went up by 24–28% compared to stories marked as human-written.
The crypto newsrooms in our sample have already moved in that direction. Investing.com, The Defiant, Benzinga, CoinDesk, and Bitcoin News each apply visible bylines or author tags when AI plays a role in the writing. Human editors still read through the drafts before anything goes live. The specifics differ across outlets, but all five follow internal rules that limit where AI fits into the workflow. The process isn’t uniform, but the principle is the same: no machine-written text gets published without someone taking a second look.
Having outlined the purpose of this report, detailed our methodology, addressed disclaimers, and reviewed broader research on AI in journalism, we now turn to our findings on AI adoption across crypto newsrooms.
AI Adoption Among Crypto Newsrooms
From January to June 2025, crypto newsrooms took different paths with AI. Some leaned heavily on it from the start. Others were just beginning to bring it into the mix.
Investing.com published the highest share of AI-written content. About 54.8% of its articles during the period were attributed to machine authorship. The Defiant came in close behind at 48.8%.
Trailing behind, Benzinga integrated such tools into 18.3% of its output, with CoinDesk and Bitcoin News being the most restrained at 13.4% and 9.6%, respectively.
On a month-to-month basis, the data reveals an upward trend in AI usage across all five crypto newsrooms from January to June. Machine-generated content captured a larger share of output by mid-year than at the start, which indicates an industry shift toward automation.
Investing.com and The Defiant reached the highest levels of AI use. Starting in February, Investing.com kept its AI use above 50% through June, with a peak of 65.36% in March. The Defiant also showed steady growth and reached a peak of 67.77% in May. Although by the end of the analysis period, it had dropped by nearly half to 32.71%, perhaps due to a rebalance in its AI and human content mix.
Benzinga and Bitcoin News had the least volatility and the lowest AI adoption. Benzinga stayed under 25% throughout the period, moving only between 15% and 25%. Bitcoin News remained below 15% from January to June. They maintained this approach likely due to a preference for human content and slower adoption of automation.
CoinDesk was the only outlet that began testing the waters with AI just recently this year, so the growth is much more pronounced in relative terms. It started at 0.41% in January and reached 35.62% by June, overtaking The Defiant. While other outlets built on existing AI use, CoinDesk introduced it from a near-zero base and scaled quickly in the following months.
Although the absolute adoption percentages reveal the current scale of AI integration, the pace of change is equally important. By average monthly growth rate, CoinDesk was the fastest adopter at 275.44%, followed by Investing.com at 84.55%. The remaining three – Bitcoin News, The Defiant, and Benzinga – were more or less in the same range, each growing at around 19% per month.
Looking at the median helps explain how averages can be skewed by sharp jumps or drops. Even then, the overall pattern holds. CoinDesk is still number one, while Benzinga and Bitcoin News remain the outlets that are more focused on human-written content.
What’s interesting is that, based on median growth, The Defiant moves into second place. This happens because its monthly increases were steady and consistent, without large spikes or drops, unlike Investing.com.
Another way to understand how AI fits into newsroom operations is to compare the volume of human-written content to machine-generated output. The same pattern shows up here as well, where a ratio reflects how many human-written articles are published for every AI-written article.
CoinDesk moved the fastest toward AI, going from 244 human articles per 1 AI article in January to just 1.81 by June. In simpler terms, if CoinDesk produced 245 articles in January, 244 were written by humans and only 1 by AI. By June, out of the same 245 articles, about 158 were human-written and 87 were AI-written.
Investing.com was next, shifting from 1.96 to 0.99 and reaching a near balance between human and AI content. Bitcoin News followed, lowering its ratio from 17.63 to 9.64 over the same period.
The Defiant and Benzinga remained more focused on human-written content, though for different reasons. The Defiant’s ratio changed only slightly, from 2.59 to 2.06, largely because it reduced its AI output in the later months. If that drop is taken out of the picture, Benzinga appears to be the most human-led newsroom overall. Its ratio moved from 5.67 to 4.21, showing some use of AI but no major shift in editorial strategy.
Newsroom Statements on AI Integration
To ensure we didn’t unintentionally misrepresent or unfairly characterize any newsroom’s use of AI, we reached out directly to numerous outlets, including those mentioned specifically in this report and others across the crypto media.
Responses came from Benzinga, Crypto.News, CoinDesk, Cointelegraph, and DL News, each outlining their specific approach to AI integration. While we haven’t yet heard back from Investing.com, The Defiant, Bitcoin News and others at the time of publishing this report, we will update this section as soon as their responses become available.
Below are the responses from each newsroom, published exactly as we received them:
Quotes
“Benzinga tries to stick to an “as much as possible, no more than necessary” policy when it comes to AI. Meaning we use AI for stories that require fast turnover and have little informational value beyond a headline or a data point. For more insightful coverage, we continue to rely on human-generated content and don’t expect major changes to this policy in the near future.”
Ivan Crnogatic
Managing Crypto Editor at Benzinga
At CoinDesk, we view AI as a powerful tool to help us cover the fast-evolving world of digital assets. Our top priority is delivering deeper, more insightful journalism to keep our readers informed on the developments shaping this space every day.
To support that mission, we’ve built AI tools that expand our coverage while freeing up our award-winning newsroom to go beyond headlines, uncovering the features, investigations, and exclusives that add tremendous value to our audience.
We use AI to surface topics, analyze data, and sift through hundreds of documents in minutes, allowing us to report on stories that would otherwise go untold. By combining public information with proprietary insights powered by CoinDesk Data, we’re pushing AI-powered reporting to the next level.
Our journalists and editors vet, edit and fact-check every AI-driven story before publication, ensuring CoinDesk’s strict quality controls are always upheld. To ensure we are fully transparent, CoinDesk has an industry-leading AI ethics policy that is cited in every article that uses AI.
AI and journalism can be a winning combination if used responsibly. Whoever can find the right balance will reap the benefits not only for their business but also for their customers. In a newsroom and industry as dynamic as ours, we want to show that CoinDesk is at the forefront of using such cutting-edge technology without compromising our journalistic integrity, which has won us many awards
Aoyon Ashraf
Aoyon Ashraf, Head of Americas, CoinDesk
Cointelegraph may use AI tools to assist with tasks such as drafting, research, translations, image creation, and workflow support, but all AI-assisted content is edited and fact-checked by humans before publication. We never use AI to invent quotes or market data, and we clearly disclose any significant AI use beyond the listed tasks through notes, captions, or byline tags. Human editors remain fully accountable for accuracy, fairness, and source protection, and all AI use is logged and monitored under strict security, privacy, and editorial standards.
Geraint Price
Head of Editorial at Cointelegraph
You may have noticed that the on-site disclaimers were not used so far. This is because while not strictly prohibited, extensive AI usage for writing content that is published is discouraged in my experience as a Cointelegraph journalist and we just avoid it. It is used in the research phase.
Adrian Zmudzinski
Journalist at Cointelegraph
If nearly half of the crypto media industry’s journalism is written by (or assisted by) AI purely for SEO, as you suggest, then you’ve got machines writing articles for other machines. It’s a race to the bottom. What happens when newsrooms get so “optimised for efficiency” that there are no human reporters left? Will AI attend conferences, publish scoops, build trust with sources, and ask difficult questions?
We also find that some outlets use AI to rewrite our original stories. What happens when there is no original reporting left because incentives are so broken and misaligned?
Given that language models are (currently!) mostly statistical pattern matchers, there’s still a categorical difference between text generation and epistemic labour. They synthesise existing data but don’t originate, say, journalistic investigation. That said, if/when we reach AGI, it might be capable of genuine journalistic work, and that’s a different conversation.
As for DL News, we don’t currently use AI to generate our articles. We don’t have a policy against our reporters using AI tools for spellchecking or brainstorming — they’re free to use them as they see fit, the same way they’d use Grammarly or Google. It’s good to experiment with technology.
I must also add that we’re also seeing a surge in AI-generated cookie-cutter PR and op-ed pitches. We reject most of them. They all sound like each other — same tone, same cadence, and same polish…
That said, we notice and welcome increasing traffic from LLMs. We’re exploring ways to make our content more accessible to AI models. If readers are asking questions into these black boxes, we want our journalism to be one of the answers they get. Perhaps the demise of SEO traffic will be a good thing for overall quality over the long run, and we’re just in the transition phase at the moment.
Ekin Genç
Editor-in-Chief at DL News
As an editor at crypto.news, I recognize that AI can be a fantastic research assistant and an awful storyteller. It’s incredibly helpful for accelerating background tasks like summarizing long research reports, extracting key points from filings, or finding stats to justify a claim. But when it comes to producing actual content, the end result is poor. AI-generated articles are, by definition, robotic, so they lack a genuine voice and can feel disingenuous to readers. The human layer is impossible to replicate (at least for now). I believe audiences can tell when something wasn’t written by a real person, whether it’s labeled or not.
Jayson Derrick
Editor at crypto.news
Conclusion
AI is no longer a future-facing experiment in crypto journalism. It’s part of the present. Some newsrooms have embraced it fully, using it to handle scale, speed, and volume. Others have kept their focus on human reporting, placing value on editorial control, voice, and consistency.
Neither approach is better. They reflect different definitions of trust, efficiency, and responsibility. What matters now is clarity-about when AI is used, how it shapes the final product, and where human oversight begins and ends.
The data shows that AI adoption is rising fast. But it also shows that human authorship still holds its ground. The future of news won’t belong to one or the other. It will depend on how well both are used together and adapted to fit the strengths of both technology and human insight, without trading trust for speed.
