The Other Side of AI

How AI-Generated Text Can Be Detected

Since ChatGPT's public release in late 2022, a parallel industry has grown around a single question: can software tell whether a piece of text was written by a human or a machine? Universities, publishers, and now blog platforms increasingly run submissions through "AI detectors" before accepting them. But how do these tools actually work, how accurate are they in 2026, and what do official research and publishing bodies say about relying on them? This article walks through the mechanics, the evidence, and the practical reality.

How AI detectors actually work

Most AI text detectors do not "recognize" a specific model the way a fingerprint scanner recognizes a finger. Instead, they analyze statistical patterns in language:

• Perplexity — a measure of how predictable each word choice is, given everything before it. Large language models tend to choose highly probable next words, which produces low-perplexity, "smooth" text. Human writing is often less predictable.
• Burstiness — the variation in sentence length and structure across a passage. Human writing tends to alternate between short and long sentences organically; AI output has historically been more uniform in pacing and rhythm.
• Watermarking and model-side signals — some newer approaches rely on statistical watermarks embedded by the model provider itself, or on classifiers trained to recognize a specific model family's output patterns.

Detectors such as Turnitin's AI Writing Indicator, GPTZero, Originality.ai, Copyleaks, and Pangram build on these signals, usually combined with machine-learning classifiers trained on large datasets of known human and AI text.

How accurate are these tools, really?

This is where the picture gets complicated. Independent academic evaluations published through 2026 consistently find that detector accuracy is far from the near-perfect numbers vendors advertise.

A 2026 study in the International Journal for Educational Integrity comparing Originality and Turnitin found that Originality outperformed Turnitin on overall accuracy (0.69 vs. 0.61) and recall (0.60 vs. 0.51) — but both tools performed poorly on "hybrid" text, meaning writing that mixes human editing with AI-generated passages, which is increasingly how AI-assisted writing actually looks in practice.^[2]

A separate 2026 systematic evaluation published on ScienceDirect concluded that current AI-generated-content (AIGC) detection tools "are not yet sufficiently robust or reliable for high-stakes academic decision-making," despite rapid improvements in the underlying detection algorithms.^[3]

A peer-reviewed evaluation in a medical education journal (PMC) testing detectors and human reviewers side by side found that while detection tools could "meaningfully distinguish plausible AI-use conditions," reliability varied significantly between tools, and human scoring accuracy was uniformly low — reinforcing that people are generally worse at spotting AI text than the software is.^[4]

The false-positive problem

Perhaps the most important finding for anyone publishing legitimate human-written content is the rate of false positives — human writing incorrectly flagged as AI-generated.

An earlier but widely cited 2023 evaluation by Weber-Wulff and colleagues, testing 14 detection systems, found that none of the tools reliably confirmed the accuracy claims made by their developers.^[5] A related study published in Patterns found that a large majority of TOEFL essays written by non-native English speakers were incorrectly flagged as AI-generated by at least one of several detectors — because formal, careful, less idiomatic writing statistically resembles AI output, even when a human wrote every word.

This matters directly for blog and journal content: polished, well-structured, formal writing — exactly what most publications ask for — is the style most likely to trigger a false positive, regardless of who actually wrote it.

What official publishing bodies actually require

Detection tools are only half the story. The other half is policy — what journals, publishers, and editorial bodies actually require from authors. As of 2026 there is a stable, cross-industry consensus:

• No AI tool can be listed as an author. The International Committee of Medical Journal Editors (ICMJE) states that AI tools cannot take responsibility for a work's accuracy or give final approval for publication — both required conditions for authorship — so they cannot be credited as authors, however much they contributed to drafting.^[1]
• Disclosure is required, not detection. The Committee on Publication Ethics (COPE) requires that authors using AI tools to draft text, generate images, or process data be transparent about it, typically in the methods or acknowledgements section, naming the tool and describing how it was used.^[6]
• Human authors remain fully accountable. Both ICMJE and the World Association of Medical Editors (WAME) are explicit that authors are responsible for the accuracy of anything an AI tool contributed, including checking for fabricated citations or incorrect claims — a well-documented failure mode of generative AI.^[7]

In other words, the publishing world's actual safeguard against undisclosed AI use isn't a detector score — it's a disclosure requirement, backed by the much simpler fact that authors are liable for what they submit, detected or not.

Comparing the major detection tools

Tool	Primary signal	Reported strength	Known limitation
Turnitin AI Writing Indicator	Perplexity/burstiness classifier	Widely deployed in higher education	Independent accuracy estimates trail vendor claims; struggles with edited text
GPTZero	Perplexity/burstiness classifier	Fast, widely used browser-based check	Sensitive to formal, non-native, or heavily edited writing
Originality.ai	Classifier + paraphrase-resistance layer	Outperformed Turnitin in 2026 academic testing	Still weak on hybrid human/AI text
Copyleaks / Pangram	Classifier + similarity/plagiarism check	Combines AI detection with originality checking	Accuracy varies by content genre and length

Practical takeaways for writers and bloggers

1. Don't treat a detector score as proof of anything. Every major independent study through 2026 warns against using a single detector result as evidence, in either direction.
2. Disclosure beats evasion. If you use an AI tool to help draft or organize content, a short, honest disclosure line is both the accepted best practice and far more defensible than hoping a detector never flags the piece.
3. Editing matters more than tools realize. Heavily revised, fact-checked, human-edited AI-assisted drafts are exactly the "hybrid" text that current detectors handle worst — which cuts both ways: it can evade detection, but it can also mean genuine human work gets wrongly flagged.
4. Accuracy and sourcing are the real risk, not detection. The bigger practical danger of AI-assisted writing is factual error or fabricated citations, not getting caught by a classifier. Verify everything an AI tool contributes before publishing it.

The bottom line

AI text detectors exist, they are improving, and some (like Originality.ai in recent testing) perform meaningfully better than others. But 2026 research is consistent on one point: none of them are reliable enough to serve as definitive proof that a piece of writing was or wasn't AI-generated, especially once a human has edited the draft. The publishing world has responded not by chasing better detectors, but by shifting the burden to disclosure and author accountability — a framework that works regardless of how good detection technology eventually becomes.

Sources

1. International Committee of Medical Journal Editors (ICMJE). Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals, Section V: Use of Artificial Intelligence in Publishing. icmje.org
2. Hadra, M., Cambridge, K., Mesbah, M. (2026). Evaluating the accuracy and reliability of AI content detectors in academic contexts. International Journal for Educational Integrity, 22(4). link.springer.com
3. Trusting AI to detect AI? A systematic evaluation of the reliability and robustness of current AIGC detection tools for student academic work (2026). ScienceDirect. sciencedirect.com
4. Ability of AI detection tools and humans to accurately identify different forms of AI-generated written content. PMC. ncbi.nlm.nih.gov
5. Weber-Wulff, D. et al. (2023). Testing of detection tools for AI-generated text. Summarized in: How Reliable Are AI Detectors For Academic Text? effortlessacademic.com
6. Committee on Publication Ethics (COPE). Authorship and AI tools — COPE position statement (13 February 2023). publicationethics.org
7. World Association of Medical Editors (WAME). Recommendations on Chatbots and Generative Artificial Intelligence in Relation to Scholarly Publication. wame.org

Why Blogging Still Matters In The Age Of AI

Type a question into an AI chat tool today and you get a clean confident answer in three seconds. No scrolling past ads, no reading five hundred words of introduction before the actual recipe or the actual fix appears. It feels like the search box has been replaced by something faster and smarter, and it raises an honest question for anyone thinking about starting a blog in 2026. If a machine can summarize the internet on command, why would anyone spend hours writing a post that a chatbot might just swallow and repackage in a paragraph?

It is a fair worry, and the numbers back it up more than most people expect. But the numbers also tell a second story that rarely gets attention, one about why original writing is becoming more valuable, not less, even as it becomes less visible.

The traffic numbers are real

The Pew Research Center tracked the browsing activity of nine hundred American adults across roughly sixty nine thousand Google searches in March 2025. Their finding was direct. When an AI generated summary appeared at the top of the results page, people clicked through to a website only eight percent of the time. Without that summary, the click rate was fifteen percent, almost double. Clicking a link placed inside the AI summary itself happened in just one percent of visits. People were also more likely to end their search session entirely once the summary answered their question, rather than continue reading elsewhere.

Ahrefs reached a similar conclusion from a different angle. Their analysis of three hundred thousand keywords found that the top ranking page for a search term lost more than a third of its usual click through rate whenever an AI Overview appeared above it. By early 2026, industry trackers estimated that AI generated summaries were showing up on close to half of all tracked Google searches, a sharp rise from under ten percent just two years earlier.

So the premise behind the worry is accurate. Fewer people are clicking through to read the original article. Search, the channel that built the entire blogging economy for two decades, is sending less traffic than it used to.

Why this does not mean blogging is finished

Here is the part that gets lost in the panic. AI answers have to come from somewhere. Every large language model and every AI search summary is built by reading, ranking and condensing text that a human being originally researched, tested and wrote down. Pew found that the typical AI summary cited three or more separate sources, and government and reference sites were cited noticeably more than average, meaning credible original material is exactly what these systems lean on. A blog is not competing against AI. In a very real sense, a blog is feeding it.

That dependence has created something new: a race among AI companies to secure reliable content, not avoid it. Major publishers including the Associated Press, Axios, The Guardian, News Corp and Conde Nast have signed licensing agreements with OpenAI, Google, Amazon and other AI developers, some reportedly worth tens of millions of dollars a year, specifically to keep their writing feeding these systems on agreed terms. At the same time, publishers who felt their work was taken without permission have gone to court. The New York Times is suing OpenAI and Microsoft. CNN sued Perplexity in 2026. News Corp sued Brave over how its content was being scraped. More than seventy separate copyright lawsuits against AI companies were active by the end of 2025, more than double the number from a year earlier.

None of this is happening because original writing has become worthless. It is happening because original writing turned out to be the raw material the entire AI industry runs on, and everyone involved, from billion dollar publishers to individual writers, is now working out what that material is worth and who should be paid for it.

What actually changes for an independent blogger

The honest answer is that the old game, where you wrote a post, ranked on Google, and collected ad revenue purely from search visitors, is weakening. Relying on search traffic alone as a business plan is riskier today than it was five years ago. But several things still work, and some work better than before.

• Being the cited source still counts. Seer Interactive found that brands mentioned inside an AI Overview earned around thirty five percent more of the remaining organic clicks than brands left out of it. Visibility inside the summary itself is becoming the new ranking position worth fighting for.
• Direct relationships matter more than search traffic ever did. Email lists, newsletters and returning readers are not affected by how Google displays its results page, because that traffic never depended on Google in the first place.
• Specific firsthand experience is difficult for a model to fake. A generic explanation of a topic is easy for AI to summarize and replace. A personal product test, an original photo, a documented case study or a strong opinion built on real experience is not something a summary can substitute for.
• Depth still has an audience. Readers who need a quick fact use AI. Readers who want to genuinely understand a topic, compare options or follow an argument still seek out full articles, and that audience tends to be more loyal and more valuable to advertisers and sponsors.

The fair question about getting paid

This brings up the second half of the worry, and it deserves a straight answer. If a blogger spends hours researching and writing something original, and an AI system reads that page and then answers a reader's question without sending them to the blog at all, how is the blogger supposed to earn anything from that work?

Right now there are three real paths being tested at once, and none of them existed in this form three years ago.

The first is licensing, the same model the big publishers are using, now becoming available to individual creators and smaller publishers as AI companies expand their programs to bring in more original content on agreed terms. 

The second is legal pressure, with the growing wave of lawsuits and the United States Copyright Office openly questioning whether unlicensed AI training qualifies as fair use, a debate whose outcome will shape whether smaller writers get paid too. 

The third, and the one within a blogger's own control today, is building an audience that AI cannot intercept, through email subscribers, a paid community, a product, a course or a service that grows out of the trust built by consistently useful writing.

None of these paths guarantee income the way ranking first on Google once did. But they describe a business built on being trusted by real people rather than one built entirely on being found by an algorithm, and that kind of business tends to survive changes in technology far better than one that depends on a single traffic source.

The honest conclusion

Blogging is not ending. 

The specific business model of writing generic articles purely to rank on Google and collect ad clicks is under serious pressure, and the data on click through rates leaves no real doubt about that. But the underlying activity, researching something carefully and explaining it clearly in your own voice, is more necessary than ever, because it is the exact thing every AI system needs and cannot fully create on its own. The blogs that struggle going forward will be the ones offering nothing a summary could not already provide. The blogs that thrive will be the ones offering something a summary never can: a real person, a real experience and a reason to keep reading.

So keep doing blogging. 

XIV verbena Republicana Valladolid

The Verbena Republicana in Valladolid is a recurring music and community festival with a republican political/cultural identity, organized by the Ateneo Republicano de Valladolid.

“Verbena” in Spain basically means a festive outdoor evening with live music, drinks, dancing, and socializing. “Republicana” refers to the event’s association with the republican movement—broadly, people who favor a Spanish republic rather than a monarchy.

For 2026, the 14th Verbena Republicana is happening today, September 9, at Plaza Andrés de Laorden in Valladolid. It is free and includes:

• Delameseta — 20:00
• Hula Baby — 21:30
• Cañoneros — 23:00
• An official address from the Ateneo Republicano around 22:45
• Music/socializing until about 01:00

It has been held for many years as part of Valladolid's September festivities; the 2025 edition was the 13th.

So if you saw “Verbena Republicana Valladolid” on today's events listings, it's essentially a free outdoor concert/party organized by a republican civic group, rather than a formal political rally.

Which Country Buys the Most Airline Tickets Each Year?

Which Country Buys the Most Airline Tickets Each Year? (2025 Data)

Every year, hundreds of millions of people around the world buy plane tickets — for business trips, family visits, holidays, and everything in between. But when you add it all up by country, one nation towers over the rest: the United States.

According to the International Air Transport Association's (IATA) latest World Air Transport Statistics report, the US remained the world's biggest passenger market in 2025, with 890.1 million passengers recorded arriving and departing. That's not a typo — nearly 900 million tickets flown in a single year, in a country of roughly 340 million people. On average, that works out to more than two and a half flights for every man, woman, and child in America.

The Top 5 Air Travel Markets in 2025

Rank	Country	Passengers (2025)	Year-over-Year Growth
1	United States	890.1 million	+1.6%
2	China	776.1 million	+4.8%
3	United Kingdom	269.7 million	—
4	Spain	—	—
5	Japan	—	—

The gap between first and second place looks large, but it's closing. China's aviation market grew nearly three times faster than America's in 2025, and it isn't a fluke — it's part of a long-running trend.

Why the US Leads

A few factors explain why America has held the top spot for so long:

Sheer geography. The US spans nearly 3,000 miles coast to coast, and unlike much of Europe or East Asia, it never built out a national high-speed rail network. For a trip from New York to Los Angeles, flying isn't just the fastest option — it's often the only practical one.

Population and wealth combined. Size alone doesn't guarantee high ticket volumes (India, for instance, has four times the US population but a much smaller aviation market). What matters is the combination of a large population and enough disposable income for people to fly regularly, whether for work or leisure.

A dense hub-and-spoke network. Major hubs like Hartsfield-Jackson Atlanta International Airport, Dallas-Fort Worth, Chicago O'Hare, Denver, and Los Angeles funnel huge volumes of connecting traffic, on top of point-to-point demand. Atlanta alone has spent years as the world's busiest airport.

Global hub status. Beyond domestic flying, the US serves as a major gateway for international travel between the Americas, Europe, and Asia-Pacific, adding to the total passenger count.

China Is Catching Up — Fast

China's rise is the other big story in global aviation. Its domestic market has exploded alongside rising middle-class incomes and an aggressive airport-construction program. Industry forecasts, including estimates from Airbus, suggest China's domestic air travel market could be roughly a third larger than the US market by 2045, with several analysts projecting China will overtake the US as the world's single largest air passenger market sometime in the 2030s.

For now, though, the numbers still favor the US by a wide margin — more than 100 million passengers ahead of China in 2025.

Beyond the Top Two

After the US and China, passenger totals drop off sharply. The UK, Spain, and Japan round out the top five, but none come close to the scale of the two leaders. Countries like Ireland, Panama, Singapore, and the Netherlands also punch above their weight in passenger traffic — not because of domestic demand, but because their airports function as major international transfer hubs (Ryanair's registration in Ireland, for example, means its flights count toward Irish totals even when neither endpoint is in Ireland).

Regionally, the picture looks a little different: Asia-Pacific as a whole is now the world's largest air travel region, driven by enormous intra-regional demand across countries like China, India, Indonesia, and Japan — even though no single Asian country yet rivals the US on its own.

The Bottom Line

For now, the United States remains the undisputed leader in annual air ticket purchases, a position it has held for decades thanks to its geography, economy, and aviation infrastructure. But the gap is narrowing, and China's rapid growth suggests the title of "world's biggest air travel market" may change hands within the next decade or two.

Data source: IATA World Air Transport Statistics, 2025 report.