Anthropic is offering new details about how it plans to watermark text written by Claude — including what happens when someone edits the output.

The company uses a simple example to explain the technology. If Claude can choose between words such as “overcast” and “grey” to describe the weather, it can make those small choices in a pattern. A reader wouldn’t notice. But someone with the right detection tool could identify the pattern as a watermark.

Anthropic described the system in a Friday TechCrunch blog post answering questions about the plan, which it announced earlier in the week to comply with the European Union’s AI transparency rules. We covered the original announcement here.

The company plans to use SynthID Text, a watermarking technology developed by Google DeepMind. Anthropic also says it will release an API to detect the watermark.

The system has limits, particularly once people start changing Claude’s words.

“Light editing probably won’t remove the watermark completely,” Anthropic said. But a complete rewrite in which every word is replaced would remove it. Anthropic said in such a case, it becomes harder to describe the finished product as AI-generated.

Whether a watermark can be detected also depends on how much Claude actually wrote. If someone gives Claude a human-written document and asks it to make only minor edits, Anthropic said that leaves “very little (if anything) for the watermark to attach to.”

Code presents a similar problem. Code gives a model fewer stylistic choices than prose. Anthropic said it can still watermark choices, including some comments and names within code, but the technology should have “a negligible effect on the actual code produced.”

The company is also drawing a distinction between watermarking and statistical AI detectors.

Those tools look for patterns associated with AI writing and estimate whether a model produced a piece of text. A watermark works differently, using a pattern deliberately inserted when the text was generated.

An AI detector makes a statistical judgment. A watermark provides evidence that a particular system marked the text when it was created.

Not everyone is happy about the change. TechCrunch reported that dozens of Claude users on X said they had canceled subscriptions over the watermarking plan, citing Business Insider.

For news organizations, the technology could make questions about AI use easier to investigate.

An editor could check whether copy submitted by a freelancer carries Claude’s watermark. But competitors, critics and others with access to Anthropic’s detection system could check published work as well.

That makes newsroom policies around AI use more consequential. Publishers may need clear rules about when it is permitted, how much AI-generated material can appear in published work and when that use should be disclosed.

Claude won’t necessarily be the only model leaving marks behind. Anthropic says other major AI developers that signed the same European code of practice are also working on watermarking systems.

What remains unclear is whether those systems will work together. Publishers find themselves checking text against a different system for every AI company.

The post Anthropic says Claude’s text watermark survives light editing but not rewrites appeared first on The Media Copilot.

Time Inc. is inserting ads into the markdown versions of its webpages, stripped-down formats that AI systems and agents can read. The ads are formatted as FAQs filled with a brand’s messaging and labeled as sponsored content. Time says it is the first publisher to sell advertising inventory in this format.

The idea has divided media buyers, Digiday reports. Some see it as a potential shortcut to a problem their clients are eager to solve. Others dismiss it as wishful thinking.

The problem is real enough. More consumers use tools like ChatGPT and Google’s Gemini for product research, and brands have almost no visibility into what those chatbots say about them. Most have turned to generative engine optimization, or GEO, pumping out branded content and using visibility trackers like Scrunch and Profound. Those tactics take time and guarantee nothing.

“Right now there’s a lot of testing based off of building content and hoping that it shows up within the LLMs,” said Jeff Eisenfeld, director of activation at Media by Mother. “Is that going to show up in the prompt? Is that going to have enough authority? Is that going to get crawled?”

That uncertainty makes a direct ad buy more appealing. Jonah Goodhart, co-founder and CEO of Mobian, the company behind Time’s inventory, told Digiday that paid ads can influence LLMs because the models seek out information from trusted brands.

Sam Huston, senior vice president of media at growth invention company Dept, said clients would consider testing the ads depending on the cost. He pointed to high-consideration purchases such as cars, where consumers may spend weeks researching a decision and increasingly turn to LLMs for help. Jaquie Hoyos, chief media officer at Moroch, said the key question is whether placing ads alongside high-quality information can actually influence how brands appear in AI results.

Skeptics are pushing back. Danny Weisman, co-founder of Obsessed Media, said brands would be better off investing in traditional brand-building advertising. He argued that markdown ads should be treated as an added value from publishers rather than a core part of an advertising campaign.

The data also raises questions about whether these ads can meaningfully influence AI visibility. A WARC study conducted by agency Charlie Oscar estimated that 63% of a brand’s LLM visibility came from long-term brand equity, compared with 26% from current marketing activity.

Stephan Kopp, managing director of Mediaplus Performance, was more skeptical, arguing that the ads are unlikely to influence AI models. He also warned that AI developers could eventually adjust their systems to ignore such tactics, much as search engines have changed their algorithms to prevent advertisers and publishers from gaming rankings.

For newsrooms, the debate matters because markdown inventory offers publishers a potential way to monetize the AI traffic already reaching their sites. Time’s experiment tests whether publishers can package and sell that traffic before regulators or AI companies set the rules.

The source of those test budgets is telling. Huston expects advertisers to shift money from programmatic display rather than search, despite the goal of improving AI search visibility. Some budgets could also come from the same experimental funds advertisers are using to test ads on ChatGPT.

Rita Steinberg, vice president of media at FUSE Create, summed up the view shared by many buyers: “You can influence AI visibility. I just don’t think you can reliably buy it yet.”

The post Media buyers debate whether ads can sway AI search appeared first on The Media Copilot.

Thirty-seven hosts for home shopping network QVC voted Thursday to unionize with SAG-AFTRA, turning concerns over generative AI and the use of their voices and likenesses into a new front in the media industry’s growing labor battles. Only nine voted against the move, giving SAG-AFTRA roughly 80% support in a National Labor Relations Board election covering full-time and part-time presenters at the company’s West Chester, Pennsylvania, headquarters, according to The Hollywood Reporter.

The vote covers hosts who appear across linear television, livestream and shortform programming. While the bargaining unit is small, the fight behind it reflects a broader industry battle over a question becoming harder to ignore: who controls a performer’s image when companies can create synthetic versions of it.

SAG-AFTRA President Sean Astin framed the vote around that concern. “QVC, HSN and Omni Channel employees deserve fair compensation, AI protections and a stronger voice in decisions that affect their work,” he said in a statement, adding that the union’s 160,000 members are proud of the hosts for organizing.

QVC said it has historically preferred a direct relationship with employees but will “respect the outcome of the election” and begin working with SAG-AFTRA toward a first contract.

The organizing effort was driven in large part by concerns over generative AI. The Hollywood Reporter reported earlier this summer that hosts feared their images, voices and likenesses could be used without consent or compensation. In a petition delivered to management, the hosts said they “should have meaningful input into [their] role in the network’s future, and that this is best accomplished through a formal collective-bargaining process.” They also sought clearer standards around pay, promotions and job protections.

Those demands mirror the fight SAG-AFTRA already waged with Hollywood studios in 2023, when AI consent and compensation protections became central issues in the film and television contract that ended a monthslong strike. QVC hosts do not have the same bargaining power as scripted actors, but the underlying concern is similar: companies gaining the ability to create digital versions of a person’s face or voice without their permission.

The timing adds another layer of complexity. Qurate Retail Group, QVC’s parent company, filed for Chapter 11 bankruptcy protection in April after cutting 900 jobs earlier this year. The company is also shifting its business model toward live shopping on social platforms and streaming services as traditional linear television declines. SAG-AFTRA will enter negotiations with a company restructuring both its finances and its distribution strategy.

For other newsrooms and media companies leaning on on-camera talent, hosts, anchors, livestream personalities, the QVC case is a preview. As publishers experiment with AI-generated avatars and voice clones to expand content across platforms, the people whose likenesses power those systems are seeking contract protections before the technology gets ahead of them. The Media Copilot has tracked similar consent and compensation disputes emerging wherever generative tools touch a recognizable face or voice.

What QVC’s first contract with SAG-AFTRA ultimately includes on AI rights, consent and compensation could become a reference point for other retail media and livestream operations facing similar questions.

The post QVC hosts vote to unionize with SAG-AFTRA over generative AI concerns appeared first on The Media Copilot.

Anthropic put Claude Opus 5 on sale today at $5 per million input tokens and $25 per million output tokens, the exact prices it charged for the previous Opus 4.8. According to Anthropic’s announcement, the model reaches close to the intelligence of its higher-end Fable 5 model at half the cost, and it now serves as the default model on Claude Max and the strongest option on Claude Pro.

The company is leaning hard on benchmark numbers to make its case. On Frontier-Bench v0.1, Anthropic says Opus 5 beats every other model and more than doubles Opus 4.8’s score at a lower cost per task. On ARC-AGI, a test built around novel reasoning problems, the company reports Opus 5 scoring three times higher than the next-best model. On Zapier’s AutomationBench, which checks whether a model can run a business task end to end, Anthropic claims a pass rate roughly 1.5 times the nearest competitor at the same cost.

The recurring theme in the launch is verification. Anthropic describes Opus 5 checking its own work before handing it back. In one Frontier-Bench task, the model was asked to rebuild a machine part in 3D code but given no way to view the drawing, so it wrote its own computer vision pipeline to pull the geometry from raw pixels. In another example, it found the root cause of a bug in an open-source package manager that the community’s own patch had missed.

Early-access customers echoed that framing. JetBrains said the model catches its own logical faults during planning rather than after. A legal-tech tester reported first-turn redline scores nearly double Opus 4.8. Box measured an 8% overall improvement over Opus 4.8, with 17% gains on due-diligence workflows. These are vendor-supplied quotes, so treat the precise figures as marketing rather than independent measurement.

On safety, Anthropic says Opus 5 is its most aligned model so far, scoring 2.3 on its automated misaligned-behavior audit, the lowest of its recent releases. The company also notes the model stays behind its Mythos 5 model on biology research and offensive cybersecurity. Notably, Opus 5 can find software vulnerabilities about as well as Mythos 5 but lags badly at writing exploits for them, which Anthropic frames as a deliberate safeguard. Its cyber classifiers block binary vulnerability scanning, penetration testing and exploit generation, with flagged requests falling back to Opus 4.8.

For newsrooms and publishers, the pricing is the story. Holding costs flat while claiming stronger reasoning and cleaner outputs matters for teams running document analysis, research summaries and data work at volume. The customer notes about tighter, more concise responses and fewer tool calls point to lower token spend per task, which is where AI budgets actually get decided. Publishers weighing model choices should still run their own tests against real workflows rather than trusting benchmark charts, a point we’ve made repeatedly at The Media Copilot.

Anthropic paired the launch with two beta features: mid-conversation tool changes that don’t break the prompt cache, and automatic API fallbacks that route flagged requests to another model instead of blocking them. A Fast mode runs about 2.5 times the default speed at twice the base price. The bet is that a cheaper, more careful default beats a smarter but pricier one for daily use, and the next few months of real deployments will show whether that holds.

The post Anthropic ships Claude Opus 5, pitching frontier work at half the price appeared first on The Media Copilot.

The New York Times and a group of other publishers are asking a federal court to sanction OpenAI, accusing the company of withholding or destroying evidence in a high-stakes copyright case over how ChatGPT was trained and used.

In a motion filed Thursday in federal court in Manhattan, the publishers alleged that OpenAI misrepresented its ability to search training datasets and ChatGPT output logs for copyrighted news material. According to Reuters, the publishers said OpenAI told the court it could not search its large language models for their work while allegedly concealing that it had already done so “even before the first News Plaintiff filed suit.”

The motion is the latest escalation in the copyright fight between major news organizations and AI companies. It also moves the dispute deeper into discovery, where the question is not just whether AI companies can use journalism to train models, but whether they can preserve, search and produce the records needed to prove what happened.

The plaintiffs include The Times, the New York Daily News and other media organizations, including Ziff Davis and the Center for Investigative Reporting, according to The Associated Press and Variety. The original New York Times article reported that the publishers are seeking legal sanctions against OpenAI, including monetary penalties and other remedies.

The filing does not ask for sanctions against Microsoft, which is also a defendant in The Times’ broader copyright case, according to The Times’ summary of the motion. Microsoft has invested heavily in OpenAI and integrated OpenAI technology into products including Copilot.

“The evidence is in OpenAI’s training data sets and ChatGPT output logs,” the publishers said in the motion, according to The Times. “But instead of just producing that evidence at the start of the case and focusing on the merits of its fair use defense, OpenAI chose obstruction.”

OpenAI rejected the allegations. “As the Times’ case weakens and they’ve been forced to drop claims against us, they’re persisting with their efforts to invade the privacy of people who have nothing to do with this case, including by making these blatantly false allegations,” OpenAI spokesperson Drew Pusateri told Reuters. “We’ll continue defending our users’ privacy and the long-established principles of fair use.”

The publishers allege that OpenAI deleted billions of relevant ChatGPT conversations or made them unsearchable. They also argue that an OpenAI employee later testified that the company had performed multiple searches for news publishers’ content, contradicting earlier representations about the company’s technical limitations.

A sanctions memorandum posted by Ars Technica says the publishers want the court to bar OpenAI from relying on a disputed 20 million-log ChatGPT sample, find that ChatGPT’s output logs include or would have shown substantial use of the publishers’ copyrighted material, instruct the jury on those findings and award fees and costs tied to the discovery fight.

Those remedies would matter because discovery disputes can shape the trial record. If the court finds OpenAI failed to preserve or produce relevant evidence, the ruling could affect what arguments OpenAI can make later and what conclusions a jury may be allowed to draw from missing or incomplete records.

The Times sued OpenAI and Microsoft in 2023, alleging that millions of Times articles were used without permission to train AI systems that now compete with publishers as sources of information. OpenAI and other AI companies have argued that training models on large bodies of text is protected by fair use, a theory now being tested across lawsuits from authors, artists, music labels and news organizations.

For publishers, the issue goes beyond training data. They argue that AI chatbots and AI search summaries can answer readers’ questions using journalism without sending traffic, licensing revenue or subscribers back to the organizations that reported the information. Media Copilot has been tracking the same pressure point in coverage of Google’s AI accuracy problem and The Times’ warnings about AI companies using journalism without permission.

At the same time, publishers are taking different approaches to the AI economy. Some are suing. Others have signed licensing deals with AI companies. The Associated Press announced a deal with OpenAI in 2023, while other media companies have made agreements with OpenAI, Google, Meta and Amazon.

The sanctions motion could increase pressure on both sides. A ruling against OpenAI would give publishers leverage in court and in licensing talks. A ruling for OpenAI would strengthen the company’s argument that publishers are using discovery to intrude into user privacy and commercially sensitive systems.

Either way, the case shows that AI copyright fights are becoming data-governance fights. The central questions are no longer only what AI systems were trained on. They are whether companies can prove it, search it, preserve it and explain it in court.

The post Publishers ask court to sanction OpenAI in escalating copyright fight appeared first on The Media Copilot.

For a while, it seemed like Google Search was in trouble.

Seemingly caught by surprise by the AI revolution that ChatGPT sparked, Google looked old and confused as upstarts like OpenAI and Perplexity pointed to a new future that replaced the “10 blue links” with question-and-answer conversations. Google’s first steps into this future were unsteady, with error-filled answers epitomized by the infamous glue-on-pizza moment. Some suspected, for all its scale and influence, a post-Google world was near.

That looks a lot less likely after this week. At Google I/O, the company confidently showed us its version of our informational future. And while it might be post-search, it’s not at all post-Google. Google is expanding its use of AI Overviews, meaning more searches will include the top-of-page summaries, and it’s adding a query box within them. When a user engages with it, they’re kicked to AI Mode, which abandons the “10 blue links” altogether.

In addition, Google.com now has a “+” icon, similar to its Gemini chatbot. If user engages with it and uploads a file or photo, that will also take them to AI Mode. It’s now extremely difficult to search on a Google product without AI being part of the result. You can still find your page of links by switching to “Web,” though that option is often buried.

So, far from the future where search is competitive again, it’s increasingly looking like a new future that’s the same as the old future. Even if you look just at AI chatbots, the Gemini app is now at 900 million users, making it about as big as ChatGPT. That doesn’t even count AI Overviews and AI Mode, which have 2.5 billion and 1 billion users, respectively, according to the company.

The bots ARE the traffic

The obvious consequence of all this is more searches will begin and end in the query. For publishers, that continues and likely accelerates the ongoing traffic apocalypse. We may, however, have to update our vocabulary: Google Zero—which was supposed to connote an environment where the clicks from Google search were basically nil—feels imprecise.

That goes double when you consider that, as humans spend more time in AI interfaces, a commensurate amount of bot activity spreads out from those queries. So the future isn’t Google Zero. It’s Google Bot Infinity.

So the future is a world where people happily chat—either via typing or speech—to Google, and those Google bots bring the right information and context to answer them. More accurately, those bots bring what they deem as the right information and context to queries. AI systems prioritize information differently from traditional search, looking for information that both fits a pattern but also includes novel and authoritative elements. This is manifesting into the new-but-rapidly-evolving field of GEO, or generative engine optimization. Google’s renewed push into AI experiences means the battle for presence in answers is no longer a side bet. It’s the game.

That’s the media story here in Google’s renewed rise. Once laughed at for how far behind it was in the AI race, it’s now architecting the future where it’s still in charge. Judging by its balance sheet—with earnings steadily increasing even as competitors rise—it’s found the right balance of building the new while preserving the old. Even as it demotes the “10 blue links” that built the company, it’s offering a bevy of new ad products in conversational search that spin up generative ads on the fly. It clearly has the confidence that it can make money in an AI world.

Brands might be less confident about that, and publishers even more so. Authority in AI answers is nice, but monetizing has so far been a challenge.

Credibility is the new click

But it’s not nothing. If Google’s AI layer becomes the place where people encounter information, then presence inside that layer becomes a form of distribution. A publisher cited consistently in answers about politics, technology, health, finance, or culture has something valuable: proof that it owns authority in a category. The old metric was how many people Google sent to you. The new one may be how often Google needs you to make its answers credible.

That may not produce the same clean, scalable ad business that search referrals once did. But it points to a different one. Advertisers have always wanted to sit next to authority. They sponsored sections, bought podcast reads, backed newsletters, underwrote events, and cut direct deals with creators because association matters. If a publisher becomes one of the sources AI systems repeatedly rely on, that authority can be sold directly—not necessarily through Google, and not necessarily as a banner ad awkwardly stapled to a webpage.

That’s the hopeful version of Google Bot Infinity. Publishers may lose a lot of casual traffic, and pretending otherwise is foolish. But the ones that produce distinctive, trusted, deeply useful work still have leverage. The job now is to make that work legible to machines without making it lifeless for people.

The post The end of 10 blue links is not the end of Google appeared first on The Media Copilot.

How can you tell something’s AI-generated? When it comes to writing, there are common tells: the excessive use of em dashes, sentences that are too rhythmically clean, and a general smoothness that feels overly engineered.

It’s hardly a perfect science, though, and most humans’ AI detection skills are based on vibes.

If humans are just relying on instinct, what are AI detectors relying on? Here, Zapier shares everything you need to know about how AI detectors work.

What is an AI detector?

An AI detector is a tool that analyzes content like text, images, or videos, and estimates the likelihood that it was generated by an AI model. Instead of giving a definitive yes-or-no answer, most AI detectors will give you:

• A probability score (for example, “74% likely AI-generated”)
• A confidence rating
• Highlighted passages that appear machine-written, if it’s text

Their goal isn’t to “catch” AI with certainty, but to flag content that statistically resembles AI-generated patterns.

How do AI detectors work?

The specifics of how AI detectors work vary depending on what type of content they’re analyzing. For simplicity, this article will focus on AI text detectors. But other types—like AI image detectors—work similarly.

Large language models (LLMs) generate text by predicting the most likely next word based on probability. It’s more nuanced than that, but that’s the idea. AI detectors reverse-engineer that idea: They look at a finished piece of writing and measure how closely it matches those probability patterns. Here are the main techniques they use.

1. Perplexity

Perplexity (not to be confused with the AI-powered search engine) measures how unpredictable a piece of text is to a language model. The lower the perplexity, the more the wording follows patterns the model expects to see.

AI-generated text often has lower perplexity because it’s built from highly common word sequences. It gravitates toward phrasing that’s safe, common, and structurally sound. Which is kind of the point. AI models are trained to predict the most probable next word, not the most chaotic or idiosyncratic—just the most likely.

Human writing, on the other hand, tends to raise the perplexity score because it’s usually less predictable. Unless you have a ruthless editor who’ll set you straight, humans use words that technically work, even if they’re not the exact right ones. They go off on tangents and litter their work with comma splices because those pauses just feel right to them.

2. Burstiness

Burstiness looks at sentence length distribution and structural variation to identify patterns that appear overly consistent.

Humans rarely write in perfect cadence. They mix short sentences with longer ones, occasionally go on tangents, and vary pacing without thinking about it. Earlier AI models, by contrast, tended to produce writing that felt evenly spaced and neatly balanced. Nothing was outright bad, just … suspiciously consistent.

That “too rhythmic” quality is often what sets off our internal AI radar. AI detectors try to quantify that instinct by measuring variation in sentence length, punctuation, and structure. If the tempo barely changes from start to finish, that uniformity can raise a flag.

3. Classifiers

A classifier is a machine learning system trained to categorize text as likely human- or AI-generated. Unlike perplexity or burstiness, which are individual signals, a classifier looks at many features at once and weighs them together.

Developers train their LLMs on large datasets of labeled human and AI text. Through that training, classifiers learn statistical patterns that tend to separate the two categories. Those patterns can include predictability scores, sentence variation, word frequency distributions, and other structural signals.

When you paste new text into an AI detector, the classifier evaluates how multiple signals interact and then produces a probability score. The final output reflects whether the writing, on average, more closely resembles patterns associated with AI-generated text or human-written text.

4. Stylometric analysis

Stylometric analysis is the study of writing style, including vocabulary richness, repetition, and sentence complexity. Think of it as your linguistic fingerprint.

The idea is that humans tend to develop quirks over time. For example, the author Fredrik Backman typically writes stories with a sort of progressive repetition that’s hard to describe, but is uniquely him. It’s what makes his writing so easily distinguishable.

AI writing, by contrast, often clusters around high-probability patterns, generating phrasing that reflects widely represented patterns rather than highly idiosyncratic ones. That’s also what makes much of AI writing feel technically solid but vaguely same-y.

5. Watermark detection

Watermark detection is a way of identifying AI-generated text by looking for a hidden signature baked into the writing itself.

Not all AI models use watermarking, and there isn’t one standard way to do it. But when watermarking is enabled, the model slightly nudges its word choices in a consistent, trackable way. The shifts are subtle enough that you wouldn’t notice anything while reading, but an AI detector that knows what to look for can spot the pattern.

In theory, that makes AI-generated content easier to trace. In reality, even light editing or paraphrasing can blur or erase the signal. So while watermarking sounds like a clean solution, it’s not foolproof.

How accurate are AI detectors?

AI detectors are probabilistic tools, not lie detectors. A detection score reflects how closely writing matches certain patterns. It doesn’t prove who or what actually wrote the text.

Here’s why accuracy gets complicated.

• False positives happen. Some human writing naturally resembles AI-generated text. If you refuse to give up the em dash and sprinkle them liberally throughout your writing, an AI detector may flag it as machine-written, even if it wasn’t.
• False negatives happen. AI models are improving at an alarming speed and learning to mimic human variability more effectively. Humans, for their part, are learning to refine their AI prompts to inject human signals—for example, telling their AI writing generator to mix up sentence patterns or intentionally include errors. As AI writing and human prompting become more nuanced, detection becomes harder.
• Hybrid content blurs the line. Most writing today isn’t purely human or AI. AI detectors struggle in this gray area because the final text contains both human and machine signals.
• Results vary across tools. Different AI detectors use different training data and different models. The same paragraph can receive dramatically different scores depending on the platform. That inconsistency makes it risky to rely on a single detection result for high-stakes decisions.

The bottom line on AI detectors

We’re no longer living in a binary world of purely human or purely AI-generated writing. A lot of content now sits somewhere in between. A draft may start with AI, a human reshapes it, AI tightens a paragraph, a human adds a lived example—the lines blur. And AI detectors have to make probabilistic guesses in that gray space.

This story was produced by Zapier and reviewed and distributed by Stacker.

The post How do AI detectors work? appeared first on The Media Copilot.

This is an all-too-common scenario: An organization is excited about the possibilities of AI. There’s tremendous internal buzz about the launch of an AI pilot. After the launch, though, there’s not much news about any real results. Eventually, the pilot winds down with little fanfare, and things go back to normal, except for a lingering company-wide fear that the organization is further behind in the AI race.

According to an IDC report, approximately 88% of AI proofs of concept (POCs) launched by surveyed enterprises never reach production. The report states, “The high number of Al POCs but low conversion to production indicates the low level of organizational readiness in terms of data, processes and IT infrastructure.” MIT Media Lab’s “State of AI Business 2025” report produced an even more stark finding: 95% of generative AI pilots in enterprises have delivered no measurable ROI.

Why is this happening? Teradata, an autonomous AI knowledge platform, suggests that AI pilots stall and fail to scale because of fragmented data silos and architectures that are designed for static reporting instead of dynamic intelligence.

The unstructured data gap enterprises can’t afford to ignore

Enterprises are looking to extract insights, but they’re not taking the steps to handle structured data alongside the torrent of unstructured data like images, audio, PDFs and customer chats. They must come to grips with the fact that unified data architecture is not merely a technical preference for AI, but a strategic prerequisite.

Structured data is relatively easy to query, parse, and analyze because it’s organized in databases with strict structures and predefined fields. But unstructured data is fueling this new era of LLMs — and to a greater extent, agentic AI.

Gartner notes that unstructured data is growing rapidly, often at a rate of 40% to 60% per year. They further estimate this unstructured data, including documents, emails, images, audio and video files, comprises 70% to 90% of enterprise information.

To understand why this gap matters, consider an airline trying to analyze customer feedback through unstructured channels such as emails, chat logs, and qualitative surveys. They tried to use an external LLM and strong prompt engineering. That approach worked well enough in development, but it broke down at scale.

They solved the problem by using open-source models to convert customer messages to vector embeddings. Vectoring is a way to numerically represent pieces of unstructured data so AI models can parse them.)Then, they were able to match conversation based on topics and sentiment, rather than simply with keywords.

The external model that the airline initially tried to use may have been perfectly capable, but the barrier to moving it into production was a data architecture problem that they had to solve first. The lesson is that many AI failures are not model failures. They are architecture failures. And solving them requires enterprises to rethink the way their data environments are built.

Traditional data pipelines were built to move information from one place to another. Agentic AI requires something much more dynamic.

From information pipelines to intelligence architecture

Information pipelines, along with extract, transform, load (ETL) processes that move structured data, are no longer sufficient. Today’s enterprise data also depends on an intelligence architecture comprising knowledge, context and measurable outcomes.

Dynamic context engines handle constant uncertainty, unpredictability and variability, particularly within the context of an enterprise’s own knowledge. Even the most advanced model is only as useful as the context it can access.

A unified knowledge layer is necessary to integrate important business context with data and insights to make data actionable — one where AI systems can reason, decide and act. Alongside measurable business outcomes, there must also be a governance layer built into the architecture, so enterprises can experiment and work safely, with built-in compliance and security.

All of this has to happen at speed, not just at scale, especially in environments where decisions must be made in real time and governance cannot be compromised. One example is defense, where structured and unstructured data needs to be processed in real time within strict security protocols. For example, if military organizations need to determine the survivability of given camouflage applications in real time, troops on the ground can use secure apps to take images of camouflaged assets and send them to be analyzed.

Combining structured and unstructured data in a single, governed database allows the system to process images alongside data such as terrain patterns and threat signatures and deliver guidance to soldiers in situ.

The agentic AI opportunity, and the gap between ambition and execution

Gartner predicts that 40% of agentic AI projects will be canceled by 2027, “due to escalating costs, unclear business value or inadequate risk controls.” Regardless, enterprises need to invest in foundational capabilities to build towards implementation, which Bain reports could demand 5% to 10% of technology spending over the next three to five years.

The Capgemini Research Institute pegs the economic value generation of agentic AI at $450 billion by 2028, even though just 2% of organizations surveyed are currently at full-scale deployment. And the Futurum Group predicts that as agentic AI replaces data pipelines, and enterprises move from experimental pilots to production, the data market could reach $541.1 billion in 2026 — and $1.2 trillion by 2031.

The market opportunity is enormous, and there are numerous real-world examples of where a unified, context-rich architecture enables agentic AI to have an impact.

The enterprises that move beyond AI pilot projects into production will be those who solve their data readiness challenges — unifying structured and unstructured data with agentic AI capabilities across any environment, and without compromising on governance.

This story was produced by Teradata and reviewed and distributed by Stacker.

The post The new agentic AI battleground: The case for unified architecture appeared first on The Media Copilot.

AI-referred visitors aren’t just increasing. They’re more likely to convert.

In an analysis of 2.3 billion sessions (January 2024 to December 2025):

• Traffic from generative AI grew 796% in two years.
• AI visitors converted approximately 1.2 times higher than organic search and at a higher rate than any other “free” channel.
• Organic and direct still dominate (63% of sessions), while AI accounts for 0.18%.

What this means for marketers:

• AI is changing when users arrive and how ready they are to act.
• Visitors from generative AI often come after researching options, comparing vendors, and narrowing their choices. This suggests they are more likely to take action when they land on a site.
• At the same time, traditional channels like organic search and direct still drive the majority of early discovery.

WebFX breaks down the data.

Note: This report was updated in March 2026 to reflect expanded data from January 2024 through December 2025. Earlier versions of this study (January 2024–February 2025) reported that generative AI traffic grew 165 times faster than organic search. The updated analysis extends the dataset and timeframe.

Generative AI has become a strategic traffic channel

By 2025, generative AI traffic was no longer behaving like a one-time spike. Generative AI grew approximately 796% from January 2024 to December 2025.

The quarterly growth pattern also shows how the channel evolved, explaining why it now deserves strategic attention. Growth in 2025 unfolded in three distinct phases: early adoption, acceleration, and maturation.

• Phase 1: Early adoption (January to April 2025). YoY growth ranged from 1,101% to 1,835%, driven by early adopters integrating generative AI platforms into research behavior alongside traditional search.
• Phase 2: Acceleration (May to July 2025). May reached a peak of 3,431% YoY, followed by elevated growth through July. This period reflects broader adoption and increased frequency of AI-assisted research.
• Phase 3: Maturation (August to December 2025). Growth moderated into the 260%–889% range. Session volume remained elevated, while the rate of increase stabilized into a more consistent pattern.

These numbers indicate the channel is maturing and stabilizing.

Traffic share remains small, but strategically meaningful

In 2025, generative AI accounted for 0.18% of total sessions. The share remains modest, yet its sustained growth and measurable conversion activity elevate its strategic relevance.

Organic Search still remains a primary traffic channel, though, holding the second-highest market share at 27.12% and trailing only Direct. Together, the two make up more than 60% of website traffic.

Traffic distribution across channels changed measurably in 2025, reflecting users’ evolving search and discovery behavior. When taken together, the quarterly growth pattern and traffic-share data show that generative AI is no longer an experimental referral traffic source. It is measurable, sustained, and tied to revenue activity.

Takeaways for marketers: Manage generative AI as a defined traffic channel

Generative AI should now be tracked, benchmarked, and forecasted like any other revenue channel.

Here’s what marketers should do.

Track AI referrals separately

In GA4, create a dedicated channel grouping or source filter for traffic from generative AI platforms so it does not merge into generic referral buckets. Doing so lets you accurately examine quarterly trends.

Monitor channel share alongside volume

Track AI’s percentage of total sessions alongside raw session growth to understand how your acquisition mix is changing. Monitoring traffic share tells you whether AI is becoming an important contributor to your pipeline or simply expanding from a small base.

Evaluate quality with scale

Session growth alone doesn’t tell you how important a channel is. Review conversion events per user and assisted conversion paths to measure generative AI’s revenue influence.

If AI-assisted sessions are high-quality, which means they lead to conversion actions, it may justify deeper content optimization or increased efforts to improve your visibility. If traffic quality is inconsistent, you may need to adjust your targeting or landing pages.

AI visitors are buyers, not browsers

From 2024 to 2025, sessions from generative AI platforms increased 796% YoY, while conversions increased by 6,432% YoY.

When conversions grow faster than sessions, it means a larger share of visitors are turning into leads, customers, or taking other meaningful actions. Generative AI traffic is not only expanding its reach but also improving conversion efficiency.

Across industries, users referred by generative AI consistently converted at higher rates than organic search throughout 2025. Industries like SaaS and Retail saw AI referrals convert at more than 50%, while organic search conversions were between 20% and 30%.

AI traffic had fewer sessions per user than organic search in both 2024 and 2025. In 2025, AI visitors averaged 1.14 sessions per user compared to 1.18 for organic search.

This pattern suggests less back-and-forth exploration. Many AI-referred visitors have already begun evaluating options elsewhere:

• Inside AI platforms
• Review sites
• Industry publications
• Community forums

When these users reach a company website, they’re confirming pricing, specifications, credibility, or contact information.

Generative AI traffic combines conversion efficiency with rapid growth

Generative AI delivered 0.79 tracked interactions per user. In practical terms, that’s roughly eight tracked interactions for every 10 visitors arriving from AI platforms.

For context, organic search generated approximately 12 tracked interactions per 10 visitors.

High-intent channels such as Affiliates and Paid Search generated even more interactions per visitor, which implies that visitors coming from these channels are in the earlier stages of their research.

Generative AI outperformed Direct, Organic Social, Referral, Paid Social, and Display in terms of tracked interactions per visitor. This places the generative AI channel in the middle tier of conversion efficiency — competitive but not the most efficient or highest-converting.

On its own, midtier efficiency is not unusual. What distinguishes generative AI is the combination of:

• Approximately eight interactions per 10 visitors
• 796% YoY session growth
• No direct media spend

No other unpaid channel grew this quickly while still driving meaningful conversion activity. This combination reflects a growing share of visitors arriving through AI platforms with meaningful conversion activity.

What marketers should do: Treat AI as a high-intent channel

Generative AI functions as a prequalification tool for prospects. For this reason, AI traffic behaves more like bottom-of-funnel traffic than early-stage discovery.

The data suggests several shifts in digital strategy.

AI as a decision-stage channel

Visitors arriving from AI platforms are often validating options rather than beginning research. Landing pages that clearly present key information—such as pricing, specifications, comparisons, and proof points—align with the verification behavior of these visitors.

AI-driven visitors are more likely to convert when information is immediate and structured.

Shifts in performance measurement

AI visitors averaged fewer sessions per user than organic search in both 2024 and 2025, yet generated several interactions with visitors. If you measure performance primarily on session depth or repeat visits, AI traffic may appear weaker than it is.

Benchmarking AI performance against high-intent channels rather than informational organic queries provides more accurate context.

Changes to reporting and attribution models

With 796% YoY session growth and meaningful interactions per user, AI is no longer experimental traffic. Tracking it as a defined channel in dashboards, revenue reporting, and forecasting models provides better visibility.

Tracking referral sources from AI platforms separately will prevent their impact from being absorbed into “referral” or “other” categories.

Content alignment with confirmation behavior

AI-driven visitors frequently arrive to confirm pricing, review technical details, or assess credibility. Landing pages that provide clear pricing and technical information, boost brand credibility with proof points, and guide visitors to next steps align with this behavior.

As AI visibility increases, the ability to appear in AI-generated responses directly influences which brands receive this decision-stage traffic.

AI compresses research and changes how users engage on-site

Generative AI accounted for just 0.18% of traffic in 2025. While small, it’s unique: What sets it apart from other traffic sources is how AI-referred visitors behave when they land on a business’s website.

In 2025, generative AI recorded a 66.48% engagement rate and a 54.15% session conversion rate. Organic search, by comparison, recorded a 70.86% engagement rate with a 45.23% session conversion rate during the same period.

Their difference shows up in how concentrated the visitors’ intent appears to be.

Organic-driven sessions include a variety of intents. Visitors land on a brand website to conduct early research, casual browsing, comparison shopping, fill out a form, or make a purchase.

On the other hand, generative AI sessions are more likely to include a measurable action. That’s why its session conversion rate is high (54.15%).

In practical terms, a higher percentage of AI-referred visits result in form submissions, resource downloads, quote requests, or other conversion events within the same session.

For marketers, that suggests something important: AI-referred users may have done some research before they click through your site. By the time they land on your site through an AI-assisted search, they’ve already learned so much about their options and are not starting from scratch.

This trend affects how you design high-intent experiences for AI-assisted visits.

Action: Optimize for decisive visitors across channels

While generative AI traffic accounts for only a small fraction today, the behaviors seen — higher session-level conversion activity — also apply to other high-intent visitors, whether they arrive via organic search, paid search, or direct.

The objective is to optimize websites so that when visitors arrive ready to act, the process is streamlined.

Making the next steps obvious and simple

When someone lands on a product or service page, the next steps should be immediately clear. High-conversion pages often share several characteristics

• Reasonable form lengths
• Nonredundant form fields
• Strategically placed calls to action (CTAs)

Adjusting messaging for returning visitors

Not every high-intent visitor converts on the first visit. Some return to confirm or compare pricing, so some organizations personalize content for returning visitors instead of repeating introductory messaging.

If someone has already viewed technical specifications, they likely don’t need a brand overview. Messaging can be adjusted by adding excerpts from case studies to provide reassurance.

Small personalization changes can support that momentum without requiring a full redesign.

Reinforcing credibility during the decision-making process

High-intent visitors — including AI-referred users — often concentrate on decision pages. Product, pricing, and demo pages often display social proof such as:

• Testimonials
• Industry certifications
• Clear deliverables

ChatGPT dominates generative AI discovery

From 2024 to 2025, ChatGPT accounted for 82.6% of all generative AI traffic. The next-closest platforms — including Perplexity and Google Gemini — accounted for much smaller shares.

When combined, the top three AI platforms generated 96.9% of all AI-driven visits. In other words, AI discovery is not spread across dozens of tools. Instead, most AI discovery happens on just a few platforms.

This concentration suggests that optimization principles remain consistent across the landscape, requiring authoritative content, clear explanations, structured information, and credible sources. While ChatGPT currently represents the largest share of AI answers, other platforms continue to play specific roles.

That doesn’t mean other platforms are irrelevant. Perplexity continues to serve research-heavy queries, and emerging assistants from Google and Microsoft are still evolving.

Pro tip for marketers: Maintain platform-agnostic optimization

Although traffic is concentrated, the foundations of AI visibility are largely universal.

AI platforms tend to reference authoritative content, such as original research, expert explanations, and clear answers to specific questions. Well-structured pages also assist crawlers in finding, extracting, and citing information. This suggests that building content robust enough for any AI system to rely on is more effective than creating tool-specific content.

Monitor emerging platforms without overinvesting

Perplexity, Gemini, and Copilot still contribute smaller shares of traffic today. As generative AI evolves as a channel, the distribution of traffic may change.

AI adoption accelerated across B2B industries

Generative AI traffic growth in 2025 was not confined to SaaS or technology companies. Adoption accelerated across research-intensive B2B sectors.

In this dataset, Manufacturing, Professional Services, and SaaS accounted for roughly 35% of generative AI traffic in 2025. These industries often require buyers to carefully compare options, validate capabilities, and align stakeholders before inquiring.

Manufacturing and Heavy Equipment showed sustained acceleration into late 2025, while Professional Services experienced an early-2025 surge followed by stabilization. As quarterly growth stabilized overall, these industries continued to see sustained increases in AI-referred sessions, showing us that technical buyers are incorporating AI tools into procurement workflows.

Home Services followed a different trajectory. AI traffic in this category moved from negligible volume in early 2024 to steady, conversion-producing streams by late 2025.

While total session share remained modest in Home Services, AI-assisted visits showed conversion activities, suggesting that AI platforms power vendor discovery and assist with initial outreach. Total session share in the SaaS and Software industry also appears small compared to other industries and is likely due to larger datasets coming from other B2B sectors.

B2B buyers are shortlisting vendors before they visit your website

B2B buyers increasingly use AI platforms to compare vendors, review specifications, and narrow options before visiting company websites. By the time they visit your website, they are confirming details, not starting their research.

If your specifications, service descriptions, or case studies are not surfaced in AI-assisted research, buyers may never discover or consider your business. That makes visibility during their early comparison critical — vendors mentioned at this stage have a chance of getting evaluated.

Strategies for B2B visibility in AI-assisted research

B2B buyers use AI platforms to gather, compare, and shortlist options before visiting vendors’ websites and inquiring. To get their attention at this stage, you must have structured, authoritative content.

Publish comparison-ready documentation

Make product specifications, service packages, compliance details, and pricing models easy to find and easy to interpret.

Front-load key information at the top of your pages. In addition, ensure product specs and key details are consistent across pages so buyers and crawlers can easily find and understand them.

Use structured data to reduce ambiguity

Structured data (or schema markup) won’t guarantee citations, but it helps crawlers extract and summarize your content accurately. For many B2B organizations, useful schema markups include:

• Organization (brand identity signals)
• Product or Service (offer details)
• Offer (pricing and packaging structure when applicable)
• FAQPage (common validation questions)
• BreadcrumbList (site structure)

Use the types that match what you actually publish to make important details clear.

Use consistent naming so you can be cited correctly

Keep product names, categories, and terminology consistent across pages. Doing so increases the likelihood that AI-generated summaries will reflect your correct offerings and details.

Earn trust with expert-backed, proof-focused content

B2B buyers look for credibility signals, while AI-powered searches look for statements that they can reference. When applicable, incorporate insights from subject-matter experts, case studies, and data-backed comparisons into your content.

For example, a manufacturing supplier can publish an engineer-reviewed specification table comparing material tolerances, performance metrics, and compliance standards across product lines, along with a case study.

By providing specific, technical details, you’re improving both buyer trust and AI interpretability.

Audit how your brand appears in AI answers

Regularly check how your B2B business appears for high-intent queries on major AI platforms. AI visibility tools can help monitor and analyze a brand’s presence on ChatGPT and other major AI search experiences.

How to optimize for AI visibility in 2026

Generative AI has not replaced traditional traffic channels, with direct and organic search still dominating with 35.51% and 27.12% of total sessions, respectively, in 2025. However, generative AI platforms are increasingly influencing how online users evaluate vendors and make purchase decisions.

This shift suggests there are different ways for audiences to discover brands and services. Appearing in traditional search results remains essential, but being mentioned in AI-generated answers is critical to getting noticed and shortlisted.

Here’s how.

1. Prioritize traffic quality along with volume

As earlier sections showed, the AI-referred visitors often arrive at websites ready to take action. Instead of focusing only on session growth, monitoring the quality of traffic arriving from different channels with metrics such as:

• Conversion events per user
• Assisted conversions
• Engagement patterns

These metrics reveal which channels drive revenue, helping you identify the optimization efforts to prioritize.

2. Track generative AI visibility as a distinct channel

Creating a separate reporting view for generative AI traffic in analytics platforms makes it easier to evaluate their influence. As AI platforms become a measurable source of discovery, isolating that traffic makes it easier to evaluate their influence.

Monitoring referral sources from major AI tools and comparing how those visits behave compared to other channels can reveal which pages, resources, and topics are most frequently surfaced in AI-generated responses.

Over time, this analysis can reveal which pages, resources, and topics are most frequently surfaced in AI-generated responses.

3. Align SEO and GEO through a “double-dip” strategy

Rather than treating generative engine optimization (GEO) as a separate initiative, it can be integrated with existing SEO strategies.

Search engines still capture a large share of discovery traffic, while AI platforms increasingly shape how buyers validate their options during evaluation. Having a strong content strategy can support both your SEO and GEO efforts.

A strong content strategy can support both. As research expands beyond traditional search, brands that get cited are those that consistently provide helpful answers backed by first-party data and experience across discovery channels.

SEO-focused content helps brands appear during early research. The same pages — when structured clearly and supported with credible information — can become sources that AI systems can cite when users ask deeper questions.

This “double-dip” approach allows a single piece of content to contribute to both discovery and decision stages of the buyer journey.

This story was produced by WebFX and reviewed and distributed by Stacker.

The post Inside AI traffic’s 796% growth, and why it converts more ready-to-buy visitors appeared first on The Media Copilot.

Artificial intelligence is doing something economists rarely see at once: shrinking employment in affected industries while pushing wages higher. New research from the Federal Reserve Bank of Dallas offers a possible explanation — and it has specific implications for anyone early in their career.

Scott Davis, an assistant vice president in the Dallas Fed's Research Department, analyzed employment and wage data across more than 200 occupations since ChatGPT's release in late 2022. The findings, published February 24, suggest AI is simultaneously replacing entry-level workers and making experienced workers more valuable.

The employment picture

Total U.S. employment has grown about 2.5 percent since fall 2022. Employment in AI-exposed sectors has not kept pace. The computer systems design sector — one of the most AI-exposed in the economy — has shed 5 percent of its workforce. Across the top 10 percent of AI-exposed industries broadly, employment is down 1 percent over the same period.

That decline is landing hardest on young workers. Research from Stanford University's Erik Brynjolfsson and colleagues finds the employment drop in AI-exposed sectors is concentrated among workers under 25. Employment totals for older workers have not declined. According to Dallas Fed economist Tyler Atkinson, the issue isn't layoffs — it's that young workers aren't finding jobs in the first place. The entry-level market in AI-exposed fields is getting much harder to break into, a trend that tracks with AI accounting for 25 percent of U.S. layoffs in March according to Challenger, Gray & Christmas.

The wage picture

Here's where it gets unusual. Despite the employment decline, wages in AI-exposed sectors are rising faster than the national average. Nominal average weekly wages across the economy grew 7.5 percent since fall 2022. In computer systems design, they grew 16.7 percent. Across the top 10 percent of AI-exposed industries, wage growth was 8.5 percent.

Davis found no meaningful relationship between AI exposure and wage growth across 205 occupations — until he added one variable: the experience premium.

The codified vs. tacit knowledge divide

Davis draws on a distinction between codified knowledge — the kind you learn from textbooks — and tacit knowledge, the kind you accumulate through years of practice. His hypothesis: AI can replicate codified knowledge but not tacit knowledge. That means AI substitutes for workers whose primary value is book learning, and complements workers whose value comes from hard-won experience.

Using Bureau of Labor Statistics wage data that separates entry-level and experienced worker pay, Davis calculated an experience premium for each of the 205 occupations. He then tested how AI exposure affected wages differently depending on that premium.

The results were clear. For occupations with a low experience premium — jobs where experienced workers don't earn much more than entry-level workers, like fast-food cooks or ticket agents — increased AI exposure was associated with lower wage growth. AI is substituting for everyone in those roles. For occupations with high experience premiums — lawyers, insurance underwriters, credit analysts, marketing specialists — increased AI exposure was associated with higher wage growth. AI is doing the entry-level work while making expert-level judgment more valuable.

What this means for newsrooms and media teams

The implications run directly through white-collar knowledge work, including journalism and media. The traditional career path — take an entry-level job, do the codifiable tasks, slowly build tacit knowledge — is precisely what Davis says firms are finding cost-ineffective to maintain. That dynamic is already visible in the 2026 journalism layoff wave, which has fallen disproportionately on junior and mid-level roles.

For experienced journalists, editors, and media professionals with deep domain knowledge, the data offer some reassurance. Their tacit knowledge — source relationships, news judgment, contextual understanding — is not easily replicated. For new graduates hoping to learn the craft on the job, the environment is harder. The entry-level work AI can do most easily is often the same work that used to teach people the fundamentals.

Davis doesn't suggest this is permanent. "Leaving new employees off the job ladder is not sustainable in the long run," he writes. AI adoption will require rethinking how entry-level employees develop on the job — but that rethinking hasn't happened yet.

The post AI is shrinking entry-level hiring while boosting pay for experienced workers, Dallas Fed finds appeared first on The Media Copilot.