Fake accounts are as old as social media itself. So when it came to light that a “hot girl” MAGA personality named Emily Hart was actually a 22-year-old male medical student in India, it could have been dismissed as just another internet deception story. Just another catfisher, another sock puppet, another scammer—the internet is full of them.

But this case was different. This one had photos. And videos. And thousands of followers across multiple networks with some posts getting millions of views. Emily Hart was a full-on influencer, not just some anonymous egg. The person who created Emily confessed to Wired that while the account was active, he was making thousands of dollars every month from posting softcore videos to an OnlyFans competitor and merchandising.

Emily’s creator is not a developer. He’s just a cash-strapped student with a good sense of American political culture and a Google Gemini account. Yet the Emily Hart story has done more than expose one fraud. It’s put a spotlight on how thoroughly AI has lowered the barrier for almost anyone to produce convincing content and manipulate social media’s engagement systems.

That reality raises a set of urgent questions. Is anyone looking out for us out there? How can you tell what’s real and what’s not anymore? And who is responsible for alerting social media users that the images they’re looking at might have come from AI?

How cheap AI tools made fake influencers scalable

The real significance of the Emily Hart story has little to do with a single fake account. The major implication is that this is the tip of the iceberg. AI has made creating online personas like Emily so easy that it’s enabled deception at scale. The Wired story points to other pro-Trump fake influencers like Jessica Foster, but you don’t have to look very far in your Instagram Explore page before you spot something AI-generated, and it’s rarely disclosed. The Emily Hart case proves that the template is cheap, fast, lucrative, and easy to copy.

Every major social network has policies that address AI-generated content. While they vary in detail, the gist is generally the same: Synthetic images must be disclosed—especially if it could be construed as real and the subject matter involves sensitive subjects like politics, health, finance, and current news. If the account doesn’t identify AI content, it could be frozen, demonetized, or banned.

In practice, those consequences almost never materialize. Enforcement is difficult, partly because detecting AI content is getting more difficult by the day. Most state-of-the-art image generators are light-years ahead of the models that created the first “Will Smith eating spaghetti” video, and telltale artifacts like extra fingers and disappearing background characters have largely become a thing of the past. Without watermarks, even automated systems have a difficult time parsing AI images from real ones just by looking at them.

Content Credentials and the AI labeling problem

A new standard was supposed to fix this. Content Credentials are a way to track how an image was created and modified throughout its life cycle. That provenance data can live in the image’s metadata, so the site displaying it can more easily tell whether it’s AI-generated, potentially passing on a label or warning to the user. The idea is that, as you scroll your social feed, any image would have a tiny icon next to it that would reveal its history when clicked.

However, even though this technology has existed for years and ostensibly has the support of major tech companies such as Adobe, Google, and Nvidia, social platforms haven’t adopted it consistently. Seeing the label is rare, and a Washington Post report found that social networks often strip out the metadata that enables Content Credentials. The stripping isn’t necessarily a deliberate act of sabotage — it follows a best practice from the early days of the web when every byte was precious. But the fact that it’s still happening shows there is little enthusiasm to make the system work.

Does labeling even change behavior? Emily’s creator says he believes many of his followers didn’t care whether the images he was posting were AI or not. That may be true for some, but data suggest labels can alter people’s propensity to engage with AI content. A 2024 study found that labels on AI-manipulated media reduced belief in the claims. The study also found that wording matters: “manipulated” or “false” were more impactful than process-based labels alone.

Put another way: labels work, but toothless labels work poorly. A buried “AI info” tag is not the same as a clear warning that an image might depict a person who does not exist.

The technical capacity to do better clearly exists. Platforms like Facebook, Instagram, YouTube, and TikTok already process and modify content at scale. They’ve spent two decades building the art of detecting copyright violations, nudity, spam, and engagement signals. It is hard to believe they are incapable of building a clearer label for AI-generated people.

Why platforms have reason to keep AI labels weak

The question then becomes: why haven’t they? The uncomfortable answer is that the incentives point the other way. While platforms want to keep bad content out, they are more motivated to keep people posting, scrolling, sharing, and buying. AI-generated material fits neatly into that machine because it is cheap to make, easy to personalize and highly compatible with engagement-driven feeds.

Mark Zuckerberg has been unusually direct about this, describing AI-generated material as “a whole new category of content” that he sees as important for Facebook, Instagram and Threads. That framing doesn’t signal that Meta or any other platform actively wants deception — deception is a subcategory of AI content, not the whole thing. But it does mean the companies have a business reason to welcome more synthetic content, and making the labels too strong or too visible could dampen the engagement they’re trying to encourage.

External pressure could shift the math, though. Europe’s AI Act includes transparency obligations for deepfakes and certain AI-generated public-interest content, with related rules taking effect this year. Should platforms start to rack up major fines for poor labeling, things could change in a hurry. Advertiser pressure would help, too, since appearing next to deceptive content is bad for business. Finally, and crucially, there’s audience behavior: if users begin to feel like they can’t trust what they’re seeing on a network, they might, over time, stop engaging with that network.

The disclosure system failure

At the moment, detecting AI content has become largely the user’s problem, with social platforms not prioritizing the technical progress that might help, and regulators only beginning to act. And you might question what’s the point — many of Emily’s followers no doubt knew she was virtual but followed, engaged, and maybe even forked over some money anyway. But that calculus depends entirely on having information. The choice to engage or not with a virtual influencer is robbed from you if you don’t know it’s virtual in the first place.

The technology industry has spent years presenting provenance as a central answer to synthetic media. Adobe, Microsoft, Meta, OpenAI, Google and others have backed standards, joined coalitions, made public commitments and embedded Content Credentials into their tools. Fine. Then show it to people. Make it visible before the share, before the follow, before the subscription, before the merchandise purchase. Because if the only way to learn that an influencer is fake is to wait for a magazine investigation, the disclosure system has already failed.

A version of this column appears in Fast Company.

The post Why AI content labels keep failing the people who need them most appeared first on The Media Copilot.

Neil Steinberg of the Chicago Sun-Times has made a habit of this. Every February, he asks the latest version of Google’s Gemini to write his column — same prompt, new model — and publishes the results. The 2026 edition is worth reading, not because the AI failed, but because it mostly didn’t.

Gemini 3.0 produced a strong headline (“The Ghost in the Machine is Just Us”) and opened with a scene: Steinberg on the Red Line, watching a young man ask AI to write a poem for his girlfriend. The prose was confident, the tone matched, and the argument landed. The problem? None of it happened. Steinberg never got on the Red Line. There was no young man. The scene was invented — and Gemini delivered it as casually as everything else.

Steinberg asks the actual question plainly: “Will we continue to care if things are true anymore? Must the news have actually happened?”

That’s the question that matters for journalism. The voice problem is largely solved. Gemini 3.0 writes competently in the style of a working print columnist — casual, self-deprecating, city-specific. The fabrication problem is not solved. And fabrication delivered in a convincing voice, at scale, is a categorically different threat than the clumsy AI slop of two years ago.

Steinberg notes, correctly, that AI didn’t create the American appetite for comfortable fiction over inconvenient fact. “We didn’t need AI to undermine the value of truth. Look at who we picked to run the country. Twice.” But AI industrializes the production of plausible-sounding nonsense — and his annual test is a useful benchmark for how fast that industrialization is moving.

For newsrooms, the lesson isn’t that AI can write columns. It’s that the output is getting harder to distinguish from the real thing while the underlying problem — hallucinated specifics presented as lived experience — remains unchanged. The voice got better. The ghost is still lying.

The post A columnist asked Gemini to write his column. The result was good, and that’s the problem appeared first on The Media Copilot.

Apple has chosen Google’s Gemini AI to power the next generation of Siri after months of weighing options from OpenAI and Anthropic. The multiyear deal, announced Monday, positions both companies to fend off competition from fast-growing AI startups.

The partnership lets Apple use Gemini and Google’s cloud technology to build what the company calls more personalized and agentic versions of its assistant. Apple Intelligence features will still run on Apple devices and its Private Cloud Compute servers, both companies emphasized.

Apple reportedly will pay Google around $1 billion per year for the arrangement, according to Bloomberg. That’s a fraction of the up to $20 billion Google paid Apple annually to remain the default search engine on iPhones.

The relatively modest price tag signals how much both sides gain. Apple gets AI technology it failed to build in-house after a year of embarrassing stumbles. Google gets distribution to hundreds of millions of iPhone users and cements its position as a go-to AI provider.

“From Apple’s perspective, it’s certainly a win if you think about the pain that they’ve had in their AI strategy up to this point,” William Kerwin, senior equity analyst at Morningstar, told The Verge. “They over-promised back in the summer of 2024, and they under-delivered.”

Apple’s AI troubles mounted throughout 2025. Its Apple Intelligence message summaries produced errors. Promised Siri features never arrived despite TV ads promoting them. The company replaced longtime AI chief John Giannandrea with Vision Pro leader Mike Rockwell.

The deal could invite the same antitrust scrutiny Google just survived over its search payments to Apple. A federal judge ruled last fall that Google could continue those payments, clearing the path for this AI arrangement.

But legal experts say regulators may still come knocking.

“It could be that the market could evolve in a way that would make a deal like this more problematic over time,” James Grimmelmann, a professor of digital and information law at Cornell Tech, told The Verge.

The partnership represents a defensive alliance between two companies threatened by AI upstarts like OpenAI and Anthropic. Neither wants to cede ground to startups that could disrupt their longtime dominance.

“Apple is concerned that the rise of AI threatens to go completely around it,” Grimmelmann said. “This is its attempt to remain in that relationship and remain relevant.”

The upgraded Siri is expected to launch sometime in 2026.

The post Siri-Gemini partnership lets Apple, Google fend off upstart rivals appeared first on The Media Copilot.

Over the past year, Google has pushed its models back to the top tier, improved the pace and confidence of AI product deployment, and emerged from its search monopoly loss with remedies that leave its core distribution assets largely intact.

It is now putting in place contractual agreements that it hopes will secure ongoing access to content to fuel its AI future. For media businesses, Google’s AI recovery means the dominant distribution platform in this era is starting to look like it might be a continuation of the dominant distribution platform in the last.

Early stumbles, public fumbles

A year ago, this didn’t seem possible. A series of high-profile missteps in 2024 eroded confidence just as the market was forming an impression of how AI might change the platform landscape. 

Gemini’s image tool produced historically inaccurate depictions (including racially diverse “Founding Fathers”), forcing Google into a public apology. And the early rollout of AI Overviews turned into a joke, with widely shared examples of bizarre and unsafe answers that pushed Google to narrow and refine when Overviews would appear. More worryingly, these AI experiments were undermining Google’s core brand promises: trust and accuracy. Instead of looking deliberate and authoritative, Google appeared rushed, giving OpenAI space to define itself as the default.

Over the last 12 months, though, it’s regained its footing impressively. And we’ve now got a clearer sense of how Google intends to win. Given both its rate of improvement and its structural advantages, it would be brave to bet against it.

On the models, whilst these are rapidly becoming commodities with little to separate the top performers, Google has demonstrated the depth of its AI research capabilities. The release of Gemini 2.5 closed much of the perceived gap on reasoning and reliability, whilst Gemini 3’s launch late in 2025 was strong enough on benchmarks to trigger a public “code red” response from OpenAI.

On image generation—even after some years of consumers having access to these technologies—the launch of Google’s Nano Banana Pro sparked a fresh wave of experimentation, flooding social platforms with genuinely novel outputs and reminding the market that technical breakthroughs can still happen and still matter.

Sitting behind all this, Google is also developing its own chips. Its in-house program to build TPUs (Tensor Processing Units), which has been active for over a decade, gives it more control over cost and capacity than competitors reliant on third-party infrastructure. This will make it easier to train models and deploy them at scale across its products.

Distribution is destiny

But its key advantage is not derived from compute. Unlike OpenAI, which needs to create new products and then drive adoption, Google already has consumer relationships at scale and understands their habits from decades of data and experience. This means usage is a function of product integrations and pulling these formidable distribution levers.

This is where its search antitrust case is important. Whilst it lost in the first phase, and was deemed an illegal monopolist, the remedies were weak, with the judge allowing its search distribution deal with Apple to continue and opting not to force the company to spin out or divest its Chrome browser.

This legal outcome is likely to embolden Google to use Android, Chrome and of course search to drive uptake of its AI services. Whilst ChatGPT holds a dominant position in the consumer chatbot market at the moment, that could change overnight if Google chose to hard-wire Gemini into these products. It is already moving in this direction with plans to complete the replacement of Android’s Google Assistant with Gemini next year.

Google holds another advantage that no other AI lab can match: access to content at scale. Through Search it continues to ingest and index the open web. Whilst website owners are blocking other AI crawlers, Google can scrape unrestricted, with publishers facing severe consequences if they restrict it.

On top of this sits a growing layer of contractual access. Showcase-style agreements, framed as partnerships around peripheral products, function in practice as broad licences that secure ongoing rights while insulating Google from legal and political risk. The result is that Google’s AI systems are fuelled by a combination of formal deals and structural compulsion. Publishers may be able to opt out at the margins, but in aggregate—and absent regulatory action—they remain locked into supplying the inputs that power the next generation of AI products.

Content without consent?

Looking at this landscape at the end of 2025, Google sits in a position of renewed strength. It has harnessed its resources to get model and product development back on track and secured the inputs required to scale AI, all while emerging from regulatory scrutiny with its structural advantages largely intact.

The risk for publishers is that there’s a strong chance the story of AI will end exactly where the last era did: The same single company controlling the routes to audiences, setting the terms of access, and offering commercial arrangements that buy off the pain without shifting the balance of power.

For a moment AI looked as if it was going to open the content discovery market. Instead, it is increasingly being layered onto the same distribution infrastructure that shaped the last two decades of digital media. 

The post Google’s AI recovery: The comeback of the content gatekeeper appeared first on The Media Copilot.

OpenAI released GPT-5.2 on Thursday, just weeks after declaring an internal “code red” in response to Google’s Gemini 3 topping industry benchmarks.

The company says GPT-5.2 Thinking beats or ties top industry professionals on 70.9% of tasks across 44 occupations, according to expert human judges. Those tasks include creating presentations, spreadsheets, legal briefs, and engineering blueprints. The model completed them at more than 11 times the speed and less than 1% the cost of human experts.

“We designed GPT‑5.2 to unlock even more economic value for people,” OpenAI wrote in its announcement. “It’s better at creating spreadsheets, building presentations, writing code, perceiving images, understanding long contexts, using tools, and handling complex, multi-step projects.”

The model also hallucinates less. OpenAI says responses with errors dropped 30% compared to GPT-5.1.

Fidji Simo, OpenAI’s CEO for applications, denied on a press call that GPT-5.2 was a response to Gemini 3, according to Axios. But the competitive pressure is real. Google’s Gemini app has grown to more than 650 million monthly active users, compared to OpenAI’s 800 million weekly active users.

Safety improvements are part of this release. OpenAI says GPT-5.2 produces fewer undesirable responses when users show signs of mental health distress or emotional reliance on the model. The company faces multiple wrongful death lawsuits over troubling conversations users had with ChatGPT.

GPT-5.2 comes in three versions: Instant for fast everyday tasks, Thinking for deeper work like coding and analysis, and Pro for difficult questions. It’s rolling out now to paid ChatGPT plans and is available in the API.

Why it matters for newsrooms: The claimed improvements in spreadsheets, presentations, and long-context understanding could make GPT-5.2 more useful for journalists handling data analysis or working with lengthy documents. But Anthropic’s Opus 4.5 still scores higher on SWE-Bench Verified, the software coding benchmark, suggesting the AI race remains wide open.

The post OpenAI releases GPT-5.2 following internal ‘code red’ scramble appeared first on The Media Copilot.