🤖 The internet loses its human touch

🤌 Taste is the new moat

🗣️ Search's chatty overhaul

🐎 ChatGPT widens lead

🚫 Fighting the machines

🔐 Subscription shakeout

🚮 AI slop hijacks GEO

✋ Creative's red lines

The internet is undergoing a foundational shift as websites evolve from human destinations to information hubs for AI agents.

Why it matters: Human discretion served as a critical safeguard for establishing trust online. Now, it's rapidly disappearing.

The big picture: While AI platforms race to establish guardrails, bad actors are leveraging AI to build new information sources aimed at manipulating large language models.

Vetting those inputs requires time and resources that competitive pressure doesn't afford AI giants in the near term.

"There's a risk that you could lobotomize all the LLMs because there's just tons of synthetic data that no human is ever going to read," said Joe Marchese, an investor and author of "Human Dot AI," a forthcoming book about the AI attention arms race. 

State of play: Premium publishers and brands are looking to get ahead of this problem in the short term by becoming reliable information warehouses for bots while serving premium experiences for humans.

At Time, a news publisher that has leaned heavily into agentic optimization, human visitors encounter articles with imagery and storytelling while AI agents receive structured information, COO Mark Howard told Axios.

Time's Agent Ads product extends that approach to advertising, providing machine-readable information about sponsors, including FAQs.

Reality check: Smaller companies may struggle to handle an influx of bot traffic or build the infrastructure needed to serve both humans and machines.

Zoom out: Even if quality publishers build the infrastructure needed to remain relevant in today's agentic web, their ability to survive depends on a critical economics shift that isn't guaranteed.

Human discretion was essential to making the internet useful. Advertising was sold against attention metrics. Websites with spammy experiences were down-ranked in search results.

The agentic web has removed those incentives, and it's unlikely a licensing model will emerge to support most of the current ad-supported web. 

The bottom line: Few consumer websites will survive as direct destinations in the agentic era.

Without a trust layer created to vet information retrieved and analyzed by machines, the internet risks becoming a breeding ground for content that ultimately becomes useless to humans. 

1 big thing: 🤖 The internet loses its human touch

Illustration of a vintage toy robot with a crown or halo of glowing binary code over its head.

Illustration: Brendan Lynch/Axios. Stock: Getty Images

As some tech giants pour billions into building AI models, others are betting that knowing what people like is a valuable business in itself.

Why it matters: Companies with proprietary data about human preferences, interests and behavior could gain a competitive advantage.

"If you're a model developer, you desperately want that taste data layer because the model itself doesn't have any taste without it, so that data is like incredibly valuable to you," Reddit COO Jen Wong tells Axios. "I can imagine that there might be a very vibrant market for that information."

Zoom in: Companies like Pinterest, Spotify, Etsy, Reddit and DoorDash have positioned themselves as experts in human taste. These platforms have amassed years of data about what people shop, save and buy.

Reddit, Pinterest and Spotify argue that proprietary taste data becomes more valuable as AI models become commoditized. Etsy, DoorDash and others see it as a way to make shopping more personalized. 

As market momentum moves toward AI companies, they are selling those narratives to Wall Street.

By the numbers: The word "taste" appeared 22 times across Pinterest, Spotify and Etsy's earnings calls in the second quarter of 2026, up from just two in the first quarter of 2024, according to an Axios analysis of transcripts.

Between the lines: Understanding what someone wants, or intent, isn't the same as understanding taste.

AI models can identify what's in a piece of content or a product, like its size or color. But they struggle to understand why a user likes it or anticipate what else they might want. 

In a study of nearly 15,000 video placements, AI audience discovery company Chronicle Media found that similarity between a watched video and a new video was a poor predictor of whether those viewers would engage with the new video.

The big picture: The long-term value of proprietary taste data depends on people continuing to participate in the platforms that collect it.

The value of Reddit, for example, comes not just from its archive of conversations but from users continually posting, voting and commenting. 

Large language models are built to analyze human preferences, not develop them. 

The bottom line: "The AI by itself doesn't have style or taste. People have style and taste," Pinterest CEO Bill Ready said on the company's August earnings call.

2. 🤌 Taste is the new moat

Major retailers and publishers are adding generative AI tools to their online experiences, enabling stronger discovery and engagement.

Why it matters: Conversational search better reveals customer intent than traditional menu navigation and keyword search, providing companies with richer insights to improve their products.

State of play: Digital companies began experimenting with chatbots last year to retain audiences as AI search threatened traffic. Now, more companies are integrating the tech into their existing sites and experiences.

SeatGeek recently launched Ask SeatGeek. YouTube created Ask YouTube. Reddit folded Reddit Answers into its main search bar as adoption grew.

Between the lines: A cottage industry of AI startups like Daydream and Whim is emerging to build and license specialized conversational-search engines to brands so they don't have to build the tools themselves.

By the numbers: User adoption of AI search and chatbots has exploded, and companies leaning into generative AI for discovery report much stronger customer engagement and conversion rates.

Alexa for Shopping drove U.S. customers to spend over 40% more per order on average, the company said in July.

Taboola said monthly active users of DeeperDive, its AI search product for publishing partners, grew from nearly 7 million in April to more than 11 million by late September. 

Online readers of USA Today, one of the largest newspapers in the country, asked more than 50 million questions using DeeperDive between last September and July, per USA Today Co. CEO Mike Reed.

Yes, but: Companies like Instacart and Target are also preparing for a future in which consumers begin and sometimes end their journeys on someone else's AI interface.

Others, such as SeatGeek, Tubi, DoorDash and Zillow, use ChatGPT as a discovery channel and then direct users to their own platforms to complete tasks. 

Spotify splits the difference, allowing users to connect their accounts to ChatGPT for personalized recommendations and playlists, while also making audio previews and some full podcast episodes available for listening on the AI platform.

The intrigue: Reddit is taking a more cautious approach. Wong says Reddit hasn't yet determined what a mutually beneficial integration would look like given that users' participation is central to the platform's value.

3. 🗣️ Search's chatty overhaul

Illustration of a 1990s desktop computer with a sparkle and rows of binary code emanating from the monitor.

Data: Similarweb; Chart: Erin Davis/Axios

ChatGPT's web traffic rebounded over the past three months after a volatile first half of 2026, according to Similarweb data.

Why it matters: ChatGPT remains the leading AI chatbot by website visits even as competitors erode its market share.

The summer rebound coincided with Google's Gemini seeing its first traffic slowdown in nearly a year and a half.

Zoom in: September was ChatGPT's strongest month since October 2025 in terms of visits, per Similarweb.

As of last month, ChatGPT accounted for 57% of visits to eight of the world's largest AI platforms tracked by Similarweb.

Similarweb's analysis does not include visits to Google's AI Overviews, which are difficult to isolate from traditional searches.

ChatGPT's share of visits has risen from a low of 53% in June but remains below its 85.1% peak in November 2024 and its 2026 high of nearly 64% in January.

The intrigue: It's unclear why visits surged for ChatGPT in September, while Gemini and Claude did not see similar gains. Back-to-school season could be a factor, considering Claude's largely enterprise client user base.

ChatGPT's dip earlier this year coincided with major marketing and product expansion efforts from Anthropic and Google.

Data: Axios research; Chart: Danielle Alberti/Axios

Zoom out: While ChatGPT's website visits have fluctuated, app engagement remains strong.

In June, the ChatGPT app made history when it became the fastest-ever consumer app to reach 1 billion monthly active users. 

According to Sensor Tower, ChatGPT was the most downloaded app in the world in 2025.

Yes, but: ChatGPT's downloads have slowed relative to competitors as it's matured, per Sensor Tower, even as app engagement remains strong, per Similarweb.

Claude has significantly outperformed peers in app downloads in 2026, per Sensor Tower. Other apps that have shown strong year-over-year gains include Gemini and Meta AI. 

What to watch: While it's too soon to draw conclusions from preliminary data on Meta's personal AI agent Muse, which debuted in early September, reports suggest early user adoption has been strong.

4. 🐎 ChatGPT widens lead

Data: Axios research; Note: "Mixed" includes sites that block/allow only part of their site and/or only select bots; Chart: Erin Davis/Axios

More than 40% of top U.S. publishers block AI training crawlers, compared with just 25% that bar AI search crawlers, according to an Axios analysis.

Why it matters: The data suggests publishers want their content to be discovered by generative AI search interfaces without giving it away to train the models powering them.

How it works: Publishers use robots.txt files to tell crawlers which parts of their website they can access.

Robots.txt files have long been used to manage search engine crawlers, but AI content scraping requires a much more nuanced playbook.

An evaluation of what those files allow today shows which types of data publishers are most focused on restricting in the AI era. 

By the numbers: According to an Axios analysis of robots.txt files across 102 top U.S. publisher domains, 31% apply partial restrictions to AI training crawlers, while 39% do so for AI search crawlers.

Search crawlers face far fewer outright blocks than those used to train large language models.

For example, the analysis found that OpenAI's training crawler is more than three times as likely to be completely blocked as its AI search crawler.

Anthropic's Claude training crawler is around 2.5 times more likely to be completely blocked compared with its search counterpart. 

Search crawlers, which scrape websites to answer user queries, typically aren't blocked outright. The vast majority of publishers partially restrict OpenAI's search bot (78%) and Claude's search bot (73%).

Between the lines: Publishers are more likely to single out training crawlers from certain AI companies over others.

OpenAI's ChatGPT training bot is explicitly named in 56 publishers' robots.txt policies, compared with 41 for the company's search bot.

A quarter of top publishers don't explicitly address AI training in their robots.txt files, compared with 35% that don't for AI search.

The intrigue: Nearly half (46%) of publishers completely block Anthropic's ClaudeBot, compared with 38% that block OpenAI's GPTBot.

While OpenAI has struck dozens of publisher licensing deals, Anthropic hasn't publicly announced any, per an Axios analysis.

Yes, but: Setting restrictions doesn't guarantee that AI companies will respect them.

Ziff Davis accused OpenAI in a copyright lawsuit of scraping its publications despite restrictions in robots.txt files.

A federal judge dismissed Ziff Davis's claim that OpenAI violated the DMCA's anti-circumvention provision while allowing other copyright-related claims to proceed.

The big picture: Search crawlers allow AI firms to retrieve data to power real-time answers to user queries. This process, called retrieval augmented generation, is what most recent licensing deals are focused on, rather than training.

For publishers with paywalled content, search crawlers are often restricted to content that lives in front of the paywall. 

This is why many search crawlers are partially restricted, instead of fully allowed. 

What to watch: Publishers are seeking alternatives to manage crawlers at the cloud level.

Cloudflare introduced tools to help website owners manage AI crawlers and explore pay-per-crawl arrangements.

The cloud provider recently introduced a setting that lets publishers block AI training while keeping their content available to be scraped for search results. 

5. 🚫 Fighting the machines

Brands have spent the past decade making it easier for consumers to buy subscriptions. Now, AI is forcing them to give customers a reason to stay.

Why it matters: AI assistants will remove cancellation friction and increase churn, forcing publishers, retailers and brands to consider new ways to drive value, including bundles and product improvements.

State of play: Wall Street is bracing for a cancellation tsunami in the wake of Meta's Muse AI agent launching last month.

Unlike most agents, which are built to help users with content discovery, Muse is designed to help consumers manage tasks, payments and logistics across accounts and devices.

That could make it easier to remind consumers about recurring charges or subscriptions they don't use.

By the numbers: Shares of several subscription-dependent companies plunged following the launch of Muse, Axios' Matt Phillips wrote.

Some stocks have since recovered, but others remain well below their pre-launch levels.

Data: Financial Modeling Prep; Chart: Erin Davis/Axios

Between the lines: Media publishers already facing declining search referrals could see another source of revenue threatened.

Publishers are leaning into bundles, family plans and more perks to get subscribers to stick around.

Yes, but: Subscription companies that demonstrate value and already make it easy to cancel are less likely to be impacted.

"The tools that help consumers shed subscriptions will work best against forgotten, low-dollar, low-engagement autopays that survive on save offers, automated retention tactics, and phone-tree gauntlets," Jefferies analyst Randal Konik remarked last month.

The other side: More subscriptions could become accessible through AI assistants, giving publishers another way to reach customers without requiring them to visit their websites, says Jodie Hopperton, who leads the International News Media Association's product and technology initiative.

The bottom line: The digital subscription economy has boomed over the past decade, thanks in large part to consumer inertia.

Now, AI is poised to wreak havoc on that model, but it may also force subscription companies to improve their offerings or offer consumers more choice. 

6. 🔐 Subscription shakeout

Illustration of a large arrow cursor behind a wall of glowing binary code.

Data: DoubleVerify; Chart: Erin Davis/Axios

AI "slop" sites, or domains with low-quality AI-generated content, rose more than fivefold in the second quarter of this year compared with the same period a year prior, according to new data from ad measurement firm DoubleVerify. Traffic to those types of sites quadrupled in that time.

Why it matters: SEO's incentive problems have been exacerbated in the AI answer engine era with brands and publishers competing to influence what chatbots recommend.

As AI citations become more valuable, companies have an incentive to optimize content for machine consumption, which low-cost content mills could exploit.

By the numbers: Ad requests associated with low-quality AI-generated content surged 354% between January and September, per DoubleVerify.

There were more than 3.2 billion ad requests tied to AI slop webpages in the first nine months of 2026.

Zoom in: Generative AI is making it easier to operate content farms designed to capture advertising revenue.

DoubleVerify identified a network of mothan 200 made-for-advertising websites filled with AI-generated articles. Some pages featured more than 50 slides, each packed with as many as eight ad banners that refreshed every few seconds, CMO Dan Slivjanovski tells Axios. 

The surge in activity toward the end of the year resembles historical patterns of advertising fraud as operators seek to capitalize on rising holiday ad spending, he adds. 

Zoom out: AI platforms are moving to establish foundational infrastructure to prevent spam, scams and misinformation from flooding their chatbots.

Google has teamed with Shopify and a slew of online retailers to develop a Universal Commerce Protocol standard for AI shopping.

OpenAI and Stripe have built an Agentic Commerce Protocol to secure checkouts.

Microsoft has piloted a premium content marketplace for publishers. Amazon is eyeing a similar effort.

Yes, but: Those partnerships may establish guardrails for large portions of the transactional web, but the universe of content created to influence large language models is growing rapidly as bad actors use AI to manipulate consumers.

The big picture: The rush to appear in AI answers risks re-creating the content arms race that plagued traditional search.

Between the lines: Brands must optimize for multiple AI models that weigh sources differently.

Appearing more frequently also doesn't equate to appearing more favorably.

What to watch: Medium offers one warning sign.

CEO Tony Stubblebine says roughly half of the 25 Medium stories most frequently surfaced in Google's AI search summaries had been excluded from Medium's own recommendation algorithm. 

"That stuff is like the most generic content mill things you could imagine," Stubblebine tells Axios.

7. 🚮 AI slop hijacks GEO

The distinction between AI-assisted and AI-made is becoming harder to define as using the technology becomes a routine part of making content.

Why it matters: Creators, platforms and consumers are drawing different lines around how much AI involvement is acceptable and when it should be disclosed.

What they're saying: YouTube CEO Neal Mohan said last month that the platform doesn't view AI-generated content as inherently low quality.

"AI versus not is kind of an artificial sort of distinction," he added. "Every video may have AI in it, just like almost every piece of music has something synthetic in it today versus 100 years ago when it didn't."

Zoom in: Music platforms have started to formalize how much AI involvement warrants a label. YouTube Music announced this week it's rolling out labels for songs created with generative AI.

Under YouTube's guidelines, songs combining AI lyrics with human performances fall into the partly AI-generated category, while pitch correction like auto-tune doesn't count as generative AI use.

Zoom out: For other formats, platforms have been developing different rules.

YouTube made labels more prominent in May for "photorealistic and meaningfully AI altered or generated content." Routine production assistance, such as AI-generated scripts or captions, is exempt.

TikTok requires labels for realistic AI-generated images, audio and video, including significant alterations to someone's likeness or speech.

Meta requires disclosure of photorealistic AI-generated video and realistic synthetic audio. It moved labels for content modified or edited with AI into posts' menus after criticism that minor edits were being treated like fully AI-generated content. 

State of play: Some creators are keeping AI behind the scenes. Dhar Mann told Axios at Cannes that his team experimented with AI-generated effects in videos but received audience backlash.

Filmmaker Zack London is preparing to release "Gods Don't Give Gifts," a feature film with AI-generated visuals but human voice actors, musicians and a creative team of roughly 25 people.

London says it's inaccurate to describe the film as fully AI generated because humans make every creative decision under his direction. He considers it AI assisted.

8. ✋ Creative's red lines

Illustration of a feather quill and inkpot. A row of glowing binary code is pushing through the feather, and drops of ink are next to the quill's point.

Axios

Axios Media Trends Executive

October 17, 2026

1 big thing: 🤖 The internet loses its human touch