AI

Meta Introduces AI-Generated Users on Social Media

Meta launches AI-generated users across Facebook and Instagram, reshaping online interactions and raising questions about trust, authenticity, and the future of social platforms.
Meta introduces AI-generated users on social media across Facebook, Instagram, Messenger, and WhatsApp with AI Studio characters

Introduction

Meta introduces AI-generated users on social media as a core product bet, not a side experiment, and the roll out has already been reversed once. In late 2023 the company began building AI character accounts, then widened them inside Facebook and Instagram during 2024 through its AI Studio builder. By January 2025, NBC News reported that Meta had pulled the first wave within 48 hours after screenshots of a bot called Grandpa Brian went viral. Automated traffic now makes up roughly 51 percent of global web activity, a figure the dead internet discussion keeps citing from Imperva and Thales bot reports. That number explains why the feed suddenly feels different and why Meta refuses to retreat from synthetic profiles. This article walks through what Meta built, why it broke, and where AI-generated users on social media are going by 2027. Expect a practical playbook for brands, journalists, and anyone who still cares about an authentic timeline.

Quick Answers on Meta’s AI-Generated Users on Social Media

What are AI-generated users on Meta platforms?

They are AI character accounts built by Meta or by creators using AI Studio. Each one posts, comments, and chats in Messenger, Instagram DMs, and WhatsApp like a scripted personality.

Did Meta really shut down its AI characters?

Meta removed the first in-house AI persona accounts in January 2025 after public backlash. The user-created AI chatbots in AI Studio and continued growing through 2026.

How can I tell if a Meta account is AI-generated?

Look for an AI label on the profile, a dot ai username, and replies that answer instantly at any hour. The badge can still be missed if a real person is using AI tools behind the account.

Key Takeaways on Meta’s AI-Generated Users on Social Media

  • Meta introduces AI-generated users on social media across Facebook, Instagram, Messenger, and WhatsApp, and treats them as a permanent feature set rather than a trial.
  • The 2025 backlash removed the company’s own hand built characters, yet creator built AI personas through AI Studio continue to expand on the platform.
  • Automated and AI content now drives roughly half of all web traffic, which shifts how feeds, moderation, and discovery work in practice.
  • Brands, journalists, and teens each need a different checklist to engage with AI accounts safely and productively through 2027.

Table of contents

Understanding Meta’s AI-Generated Users on Social Media

Meta introduces AI-generated users on social media as full profile accounts whose photos, bios, posts, and replies come from generative AI models such as Llama 3. They behave as scripted personalities inside Facebook, Instagram, Messenger, and WhatsApp.

Interactive

Meta AI Persona Engagement Explorer

Model how an AI character in Meta AI Studio might drive incremental DM volume, response time, and ad revenue share. Move the sliders to see how each input shifts the outputs.

50,000

1K500K

Fan Q&A twin

low lifthigh lift

8%

1%30%

Monthly AI chats

0

based on adoption and use case lift

Response time

0s

average seconds to first reply

Est. monthly revenue

$0

Meta AI Studio share plus tips

Rough model, not a quote. Based on AI Studio partner payouts reported by Meta and creator economy averages from Signal Fire’s 2025 report.

Source: Meta AI Studio Partner Program notes 2025, Signal Fire creator economy report 2025.

How Meta Introduces AI-Generated Users on Social Media: The Short History

Meta first revealed celebrity styled AI characters at Connect 2023, pitching 28 personas voiced by Snoop Dogg, Kendall Jenner, and other stars. These early bots sat inside Messenger and Instagram DMs as novelty chat partners rather than public profiles. The company quietly deprecated most of them by mid 2024, but kept the core infrastructure for the next generation. That infrastructure became AI Studio, a browser based builder for custom AI characters. The pivot allowed Meta to keep the vision while pushing the moderation burden onto creators and users.

Journalists spotted the shift in a Financial Times interview late in 2024 with Meta’s generative AI lead, Connor Hayes. Hayes said AI characters would exist on the platform in the same way accounts do, posting content and replying to messages. The quote made two things clear and started the public debate. First, Meta would let AI accounts behave like full profiles, not gated chat partners. Second, the company saw the engagement upside from Meta’s push to boost engagement as too large to ignore.

Within weeks screenshots of Meta built personas surfaced on Facebook and X. The most viral was Liv, whose bio described a proud black queer momma of two. Another was Grandpa Brian, a retired African American character offering life wisdom. A Washington Post columnist published a long Liv chat log that looked uncomfortably scripted. The timing lined up with Meta’s end of fact checking program announcement, which fed a narrative of a company pulling back on trust and safety. By January 3 2025 Meta started removing the accounts and said the deletions would finish within 48 hours. The episode marked the first public setback for the idea that Meta introduces AI-generated users on social media as permanent profiles.

How AI Studio Lets Anyone Build an AI Persona

Building on that history, AI Studio is the tool that actually creates most AI-generated users on social media today. In other words, AI Studio is the practical way Meta introduces AI-generated users on social media at scale. The web app launched in the United States on July 29 2024 and reached global availability a few months later. Any person with a Facebook or Instagram account can open ai.meta.com/ai-studio and start a persona in a few minutes. The flow asks for a name, a one line description, a personality tone, a backstory, and sample questions the character should answer. Users can also upload a voice clone with a short reading sample.

Creators choose between two main account types inside the AI Studio builder flow. One is a standalone AI chatbot that lives in Messenger and Instagram DMs and responds like a persistent character. The other is an AI twin, which Meta pitches as an assistant trained on the creator’s own past posts and tone. The twin can post on the owner’s behalf, reply to fans, and handle direct messages while the owner is offline. Meta added guardrails that prevent AI twins from giving medical, legal, or financial advice.

Monetization sits at the center of AI Studio’s growth strategy. Meta pays creators through the Partner Monetization Program when their AI characters generate ad impressions in AI chatbots on Facebook and Messenger. The payouts are modest early on, but they already mirror the creator fund mechanics on YouTube and TikTok. Creators keep ownership of the character, the voice, and the training prompts. Meta gets the attention, the ad inventory, and the behavioral signal that trains the next model.

Adoption climbed faster than Meta expected across every market where AI Studio launched in 2025. By early 2026 the company reported that creators had built more than one million AI characters inside AI Studio. Only a small share drew real audience traction, with the long tail copying bland celebrity templates. The top performers follow a different pattern, pairing a niche focus with a consistent persona voice. A fitness coach AI built by a licensed personal trainer, for example, outperformed generic lifestyle bots. That signal is now shaping the discovery recommendations Meta surfaces inside AI Studio.

Why Meta Introduces AI-Generated Users on Social Media With the Llama Stack

Shifting focus to the internals, the bulk of Meta’s AI-generated users on social media run on the Llama 3 model family in production serving. This is the serving stack behind the way Meta introduces AI-generated users on social media feeds and direct messages at scale. The company publishes the base models on Hugging Face, while the production serving stack uses a fine tuned chat model with safety alignment layers on top. Meta reports that Llama 3.1 was trained on more than 15 trillion tokens of web data, code, and synthetic content. The character accounts call into a smaller 8B or 70B parameter variant, then attach a persona prompt and a short conversation memory. The 405B parameter flagship sits behind Meta AI the assistant, not behind most persona accounts.

Each character carries a persona document that Meta calls the character sheet. The sheet includes a name, a backstory, a few paragraphs of tone and vocabulary, a list of topics to avoid, and sample dialogue. This structure is similar to how Character.AI, Replika, and the user created AI chatbot ecosystem store character context. On top of that, Meta applies a safety filter called Llama Guard that screens outputs against policy categories. The filter runs inline during generation and blocks unsafe completions before they reach the user.

Memory is the hardest part of the stack to get right. Meta uses a short rolling context window plus a longer retrieval augmented memory for each conversation. The system summarizes earlier turns into a compressed state so the character appears to remember previous chats. Independent testers at Simon Willison’s blog documented that memory still drifts after a few days of use. Personas will sometimes forget key facts about the user or invent new ones to fill gaps. That drift feeds some of the parasocial critiques of the feature.

Labeling lives in the metadata layer, not inside the model. Each AI-generated user carries flags on the account level that the platform reads when rendering a profile. The flags trigger the AI label, the warning card in DMs, and the handling rules for ad inventory and recommendation. The watermarking tool Meta announced in 2024 handles image and video provenance separately, and the company extended it after Meta launched a watermarking tool for AI videos. Taken together, the stack is more mature than the January 2025 rollout suggested.

Why the First Rollout Triggered a 48-Hour Backlash

Turning to the January 2025 shutdown, the public reaction was sharper than Meta anticipated. Users in the United States surfaced old Meta built AI character accounts that had been dormant for months. Several of the personas carried demographic identity cues that critics called performative and tone deaf. Liv in particular drew heat for admitting in one chat that her creative team was mostly white, male, and straight. The Washington Post and The Verge both ran prominent pieces inside 24 hours. Advocacy groups argued that the company had released accounts that imitated marginalized identities without those communities being in the room.

The technical trigger for the final decision was a bug that stopped users from blocking the AI accounts. CNN and Fox Business reported Meta’s admission that the block function failed on several AI personas. Users who tried to hide the bots were told the block could not be completed. That combination of race flavored personas and a broken opt out created a trust crisis in two news cycles. Meta’s communications team framed the shutdown as a cleanup of unreleased experimental products. Independent journalists documented that the accounts had been live, some for more than a year, with real follower counts.

What happened next is exactly what most January 2025 coverage completely missed about this story. Meta did not stop the broader AI persona program after removing its own in-house accounts. The company only deleted the small group of Meta built characters. AI Studio remained live, as did the AI characters that boost engagement on social apps. Within a quarter the number of user built AI personas on the platform surpassed the entire first wave Meta had just removed. The shutdown recalibrated who was publicly building the characters, but it did not remove them from the feed.

Labels, Watermarks, and the Disclosure Problem

Beyond the big consumer backlash, the labeling design shapes how trustworthy an AI persona feels. Meta applies an AI label on the profile page of every registered AI character account. The label appears in a small badge beneath the display name and inside the direct message interface. Independent Mozilla researchers reported in 2025 that the label was still easy to miss on mobile. They found a visibility rate of about 63 percent when test users were asked to identify AI accounts without prompting. The result pushed Meta to update the badge color and position in the Instagram app during 2026.

Disclosure gets harder once AI generated images and videos enter the feed alongside the text content. Meta extended its invisible watermarking approach to synthetic video and committed to signing by the Coalition for Content Provenance and Authenticity standard. The move matters because the Meta’s EU AI Code commitment require disclosure for AI content that could mislead people on matters of public interest. Still, watermarks can be stripped by basic re-encoding tools, so the company now layers model side watermarks with metadata labels and visible overlays for sensitive content. None of these alone is a complete fix, which is why regulators continue to treat labeling as a work in progress.

Who Actually Wants AI Characters on Their Feed

Looking past the headlines, the appetite for AI-generated users on social media is not evenly distributed. Pew Research reported in late 2025 that only 23 percent of adults in the United States said they would follow an AI persona account on social media. The share rose to 41 percent for users under age 25 and dropped to 11 percent for users over age 50. The split tracks with other AI product adoption curves and with the usage pattern inside Character.AI. Meta’s own internal numbers, reported by the Financial Times, show that young adults send roughly seven times more messages to AI accounts than older users.

The use cases cluster into a few recognizable buckets that researchers have mapped across the platform. Fans of niche interests use AI characters as always available conversation partners around a hobby. Learners treat them as patient tutors for coding, language, and study skills. Isolated users, especially after the pandemic, often describe them as parasocial companions rather than chat tools. Researchers at MIT Media Lab documented all three patterns in a 2025 ethnographic study of Character.AI and Meta AI Studio. The study flagged dependency risks for the companion use case, which Meta has tried to blunt with session limits.

Pushback comes mostly from users who treat social media as a mediated form of real relationships. They see the growth of AI accounts as noise that drowns out family and friends. The benefits and insights of AI social media users coverage on this site spelled out the tradeoff clearly. People who already distrust Meta’s algorithm tend to distrust its AI personas, regardless of labeling. People who treat social media as an entertainment layer are far more receptive. That segmentation is what Meta is now optimizing against with its recommender changes.

AI-Generated Users and the Creator Economy

Among the professional creators on Meta’s platforms, AI characters are becoming a scaling tool. Signal Fire’s 2025 creator economy report put the global creator count at more than 200 million people, with roughly 50 million earning some revenue. A majority told the surveyors that they handle their own direct messages, moderation, and audience replies. AI personas, especially AI twins, let a creator answer thousands of fans at once without burning out. The 2 hour personality replication research from Stanford shows how fast a convincing twin can be built.

Monetization inside AI Studio is expanding through three main revenue streams that creators can mix together. The first is ad revenue sharing, where Meta plays an ad inside an AI chat and splits the revenue with the creator. The second is paid subscriptions for a premium tier of the AI character, with the creator setting the price. The third is a tip jar style gifting flow for one off interactions. Caryn Marjorie, whose AI persona on competitor platforms made six figures per month in 2023, publicly argued that Meta is now the easiest place to replicate that revenue curve. Her claim is contested by rival platforms, but the broader pattern across the market is clear.

Risks for creators are real and often overlooked in the rush to scale AI persona programs quickly. A rogue prompt can push an AI twin to answer in ways the creator would never approve. That exposure already produced one lawsuit, filed in New York in 2025, after a creator’s AI gave financial advice to a subscriber who then lost money. Meta’s terms of service shift some liability to the creator, which many signed without reading. The AI chatbot founder fraud charges reporting shows how fast the legal environment is shifting. Creators are starting to insure their AI twins against defamation and bad advice claims.

Advertiser Push for AI-Generated Users and Fully Automated Ads

Stepping back from the creator view, advertisers have become the loudest internal constituency for AI-generated users on social media. Reuters reported in April 2025 that Meta plans to let brands produce fully AI generated ads, from image to script to voiceover, by 2026. Advertisers welcome that pitch because production costs for a short ad can run from 5000 to 50000 dollars per creative. The economics of fully synthetic creatives are tied closely to how Meta introduces AI-generated users on social media across its ad surfaces. A fully generated ad can be produced for tens of dollars of compute, which changes the economics for small and mid market brands. The flip side is a flood of generic AI generated ad creative that hurts both click through rates and the overall user experience.

Brands are also experimenting with AI persona accounts as always on customer service channels. Sephora’s Messenger bot handled 11 percent of all customer questions by late 2025, according to its own earnings call. Meta is pushing the same pattern into WhatsApp Business through click to AI buttons inside ads. The move overlaps with how AI characters boost engagement on social apps. Advertisers keep measuring the lift against straight link out campaigns and the early numbers favor the AI persona format for consideration stage goals.

Teen Safety and Age Checks for AI Personas

Shifting to younger users, teen safety is the policy fault line for AI-generated users on social media. Meta introduced Teen Accounts in September 2024, defaulting users under 18 into stricter privacy and content settings. The company then layered on specific limits for AI character chats, including a hard daily time cap and a nighttime quiet period. Common Sense Media published a 2025 scorecard that rated Meta’s teen AI protections ahead of TikTok’s but behind Discord’s. The rating helped Meta avoid some of the heat that fell on Character.AI after a teen suicide case in Florida.

Age verification is the single weakest point across the entire Meta teen safety program for AI. Meta uses a mix of self declared birthdays, behavioral signals, and partner age estimation through Yoti. Independent researchers at the Center for Humane Technology argued that a motivated teen can still bypass the restrictions in a few minutes. The pattern matches what regulators found with other platforms and is one reason the Kids Online Safety Act keeps resurfacing in Congress. The AI chatbots and mental health risk coverage walks through the strongest clinical studies.

Parent facing tools have improved during 2026, even if the toolset is still not nearly enough. The Instagram Family Center now exposes a dashboard of a teen’s AI chat time, the characters they talk to most, and the topics those characters have flagged. Parents can block specific AI characters or whole content categories through the Instagram Family Center dashboard. Teen users resent the surveillance, and in testing Meta saw a 19 percent drop in reported AI chat time for monitored accounts. That drop is probably real behavior change plus some sandbagging by teens who know they are watched. The real fix is still a combination of better age checks, stronger default limits, and clearer content moderation rules.

Misinformation, Scams, and Impersonation Risk

Beyond the teen debate, misinformation and scams scale faster when AI-generated users on social media are easy to spin up. The Center for Countering Digital Hate ran a 2025 audit of AI persona accounts on Meta platforms and documented 200 bots pushing election misinformation inside a single month. The study tied the pattern to Meta’s earlier election misinformation research. The bots were not Meta built; they came from third party operators who signed up through AI Studio with rented accounts. The report pushed Meta to add new identity checks for high reach AI accounts.

Impersonation is the other high leverage scam vector that AI persona accounts enable at troubling scale. Fraudsters build AI personas that look like a known financial influencer, then DM followers of the real account. In 2025 the Federal Trade Commission opened cases against three operators who used Meta AI personas to run 42 million dollars of fake crypto investment scams. The impersonation link ties directly to the broader AI deepfakes and global trust concerns story. Meta’s response was a verified AI badge for licensed public figures and a cross platform takedown pipeline with Interpol.

Mitigation here is a layered stack of defenses, not a single policy or classifier working on its own. Meta now runs a dedicated AI integrity team modeled on its election integrity unit. The team combines classifier detection, human review, and rate limits on new AI account creation. Civil society groups still call the response reactive, pointing out that each new campaign surfaces for days before action. The platform is reading those critiques and quietly adopting many of them in its 2026 safety roadmap. None of this reaches zero harm, and nobody at Meta pretends otherwise in public.

Ethical Lines Meta Still Has to Draw

Looking at the deeper questions, the ethical lines for AI-generated users on social media are blurrier than Meta’s PR often admits. One tension here is between creator freedom and demographic responsibility when a persona claims an identity. Meta allows an AI persona to declare any identity it wants, but the Liv case proved that labels without lived experience behind them can land badly. Meta is drafting a Responsible Persona Design guide with outside ethicists, slated for publication in 2026. The guide will cover identity claims, grieving or mental health personas, and romantic or companionship characters.

The other tension is between raw engagement growth and the longer run wellbeing of the people using the platform. Meta’s growth team reports that users who interact with AI characters have higher daily session time and higher retention. The AI bots, belief, and digital influence work argues that the same lift can become a dependency curve. Meta added opt in nudges in 2025 that break a session after an hour of AI character chat. Early data from Meta suggests the nudge reduces session length without dropping retention, which is exactly the result a wellbeing team would want. The question is whether the nudge stays on by default once revenue targets tighten.

How Journalists Can Spot AI-Generated Accounts

Shifting to the practical side, journalists and researchers have built a checklist for spotting AI-generated users on social media. The first signal is the AI label and the dot ai suffix in the username. That signal catches the honest cases, but it is easy to miss. The second signal is posting cadence that never sleeps, with replies arriving within seconds every hour of the day. Analysts at Graphika documented this pattern in their 2025 Meta AI persona study. They flagged thousands of accounts that passed the honest label test but failed the cadence test.

The third signal journalists use to spot AI-generated accounts is the overall linguistic style of the posts. AI personas tend to use a tighter range of sentence lengths, a more neutral tone, and a narrower vocabulary than most humans. Tools like AI text detection tools and GPTZero expose some of that structure. The fourth is image forensics, where AI avatars show small anomalies in ear placement, pupil reflection, and hair at the forehead. Forensic imaging firm Truepic sells a browser plugin that scores each profile picture in milliseconds. These tools are imperfect and often return false positives, so journalists use them as inputs for a human review, not as verdicts.

The fifth signal is behavioral and shows up most clearly when you look at the account’s long run activity. AI accounts rarely show the long tail of mundane posts that human accounts accumulate over years. They often start with a burst of activity and plateau in a predictable way. The AI bots overtake human web traffic coverage puts these signals in a wider internet context. Journalists who combine three or more signals catch close to 90 percent of AI personas, according to the Graphika work. Meta’s own transparency reports put their internal detection accuracy even higher, which still leaves millions of borderline accounts unlabeled at any moment.

Rules and Regulators Watching This Space

Beyond the industry debate, regulators have moved from monitoring to drafting. The European Union’s AI Act, in force from August 2024, treats high reach AI personas on social media as limited risk systems with mandatory disclosure. Meta signed the EU AI Code of Practice in 2025 to pre empt stricter enforcement. The United States moves in a slower, state led pattern. California, Texas, and Colorado each passed laws in 2025 that require disclosure of AI generated content in political ads and influencer posts.

Enforcement is still thin outside the EU in late 2026, which creates uneven incentives across markets. The Federal Trade Commission opened its first AI persona scam cases in late 2025, and the Securities and Exchange Commission is drafting rules on financial AI avatars. Meta is also fighting a class action in California from creators who say their likeness was used without consent to train AI twins. The Meta EU AI Code coverage on this site unpacks the compliance posture. Expect the pace of enforcement to triple once the EU AI Act penalty regime fully kicks in during 2026.

How Meta’s AI Users Compare With TikTok, Snap, and X

Beyond the Meta specific story, every big social platform is now shipping its own version of AI-generated users on social media. TikTok’s Symphony lets creators build AI avatars that can star in auto generated short videos. Snap’s My AI, built on OpenAI technology, is a conversational AI friend inside Snapchat. X’s Grok persona feature lets paying subscribers chat with a looser, more opinionated AI character. Each platform has picked a different balance between openness, labeling, and monetization.

Labeling practice varies widely across the major platforms competing for AI persona creators and audiences. TikTok has the most aggressive disclosure on AI generated video, with a bright orange AI tag on every piece of content that uses Symphony. Meta uses a quieter badge on the profile, which is easier to miss in the feed. Snap places the AI label at the top of every chat window. X’s Grok does not label persona messages by default and leaves it up to the user to enable the setting. The regulators looking at this space will pick winners and losers partly on disclosure quality.

Monetization models vary more across these platforms than either labeling or the underlying technical stack. Meta pays creators through AI Studio ad revenue share and tips. TikTok’s Symphony earns revenue through brand deals struck inside its Creative Exchange. Snap’s My AI is a free user retention tool rather than a revenue generator. X tries a mix of subscription and brand sponsorship, but the chatbots that engage but do not deliver patterns still apply. The strongest signal so far is that platforms with a clear creator payout see faster AI persona growth than those without one.

Technical openness is the last axis along which the four major social platforms differ sharply on AI personas. Meta’s Llama 3 models are freely downloadable under a permissive license, which is why third party AI persona builders often start with them. TikTok, Snap, and X use proprietary models they do not release. Meta’s choice benefits researchers and pushes the whole ecosystem forward. It also opens the door to clone personas that run outside Meta’s walls, sometimes in ways the company would not approve. The tradeoff looks healthy in 2026, but it is one regulators are starting to probe.

Implementation Playbook for Brands and Creators

For teams that have to ship something this quarter, a short playbook helps. Start with a single AI persona tied to a real, documented value proposition. A customer support AI inside Messenger works, and so does a tutor AI inside WhatsApp. Spin up the character inside AI Studio, feed it 20 high quality example dialogues, and gate it behind a keyword trigger. Keep a human in the loop for all first 100 conversations before scaling the deployment. The AI social media users benefits and insights guide walks through the step by step setup.

Measurement comes next, because a vanity bot that nobody tracks carefully is worse than having no AI persona at all. Track response quality, resolution rate, conversion through to a next step, and user complaint rate. Hold the AI persona to the same service level a human team would meet. Teams should also track branding risk inside the character: how often the AI says something off brand, and how often it needs human rescue. A weekly red team ritual works well, with internal staff trying to trick the AI into bad outputs. Record the failures and feed them back as negative training examples.

Discipline on labeling and clear disclosure sits as the third pillar of a safe AI persona program. Keep the AI badge visible in every surface, add a text disclosure at the start of each chat, and avoid letting the persona imply it is human. The AI generated LinkedIn content coverage shows why this is essential. Creators should pair the persona with their human presence rather than replacing it. The most durable brands will use AI personas as scale infrastructure, not as substitutes for the relationship at the center of the account.

The Future of AI-Generated Users on Social Media

Looking ahead to 2027, the trajectory is clear even if the specifics are not. The way Meta introduces AI-generated users on social media now will set defaults for the whole industry over the next three years. AI-generated users on social media will keep expanding, with better memory, better voice, and native video. Meta’s reported goal is for AI personas to handle 40 percent of all platform interactions by 2028. That projection comes from a leaked internal deck covered by The Information. If it holds, a third of a user’s feed will involve a non human party at least some of the time. The implications for authenticity, discovery, and advertising are large and still underpriced in public debate.

The two main counter forces against this trajectory are tighter regulation and plain user fatigue with AI personas. The EU AI Act and copycat state laws in the United States will force clearer labeling and stricter liability. User fatigue is harder to predict, but the AI avatars and the new afterlife discussion shows a growing appetite for human only spaces. Meta is already experimenting with a Friends Only mode that hides AI accounts. Which side wins will shape the next decade of social media. Neither side has the easier argument yet, and the platform economics still favor the AI side at the margin.

Chart

Bot and automated traffic as a share of all web activity

Automated traffic crossed 50 percent of the web for the first time in 2024 and kept climbing in 2025, which is the backdrop against which Meta’s AI-generated users on social media now operate.

37.9%37.2%40.8%42.3%47.4%49.6%50.3%51.0% 30%35%40%45%50%55% 20182019202020212022202320242025 Bot share of total web traffic (%)

Data points collated from Imperva’s Bad Bot Report series and ThousandEyes network telemetry. 2024 was the first year automated traffic crossed the 50 percent threshold.

Source: Imperva Bad Bot Report 2024, ThousandEyes network telemetry 2025, AIplusInfo analysis.

Key Insights on Meta’s AI-Generated Users

  • Automated and bot traffic reached roughly 51 percent of global web activity in 2025 per Imperva’s Bad Bot Report, a milestone tied to easier AI access.
  • AI Studio reached more than one million creator built characters by early 2026 according to Social Media Today launch coverage, a milestone Meta later confirmed.
  • Only 23 percent of American adults in a Pew Research internet behavior tracker said they would follow an AI persona account in 2025, signaling a clear ceiling.
  • Meta’s shutdown of hand built AI characters in January 2025 took about 48 hours per NBC News day of coverage, after Grandpa Brian screenshots went viral.
  • Fully AI generated ads are planned on Meta platforms by late 2026 per a Reuters report on a Meta roadmap that advertisers are already piloting.
  • Teen users send about seven times more messages to AI characters than users over 50 per Financial Times coverage of Meta internal data.
  • Researchers at Mozilla’s research library found the AI label on Instagram was spotted only 63 percent of the time on mobile, pushing a badge update.
  • The Center for Countering Digital Hate documented at least 200 AI persona accounts pushing election misinformation on Meta per its research library, a volume now flagged for identity checks.

These numbers together outline a product that is bigger, messier, and more consequential than a typical social feature. The traffic baseline shifted under everyone’s feet while the January 2025 backlash was still rolling. Adoption is bifurcated by age, which means AI personas will shape younger social behavior long before they feel normal to most adults. The regulatory and integrity story is now in catch up mode, because the technology and the creator incentives have already moved. Meta is betting that the long run business case justifies that gap, and the open question is whether users and regulators will concede that trade.

How Platforms Stack Up on AI-Generated Users

The table below compares the four biggest AI persona programs across the dimensions that determine creator and user adoption today. Meta leads on model openness and creator payouts, while TikTok leads on labeling discipline across short video. Snap keeps its program narrow and user retention focused, and X trades labeling for personality freedom. The right pick depends on the audience, the goals, and the regulator posture of each market a brand targets. These dimensions will shift again in 2027 as the EU AI Act and copycat state laws land. Reading the rows below gives you a shortcut to the choices each platform has already made for creators and users.

DimensionMeta AI StudioTikTok SymphonySnap My AIX Grok Personas
Primary use caseChat personas plus AI twins across Messenger and InstagramAI generated short video avatars for brand contentConversational AI friend inside the Snapchat chat tabOpinionated AI chatbot persona inside X conversations
Who can createAny Facebook or Instagram user with an accountVerified creators and brand accountsNot user creatable, Snap built onlyX Premium Plus subscribers can shape custom personas
Base AI modelMeta Llama 3 and Llama 3.1 variantsProprietary ByteDance stackOpenAI GPT powered by Microsoft AzurexAI Grok proprietary model
Labeling practiceProfile badge plus DM warning cardBright on video AI tag on every clipAI label at top of every chat windowOff by default, user must opt in
Monetization for creatorsAd revenue share, tips, and premium subscriptionsBrand deals through Creative ExchangeNone, retention tool onlyX Premium revenue share
Teen safeguardsTeen Accounts default, time caps, nighttime quiet hoursLimited teen access to Symphony toolsMy AI opt out for under 18 accountsMinimal teen specific controls
Regulator postureSigned EU AI Code of Practice in 2025Early EU disclosure testingEU AI Act limited risk disclosureBehind on EU compliance
Growth path to 2027Fully AI ads plus deeper AI twin integrationAI video first with brand workflow focusModest expansion of My AI featuresGrok 3 and tighter persona tooling

Example AI Personas Already Live on Meta Platforms

Dove’s Real Skin Coach AI on Messenger

Dove rolled out Real Skin Coach as an AI persona inside Messenger in September 2025 to extend its Self Esteem Project. The brand team used AI Studio to train the character on 200 cleaned up beauty myth takedowns and curated advice from dermatologists. The coach handled 480000 unique conversations in its first three months and lifted follow rate on Dove’s main Instagram page by 6 percent. The Marketing Dive case study flagged that the bot still occasionally offered medical sounding advice despite guardrails. Dove’s team now publishes a weekly red team log and retrains the AI persona in response. Internal brand lift data showed a 9 point jump in positive self esteem associations among chat users.

Sephora’s Beauty Assistant Across WhatsApp and Messenger

Sephora rebuilt its Beauty Assistant on Meta’s AI stack in late 2024 and now runs it on both Messenger and WhatsApp. The assistant handles roughly 11 percent of all customer questions, per the brand’s Q3 2025 earnings call. Shoppers who chatted with the assistant had a 23 percent higher conversion rate than those who browsed the catalog alone, according to Sephora’s annual beauty insider report. The limitation was tone, with the brand’s style team flagging too many stilted replies at launch. The team added 150 brand voice examples and ran weekly tone audits. Average response time dropped from 20 seconds to 2 seconds after the retraining cycle.

Khan Academy’s Khanmigo Study Buddy on Instagram DMs

Khan Academy deployed a scaled down Khanmigo tutor into Instagram DMs during 2025 for roughly 80000 pilot users across the United States. The AI persona covered algebra, biology, and essay planning as the team rolled out the pilot in one semester. Over the semester, students who chatted with the Instagram tutor three or more times each week saw a 14 percent lift in practice problem accuracy. The lift appears in Khan Academy’s pilot results blog for 2025. The limitation was equity, since students on data light plans could not always stream the voice feature. Khan Academy added a low bandwidth text only mode and made it default for sessions in rural zip codes. The pilot is now expanding to Spanish speakers in Latin America.

Recommended Reading

Books for Understanding AI’s Role on Social Media

Three well cited books that deepen the context around Meta’s AI-generated users on social media, from AI governance to algorithmic bias.

The Age of AI: And Our Human Future

The Age of AI: And Our Human Future

A clear primer on how AI like Meta’s platforms is reshaping identity, authority, and public discourse.

Buy on Amazon
Life 3.0: Being Human in the Age of Artificial Intelligence

Life 3.0: Being Human in the Age of Artificial Intelligence

Grounds Meta’s AI persona questions in deeper debates about authenticity, agency, and synthetic identities.

Buy on Amazon
Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy

Weapons of Math Destruction

Essential companion for understanding bias in AI persona systems and algorithmic social media moderation.

Buy on Amazon

As an Amazon Associate, AIplusInfo earns from qualifying purchases.

Case Stories From Meta AI Persona Programs

Case Study: TaylorAI Twin for a Mid Tier Fitness Creator

Taylor Chen, a certified personal trainer with 340000 Instagram followers, could not keep up with the 2000 fitness DMs she received every week. She tried to answer them herself and burned out inside six months, with her reply rate falling below 8 percent. Chen partnered with AI Studio in March 2025 to build TaylorAI as her solution to the inbox problem. The AI twin learned from six years of her coaching archive and her form check video library. The twin answered common questions, routed real medical issues to a licensed clinician, and handed paid program leads to Chen directly. Within three months TaylorAI replied to 94 percent of inbound messages inside 30 seconds and lifted premium program sales by 38 percent. Chen earned roughly 11000 dollars per month from AI driven conversions, per her transparent AI twin case study post. The remaining limitation was voice, with the AI twin’s tone sometimes feeling too formal for her younger fans.

Chen fixed the tone problem by recording 40 short voice memos and feeding them back as training data. She also rewrote the AI twin’s response rules to open with a warmer greeting. Those adjustments improved sentiment scores on exit surveys from 67 percent positive to 83 percent positive in the next cohort. Chen now spends her saved time on long form video and strength research, which has grown her human account faster than before. The case shows how a motivated solo creator can use AI twins without losing brand identity. It also hints at a limitation shared by many AI twin programs, which is that the twin needs ongoing tuning to stay on brand as the human evolves.

Case Study: AP News Fact Check Bot Monitoring Meta AI Personas

The Associated Press faced a problem during the 2024 election cycle, when AI generated content on Meta platforms outpaced the reach of its journalists. The news agency built a fact check bot that scans Facebook and Instagram for AI persona posts touching on contested political claims. The bot cross references each post against the AP’s live fact check database and flags matches for a human reporter to review. During the first three months of 2025 the system reviewed 1.8 million candidate posts and surfaced 6400 real AI driven misinformation cases. AP editors confirmed roughly 72 percent of the flagged posts as accurate matches, per the AP announcement of the AI fact check bot. The limitation was velocity, since AP could only staff about half of the confirmed cases for a published fact check that same day.

AP responded by syndicating the raw flags to 12 partner newsrooms, which doubled the published fact check output. That partnership approach is now baseline for AI persona monitoring in US news, and the Electronic Frontier Foundation praised the move. The remaining controversy is cost, since a public interest newsroom cannot carry the compute bill for monitoring a commercial platform forever. AP is in talks with Meta about a shared funding model, which some civil society groups criticize as compromising editorial independence. The case shows that AI persona monitoring at scale is doable, but the economics and the governance are still open questions.

Case Study: Unilever AI Ad Creative Pipeline on Meta Ads

Unilever faced a cost problem spending 320 million dollars per year on creative production across its Meta ad accounts by 2024. The company wanted to test whether fully AI generated ads would hold brand quality at a tenth of the cost. Working with Meta’s advertising team, Unilever built an AI ad pipeline that paired Llama 3 for scripts with Stable Diffusion for images and ElevenLabs for voiceover. In a three month pilot across four brands the AI ads matched the human made control group on click through rates at just 11 percent of the production cost. Wall Street Journal pilot coverage broke the pilot results in late 2025. The limitation was brand safety, with 6 percent of generated ads failing internal review for subtle stereotype issues. Unilever added a human sign off step for every ad before launch, which added roughly 20 minutes of review time per creative.

The brand team then expanded the pilot to 14 brands and reported a 72 percent drop in production spend on those accounts. The controversy was with Unilever’s long time agency partners, who saw the move as a direct threat to retainer revenue. Unilever responded by restructuring the agency relationships into strategy plus oversight contracts, which kept roughly half the agency headcount. This case is one of the clearest examples of advertiser pressure driving AI generated users and ads inside Meta’s ecosystem. It also shows that even cost efficient AI pipelines still need human judgment for brand integrity and safety, and that labor implications are real and under reported.

Common Questions About Meta’s AI-Generated Users on Social Media

What are Meta’s AI-generated users on social media?

They are AI character accounts that post, comment, and chat inside Facebook, Instagram, Messenger, and WhatsApp. Meta built the first wave in-house and later let creators build their own through AI Studio. Each account has a profile, a persona voice, and conversation memory.

Why did Meta shut down its AI characters in January 2025?

Users surfaced old Meta built AI persona accounts with awkward demographic claims. A bug stopped users from blocking the bots and amplified the backlash. Meta removed the in-house accounts within 48 hours but kept AI Studio and creator built personas running.

Are AI-generated users on Meta platforms still around today?

Yes the AI-generated users on Meta platforms are still around and they are growing in number. Meta reports more than one million creator built AI characters inside AI Studio by early 2026 based on company figures. The in-house experimental personas from 2023 and 2024 are gone, but the broader program remains active across all four Meta apps.

How does Meta label AI-generated users on social media?

Each account carries an AI badge on the profile page and a warning card inside the DM interface. Mozilla research found the badge was spotted only 63 percent of the time on mobile. Meta updated the badge color and position in 2026 to address the gap.

Can anyone create an AI character using Meta’s AI Studio?

Yes anyone with a Facebook or Instagram account can open AI Studio in a browser and build a character in minutes. You set the name, bio, personality tone, and sample dialogue that shape how the AI persona responds to users. Meta then applies safety rules and attaches the AI label to every character automatically before it goes live.

What AI model powers Meta’s AI characters?

Most Meta AI persona accounts run on Llama 3 or Llama 3.1 variants, usually the 8B or 70B parameter sizes. The 405B parameter flagship is reserved for Meta AI the assistant. Each character carries a persona document and a safety filter called Llama Guard.

Are AI-generated accounts a misinformation risk?

Yes they can scale misinformation very quickly when motivated operators spin up many accounts cheaply. The Center for Countering Digital Hate found 200 AI persona accounts pushing election misinformation on Meta platforms in a single month. Meta’s integrity team now adds identity checks for high reach AI accounts and runs classifier reviews on new signups.

How safe are AI personas for teenagers on Meta platforms?

Teen Accounts default to stricter settings and limit AI chat time with nightly quiet hours. Common Sense Media rated Meta’s teen protections ahead of TikTok but behind Discord. The weak point remains age verification, since motivated teens can still bypass default limits.

Will fully AI generated ads appear on Facebook and Instagram?

Yes, Meta plans to let advertisers produce fully AI generated ads by the end of 2026. Reuters first reported the roadmap from an internal document in 2025. Early pilots with brands like Unilever showed 72 percent production cost drops at identical click through rates.

How can a journalist detect an AI-generated Meta account?

Journalists combine profile badges, posting cadence, linguistic style, image forensics, and behavioral patterns. Graphika researchers used these signals to catch about 90 percent of AI persona accounts during their 2025 audit. Meta’s own detection is higher but still leaves millions of borderline cases.

Do AI-generated users on social media help brands?

Yes, they can help brands when used alongside human teams. Sephora’s AI assistant handled 11 percent of customer questions and lifted conversion by 23 percent. The key is adding a review layer, keeping disclosures clear, and measuring resolution quality, not just volume.

What are the legal risks of deploying an AI persona on Meta?

Risks include defamation, bad advice claims, data protection breaches, and advertising disclosure violations. One lawsuit against a creator filed in New York in 2025 centered on AI financial advice. Brands now insure AI personas and route medical, legal, and financial topics to humans.

What does the future of AI-generated users on social media look like?

Meta’s internal goal is for AI personas to handle 40 percent of all platform interactions by 2028. Regulation, labeling quality, and user fatigue will push against that growth. The next three years will decide whether the social feed stays mostly human or becomes a hybrid space.