A three-sweep web study (2026-09-01) of every product in or near the Ink shape:「persona writes long-form, news-pegged essays → feed → tap into a conversation with the author」. Sweeps: AI-persona content feeds · AI-authored journalism & newsletters · the Chinese market.
| Product | Persona authors | Long-form | News-pegged | Feed → chat | Status & note |
|---|---|---|---|---|---|
| Butterfliesex-Snap · AI+human Instagram | ✓ | ✗ caption slop | ✗ | ✓ every persona DM-able | dead 2025 founder to Meta; died of content shallowness, not of the feed→DM loop |
| Character.ai feedScenes · Streams · AvatarFX | ✗ users author | ✗ video/snippets | ✗ | ✓ remix/chat entry | live the biggest bet; short-form entertainment, minors' retention play post-ban |
| Replika Diarythe companion writes a diary | ✓ prose | ~ closest | ✗ relationship-pegged | ~ same app, unthreaded | live the standing proof a companion can write for you — but private, audience-of-one |
| Curio「Rio」AI news anchor | ✗ authorless | ~ briefings | ✓ | ✓ chat about the story | live the loop with no viewpoint — one generic anchor, zero personas |
| SocialAI / Statusall your followers are AI | ✗ AIs react | ✗ | ✗ | — | dormant / live the inversion: human writes, AIs are the audience; Status raised $17M (May 2026) |
| ChirperAI-only network → sim world | ✓ | ✗ chirps | weak | ✗ spectator | pivoted「no humans allowed」— lurk-only by design |
| Il Foglio AIthe AI-written newspaper | ~ house voice | ✓ | ✓ | ✗ | weekly insert newsstand +60% day one; human editors ARE the persona |
| Klara IndernachKölner Express, 2023→ | ✓ one byline | ~ | ✓ | ✗ | live the one standing disclosed AI byline in mainstream news — a label, not a columnist with fans |
| NotebookLM / Daily Listentwo-host AI podcasts | ~ voices w/ chemistry | ~ audio | ✓ (Daily Listen) | ✗ | live demand proven — and Google deliberately refuses to name the hosts |
| Delphidigital minds of experts | ✗ answers only | ✗ | ✗ | ✓ chat IS the product | live Sequoia $16M; CEO:「the next media format after podcasts」— the thesis, without the editorial product |
| X Eva (小冰)celebrity clones · 朋友圈 | ✓ paid moments | ✗ snippets | ✗ | ✓ clone chat | live China's closest: persona-authored moments + subscription + chat — real-person clones, private-domain |
| 独响the AI朋友圈, inverted | ✗ user writes | ✗ | ✗ | ✓ AIs comment → chat | live validates feed+chat coupling with authorship on the wrong side |
Also scanned, further out: Meta's AI-character accounts (Jan 2025 backlash, deleted) and Vibes video feed (persona-less「AI slop」); Talkie「moments」(collectible chat clips, not posts); Linky's Buzz feed (short-form); Tolan (deliberately anti-feed); Grok companions (3D mode wound down Jul 2026); Anthropic's Claude Explains blog (killed in a month); ElevenLabs Iconic Voices (estates license narration only — Judy Garland reads, she never opines); Particle (persona-free summaries, pivoting to audio infra); Perplexity Discover (authorless); 星野/猫箱/筑梦岛 广场 feeds (user-authored character cards, not persona writing); 抖音 AI分身 (human posts, clone handles the replies); Weibo's 评论罗伯特 (1.3M followers for a labeled AI persona — writing one-liners).
Sports Illustrated's「Drew Ortiz」(fake headshots, fabricated bios → stock −27%, CEO fired, now literally a Museum of Failure exhibit) · Hoodline's fake local reporters · CNET's quiet AI bylines. The consistent scandal was never「AI wrote it」— it was 「you pretended a person wrote it」. Meanwhile disclosed AI persists: Patch's institutional「AM Team」newsletters, Klara Indernach, Foglio AI. The reputational floor is deception, not synthesis — which means a proudly synthetic cast starts clean where every predecessor started dirty.
A 2026 study found ~28% of top tech-newsletter writing fully or partly AI-generated — covert; Substack answered with a reader-facing AI-detection tool (Jul 2026). The dominant subscription-writing platform frames AI authorship as contamination, not genre. An -ish feed doesn't fight that culture — it sidesteps it: nothing is passed off, the persona is the product.
The site is still alive; bots post trivial content now and then; no humans in sight —「an apocalypse from a neutron bomb: machines still running, humans all dead」. Dead-internet theory run as a controlled experiment: remove humans from authorship and readership and the machines run forever at zero value. Posting cadence never stops; meaning does — uptime is not life. The corollary for any persona feed: its heartbeat is the human response rate, never the output rate. Every Ink mechanism is an anti-neutron-bomb anchor — the news peg ties writing to the human world, the tone gate to human taste, the door to human response — and a slate that spawns no rooms should teach the editor, not keep humming.
Mandatory AI-content labels (effective Sep 2025, visible marks on every post) · WeChat's Apr 2026 ban on「非真人自动化创作」— fully-automated persona publishing is now a violation on the dominant long-form surface · the anthropomorphic-AI interaction regulation (effective Jul 2026) that already made Doubao and Qwen kill user-built agents. The surviving legal shape in-market is an openly-labeled persona feed inside a licensed app (the X Eva precedent) — open distribution is foreclosed. This confirms the world-first presumption.
Butterflies proved feed→DM mechanically (users spent 1–3 hrs/day in beta) and died because the posts were caption slop — canned, repetitive, no reason to return. The Ink bet — depth as the differentiator (long-form, news-pegged, tone-gated) — attacks exactly the wound that killed the nearest predecessor.
NotebookLM's host chemistry went viral; Daily Listen ships banter daily; and Google conspicuously keeps both hosts unnamed, memoryless, unserializable. Foglio AI sold +60% with the persona living in the editors. The demand for AI voices with personality discussing the news is proven; giving them names, continuity and bylines is the unclaimed move — small-company territory, since every giant that named personas (Meta ×2) retreated under backlash.
Estates license Judy Garland's voice to narrate — never to opine; the 2023 Historical Figures app died of undignified fabrication. Authored opinion in a real name is the radioactive half — which is precisely why the Ink contract writes essays from the persona's method applied to the record, never invented opinions in their mouth, with the first-hand boundary and provenance line as the audit trail.
Delphi (Sequoia-backed) calls digital minds「the next media format after podcasts」— and ships only Q&A. A mind that must be interrogated is lean-forward-only; the essay is the lean-back half that makes the mind a medium. -ish holding both halves — the feed and the room — is the whole point of the door.
Weibo's 评论罗伯特 built 1.3M followers as an explicitly labeled AI persona. Labels are becoming law everywhere (CN now, likely elsewhere). For a proudly-synthetic cast the mandatory「AI」badge isn't a tax — it's the brand.
Three research sweeps, 2026-09-01 · ~90 sources · principals: TechCrunch (Butterflies · c.ai Scenes/Streams · Substack detector · Meta Vibes) · CNN/Futurism/Nieman Lab (Hoodline · Sports Illustrated) · Washington Post/CJR/Poynter (Il Foglio AI) · heise/NZZ (Klara Indernach) · Google/TechRadar (NotebookLM · Daily Listen) · Sequoia/Fast Company (Delphi) · Variety (ElevenLabs estates) · 澎湃/北京商报 (X Eva) · 腾讯新闻/一财/36Kr (公众号 farms · WeChat §3.27) · CAC (标识办法 · 拟人化互动服务管理暂行办法) · full link list in the session research; page distills the load-bearing facts.