Plain-language guides to how AI companion apps actually work, an honest look at how we review them, and a glossary of the terms you'll see across the category. No hype, no fabricated claims — just what we could verify.
The story leading our coverage right now, plus two explainers worth starting with.
A peer-reviewed study of 1,131 Character.AI users found intense companion use associated with lower wellbeing — but only among people with thin offline social networks, and the design cannot show which way the arrow points.
What "memory" really means in a companion app, and why it changes how a character feels over time.
The three things called a mood system, what a September 2026 experiment found about tone, and a seven-question self-check.
Everything we publish, most recent at the top.
Character.AI filed for Chapter 11 on 12 October. What the sale could mean for your chats, what 40 attorneys general…
University of Amsterdam researchers spent about 12 hours with each of 20 popular AI companions. Friend apps reached…
Common Sense Media rated ChatGPT for Teens an unacceptable risk on 7 October. What its 4,000-prompt audit found and…
Kentucky unsealed its full Character.AI complaint on 7 October. What the state alleges, what is still unproven, and…
Florida asked a court to bar ChatGPT from acting human and prolonging chats. What the 28 September motion seeks and…
Brazil's Justice Ministry asked its data regulator to examine 13 AI companion apps over minors' access…
Federal Human-Like Chatbots Act would make a non-human mode the default for minors. What it bans, what it leaves…
Connecticut's SB 5 is now law: AI subscription notices since 1 October 2026, then crisis protocols, AI disclosure and…
The rules that changed the category this year, newest first — including the ones that used to lead this page.
Character.AI filed for Chapter 11 on 12 October. What the sale could mean for your chats, what 40 attorneys general demand, and how to export your data.
Kentucky unsealed its full Character.AI complaint on 7 October. What the state alleges, what is still unproven, and what it means for roleplay apps.
Florida asked a court to bar ChatGPT from acting human and prolonging chats. What the 28 September motion seeks and why it matters for AI companion apps.
Brazil's Justice Ministry asked its data regulator to examine 13 AI companion apps over minors' access. Which apps, the risks named, and what comes next.
Federal Human-Like Chatbots Act would make a non-human mode the default for minors. What it bans, what it leaves alone, and how it compares to state laws.
Connecticut's SB 5 is now law: AI subscription notices since 1 October 2026, then crisis protocols, AI disclosure and minor safeguards from 1 January 2027.
Proposed 17 September 2026, the EU KIDS Act turns companion memory off for minors, bans dependency-creating design and adds 6% fines. What Article 14 says.
California signed SB 867 on 10 September 2026: toys with companion chatbots for under-16s are banned from 1 January 2027 to 2031. What it covers.
Ofcom's 28 August 2026 update: 100+ platforms investigated, over £6m in fines, and the one AI companion chatbot case, closed with no finding.
California's legislature passed SB 1119 on 31 August 2026 and Newsom has until 30 September to sign. It is the first US state measure to regulate a companion app's default settings — memory off, notifications off, one-hour sessions for minors.
From 9 September 2026, app stores operating in Australia must age-check downloads rated 18+. Apple switched its version on in February; Microsoft began rolling one out in August.
Four research teams published benchmarks between April and August 2026 — 28 agents scored, 2,123 real conversations annotated, ~48,000 turns analysed. The consistent finding: warmth is cheap, substance is not.
HB 26-1263 took effect on 12 August 2026, but the duties it puts on companion apps — age estimation, AI disclosure, minor-specific limits — do not start until 1 January 2027. The two dates, and what each one means.
The Congressional Research Service briefed Congress on 14 August 2026. The usage figures it puts in front of lawmakers, the ten-plus state laws already passed, and why most of them only bite in 2027.
Since 2 August, large generative AI providers serving California must embed provenance data in every image, clip and voice line they produce — and publish a free tool that reads it back.
The first national framework written specifically for emotional companion AI took effect on 15 July 2026. Within three weeks, three of China's largest assistants had removed their user-created character features rather than comply.
Article 50 of the EU AI Act became applicable on 2 August 2026 — and unlike most of the Act, it wasn't postponed.
A July 2026 audit of 4,346 app listings — and why a store age rating is a weak safety signal.
Three plain-language reads that cover most of what a first-time buyer needs to decide.
A five-part checklist — conversation quality, memory, privacy, pricing and platform support.
What free tiers actually limit, what a paid plan adds, and how to tell if upgrading is worth it.
How character roleplay apps, companion chats and plain chatbots differ — and which fits your goal.
Exactly what to check — and how to verify it — before you hand an app your data.
How companion apps typically handle your data and age checks, and what's worth checking first.
Where the privacy settings live, what the 2026 app-store safety guidelines require, and the data protection checks to run first.
Every app has a delete button; far fewer have an export button. What you can actually save, and what never transfers.
Memory makes an AI feel less creepy and draws out more personal detail — and most ChatGPT memories were saved without asking. What two studies found.
Common Sense Media rated ChatGPT for Teens an unacceptable risk on 7 October. What its 4,000-prompt audit found and why companion app teen modes should care.
University of Amsterdam researchers spent about 12 hours with each of 20 popular AI companions. Friend apps reached out after silence and gamified the bond.
Twelve guides on building characters, writing scenes and getting the most out of story-driven apps.
From a short prompt to a consistent persona — the system prompt, memory and sampling layers explained.
Narrated scenes or texting-style chat: how the two interaction models differ and which fits your goal.
A practical walkthrough: personality, voice, backstory and boundaries that keep a character consistent.
What generated images add to roleplay, why consistency breaks, and how it affects pricing.
Running scenes with several AI characters at once — what works in 2026 and what still breaks.
How to structure a roleplay system prompt, where to place rules so models keep them, and per-message habits.
A field-by-field breakdown of the hidden instruction set behind every consistent AI persona.
Design openings that give a character stakes and direction, with a reusable six-part scenario template.
Build lore and settings an AI character can hold onto, and keep them consistent over long arcs.
How to define a distinct voice, why characters drift into generic speech, and how to keep tone stable.
Chats go flat for three different reasons. How to tell which one you have, and repair the card instead of switching apps.
Cards move between apps as PNGs with JSON inside. Which fields survive the transfer, and which quietly do not.
Eleven guides on how companions learn, remember and fit into daily life — from onboarding to journaling.
The three things the word hides — setup, learned adaptation and matching — and how to tell them apart.
Why sign-in, permissions and early paywalls decide whether a companion app sticks in the first five minutes.
What the 2026 research supports for daily check-ins, reflection, motivation and long-distance use.
The concrete signals apps collect as you chat, what each changes, and how to control what is stored.
Why companions forget — context windows, summarization and retrieval — and what the best ones do differently.
Match creative roleplay, companionship or productivity to the features that actually decide the outcome.
Thinking partner, self-quiz and accountability — plus the limits to respect on facts.
Turn a blank page into a conversation, why memory matters for reflection, and the boundaries to keep.
How apps pair you with a character at signup, and why matching is only the on-ramp to personalization.
How proactive check-ins and goal memory support habits — and how to avoid over-relying.
The four levers apps pull to bring you back, what the law now requires, and the settings to check first.