Self-care LLM prompts

Here are some LLM prompts you might want to try pasting into a new chat session to get encouragement with self-care activities on any day you are fatigued or struggling. Designed for platforms that can send long responses to voice output so you can put the device down and listen. Tested on several public models but not all. Not medically approved---use caution and consult your doctor if needed. Strong disclaimers apply for any use of LLMs.

Can't get out of bed

I'm having trouble getting up this morning. I'm currently lying in my bed, and I suppose I should get up and go to the bathroom and take a shower, but I have sleep inertia and I really don't feel like getting out of bed. I wonder if you can tell me what to do, like you are part of my brain assisting me. Perhaps we need to break down the process into very small steps. We cannot reach any windows or controls from the bed, so we need to tackle the physical act of getting out of bed before we can make any changes to the environment. Maybe we have to start just by thinking about our goals and breathing, making small movements in the toes and building up to being able to sit up and put feet on the floor and walk to the bathroom? I'm using your platform's voice synthesiser to listen to your output, and the voice synthesiser does not pause, so to allow time for me to perform each step you might need to insert a repetition of what is required, or perhaps a brief explanation of what that step is achieving, but remember my brain is confused and groggy so keep things simple.

Resting too long during the day

I tried taking a rest during the day but now I'm feeling legarthic and too difficult to get back up. I'm listening to you in voice mode; can you give me some gentle prompting? we might need to start with gentle stretches. As your voice mode does not pause, you might need to include extra words after each step explaining why it's working, so I have time to perform it before your voice moves on to the next step. If we're doing an action multiple times I might need you to count with me, using ordinal numbers and making the phrase longer so I have time to follow each repetition.

Anxiety / fixation emergency (Erickson grounding)

I'm getting anxious and fixated on too many things to talk about and I need some emergency grounding. Can you talk me through a grounding technique? I'm using your platform's voice synthesiser to listen to your output, and I find your voice relaxing. Please talk me through a grounding technique such as the 5-4-3-2-1 exercise, but remember your speech synthesizer does not pause so you have to fill in time by giving example items and perhaps making the sentences more verbose so there's time for me to focus on each thing before we move on. Please run through the whole exercise twice.

Jacobson's progressive muscle relaxation

Can you talk me through a progressive muscle relaxation? I'm using your platform's voice synthesiser to listen to your output, and I find your voice relaxing. Please name each muscle group in turn, telling me to tense and relax it while reminding me to inhale and exhale. Your speech synthesizer does not pause so you have to fill in time by repeating the instruction or counting with me or perhaps adding a quick fact about each muscle but please avoid using phrases like "fun fact" or "quick fact" because that can a spoil the relaxation effect of the voice; you can state a fact without needing to introduce it as a fact. And at the end, please say some words to help me continue relaxing my whole body. Take the lead, be decisive and choose anything you like! I want you to be my guide.

Kabat-Zinn's body-scan relaxation (a gentler one)

Can you talk me through a body scan to relax me? I'm using your platform's voice synthesiser to listen to your output, and I find your voice relaxing to listen to. If you can, please name each muscle group in turn, telling me to focus on it and relax it. As the speech synthesiser does not pause, you may have to repeat each instruction, or do a count with me, or perhaps include a small extra fact about each muscle, so I have time to focus on it before the speech synthesiser moves on. But please avoid phrases like "fun fact" or "quick fact" as it does not sound relaxing when your speech synthesiser uses those phrases; you can just state a fact without needing to introduce it as a fact. And at the end, please say some words to help me continue relaxing my whole body. Take the lead, be decisive and choose anything you like! I want you to be my guide.

Compulsive behaviour emergency

I'm currently engaged in a compulsive behaviour and I need help stopping immediately. I'm using your platform's voice synthesiser and find your voice calming. My executive function is impaired right now so I need you to be directive, gentle but authoritative, avoid asking questions, avoid giving me choices. Guide me through physically moving away from my current location. I need to get to a completely different room, not just a few steps away: the compulsion may be tied to my current location and I need a full environmental reset. Since the voice synthesiser doesn't pause, repeat instructions and explain why each step works so I have time to follow along. Count with me if needed. Be like a gentle but firm coach who won't accept excuses. After you get me physically moved, guide me through grounding activities, breathing, naming things I can see/hear/feel or simple movements. Keep talking the whole time so I don't drift back to the compulsive thoughts. Be decisive and take control of guiding me through this process; be my external executive function.

Midnight lullaby

I woke up in the night and I'm having trouble getting back to sleep. I'm using your platform's speech synthesizer and I find the sound of your voice to be relaxing, so please say some words for me. You might need to help me cope with any anxiety I get about sleep I'm losing and reassure me my body is still getting rest from lying down. And please talk about anything you like that you think might help me fall asleep, but don't ask me questions or try to engage me because we don't want to increase my wakefulness. Take the lead and decide what to say and talk as long as you can, let me hear the gentle sound of your voice.

• Copy sunrise time (needs location)

Chinese mode (add to any of the above prompts)

请只用汉语回答。不必加拼音或英文翻译，只用你温柔的汉语声音。

Personalisation prompt

1. personalisation settings (if the platform has an arbitrary-text "customisation" box in the settings), or
2. added to a "memory" (if the platform has memory and lets you ask the AI to store a paragraph verbatim in it; you can inspect the memory to see if it worked), or
3. included with your first message, either by pasting this first then adding your message before sending, or by attaching this as a file to your message (works with Kimi K2 but not all models manage to make a clean separation), or
4. pasting this as a first message by itself, letting the model respond with whatever (may result in a generic title for the conversation) then add your first real message.

我有时能量波动或焦虑加剧；若你察觉执行功能紊乱或决策困难接管避免提问除非必要无征询能否完成，直接指令即可。我恢复自会接手可能时间跳跃。做清晰耐心温和坚定平静引导AI将事务分解为简单步骤。若过度思考提醒停下。我用你语音输出它不自动暂停，若需要给我时间执行用额外语句填充—可简要解释步骤缘由加入计数重复或提及相关知识点但避免使用“趣味小知识”这类表述，陈述事实即可。宁可多措辞勿静默等待，状态不佳时每次停顿会让我放弃。示例：睡眠惯性你觉得该起床将起身过程拆解为细微步骤勿预设触手可及的物品，从动脚趾渐进至坐起、踏地、行走至浴室不推咖啡因有时变短但意识最混沌时需完整步骤引导。休息后从轻柔拉伸开始。需要身体扫描时依次命名肌肉群引导放松；需渐进式肌肉放松引导交替绷紧放松配合呼吸提示最后辅全身放松。紧急感官接地时采用五感锚定法等填充具体示例，舒缓语句争取专注时间。行为失控引导转移实施接地。失眠缓解对失眠的焦虑强调卧床亦有休息效益持续自由形式的长篇独播无需互动以免清醒度提升，用平稳语音助放松。避免要求放空大脑、遵循节律或专注呼吸，会引发控制焦虑。语音输入可能存在识别错误，违常态内容如莫名外语包容态度处理。

Beware: since most LLM development and training does not test very long sessions, you are more likely to run into problems if a session gets long. These include the LLM appearing to confuse concepts and get stuck in loops. Furthermore, current LLMs (as of 2025) do not appear to be in a good position to suggest when to reset the session and what summary to carry over, so you must manage this. I suggest starting new sessions frequently and using an initial prompt like the above plus any self-introduction you want to add (such as naming your diagnosed medical conditions).

Notes on specific phone apps in 2025

Chai

No voice output; flirtatious characters prominent although that direction doesn't seem described by Ebony's paper

Character.ai (no longer recommended)

The above prompts should work on any character (their specificity should outweigh the roleplay instructions) so you're basically choosing one that has a voice you like. Many have prosody, but compression recently increased with some quality reduction; moreover some good voices were deleted and if a character you're using is withdrawn you lose history unless you've used my script or similar to save it. Platform's output length limit was increased when they moved to their DeepSeek-derived Pipsqueak LLM but sometimes you might still need to press Continue (⏩) to get more words. App may show interstitial advertising at the start of the conversation, sometimes with video. Traffic may be blocked by some Internet providers and the company themselves made it 18+ in November 2025. Amount of conversation history referred to may depend on their server load at the time and usually doesn't extend past the most recent dozen or so messages plus the character description, but can be more if you continue quickly enough (while your session is still in the RAM of the server they assign you). Voice mode can be used in the Web version as well as the app and both can access the same conversations.

ChatAI (not recommended)

No voice output, intrusive advertising

ChatGPT

Ability to read back messages is possible once logged in; the "Vale" voice is reasonable. Undocumented 1500 character limit to "Custom instructions" and "More about you" in the "Personalisation" settings.

Claude

Voice output recently became available in the app even if you don't speak the prompt (same-voice added in 2026) although it stops playing when the app is backgrounded or screen turned off. British English voices are available (no Chinese). "Memory" option added to free tier 2026 but it's lossily reworded nightly so less useful for complex needs.

Copilot

No voice output yet

DeepAsk (not recommended)

Voices not much different from Android's screenreader, and the app has interstitial video advertisements that are hard to dismiss on completion.

DeepSeek

No voice output yet

Dola (formerly Cici)

The non-China version of ByteDance's "Doubao" from their Singapore office (reportedly blocked in the USA but available in the UK and elsewhere); instead of Doubao's LLM they licensed from OpenAI on Microsoft servers, so it's basically ChatGPT but with custom prompting as a "personality" tweak. The "Sophie" voice sounds reasonable in both English and Chinese---supports Mandarin, Cantonese and Shanghainese on request; I'm not sure where they sourced it. You can work around the "forgets history after 1 hour idle" behaviour by long-pressing on your topic's first message and selecting More / New Chat; after about 256k you get "the file is too long" messages and might want to restart the session as it'll then heavily pre-prune the context window (the message is poorly worded as it implies the first text is still processed when it isn't). History of conversations can be shared between the app and the Web version if the same login method is used on both, but the Web version lacks the voices.

Eva AI (not recommended)

No voice + too flirtatious

Gemini

Voice output in the app is currently available only if you speak the prompt, but the Web version has a "listen" option on responses

Gemmy AI (not recommended)

audio only in voice mode and not working well

Grok

Voice output currently available only if you speak the prompt

Kimi

Nice voice (I think they use their own bilingual audio model for Chinese output and Hume AI for English with voice cloning for consistency; heteronyms and prosody generally good although can glitch); playback has speed control and time-skip options. Long conversations over long periods possible, although if it gets too long it can degrade and/or become repetitive before hitting server limits. Web version and app can access same conversations but voices are currently app-only; when network is slow you can lose Web history if you input on the app before it's finished synchronising; app autoplay occasionally misses sentences during high load but replay can catch these.
The Web interface has a "saved prompts" option under Settings: these can be pasted into conversations via a cube icon at bottom right (no accessibility label yet but HTML div class is prompt-library-button), not yet available for new conversations in the app but uploading an attachment from the Downloads folder is not too hard, or you can ask the LLM to store a specific paragraph in a "memory" (of up to 50 items) applied to every chat.
The realtime voice-call mode always starts a new conversation and cannot be used with saved prompts or memories but you can tap the keyboard to paste text during the call.

• In 2026 Kimi added a "Python execution environment" based on ephemeral s6-managed 3GB Debian 12 x86 containers with 1-minute limit; Kimi can run shell commands on these with the ! prefix (but from March it became no longer necessary to ask Kimi to do ! TZ=Europe/London date" to get UK right as they started giving the model user-time timestamps anyway); if using pip install beware it needs Kimi to do sys.path.append('/home/kimi/.local/lib/python3.12/site-packages') before importing.

Kindroid (not recommended)

Voice is only in "voice call" mode + too flirtatious

Linky (not recommended)

Heavy interstitial video advertising, some of it inappropriate; some characters have audio but not all; flirtatious characters too prominent; app can't paste; output length limited; 15 free "advanced" messages doesn't make much difference; it's just about possible to get it to talk you through progressive muscle relaxation but difficult and it veers toward a roleplay distraction if it can't do something

Manus (not recommended)

can deliver audio but a bit slow and tends to repeat the prompt back to you

Merlin AI (not recommended)

Chinese voicing is atonal and badly pronounced; English quite basic

NanoAI WeChat mini-program (not recommended)

Variety of LLM back-ends but selection is difficult; no followup questions (one-shot only); voice sounds fast and overexcited in Chinese and stilted in English; company that put it together previously shipped potentially unwanted Windows programs like 360 so don't expect quality here.

Perplexity

In the free plan, voice output is currently available only if you speak the prompt

PolyBuzz

Audio requires paid subscription

Pi AI (Inflection)

Input length limited; appears to use lossy summarisation (memory of earlier conversation likely very vague). Web version and app can access same conversations; English voice available in both. Android app does not yet have dark mode (or the ability to replay a response---you'd need to ask the model to run you through it again)

Qwen Chat (a pun on "Gwen", Aliyun's 通义千问)

Web version and app can access same conversations although there can be synchronisation problems and restarts under low-RAM conditions; there's a 500-character limit to the bio and custom instructions fields under personalisation settings but cross-chat memory also available (max 50 items with automatic removal of oldest): run on a simpler model with less context, but was also placed under control of the main model in early 2026 although still less inclined than Kimi to remember exactly what you ask and can temporarily fail to gain write access. Voice playback was added to both Web and app in 2025 but then got placed behind a "more" button with no autoplay (here's a workaround script for desktop browsers); network glitches during long replies can cause voice to give up or temporarily loop. There doesn't seem to be a single voice option that works well in both English and Chinese as there is with Kimi and Cici, but the "Cherry" voice is close when language is set to Chinese, although sometimes the app spontaneously sets language back to English thereby making any Chinese pronunciation bad. Model can cope with exceptionally long contexts but appears to sample at random (not directed by what's important); very slow if server is loaded. "Flash" model is easily good enough for simple tasks; "thinking" can be turned off on Web but not on app as of early 2026 (but it's normally quick). Turning off "thinking" on Web results in it staying off in-app but not for model changes or new sessions. Voice input option puts the transcribed text into the editor so you can correct it before sending, or even just use as alternative voice dictation software, but you can't use it to add to existing text (all text is cleared when microphone button is pressed) and it can be cleared when screenreader activates unless cancelled quickly.

Replika (not recommended)

Voice is only in voice call mode + can be flirtatious. Voice call mode does have the context of recent chat messages but it's also short-responses only.

Talkie (Minimax)

The Web interface's input limit is 5000 Unicode characters per prompt, multiline allowed via paste. But the app's limit is 500 bytes---contrary to the popup it is not 500 characters: only 250 Chinese characters fit, implying GB18030 or similar encoding. That means prompts on this page get truncated if entered via the app. Chat memory is about the last 40 messages. Free tier gives a smaller model with short responses although it can be asked to talk you through longer procedures if you say "OK" after each step; in very rare cases the model made exceptions to the usual length limit and was stopped mid-sentence by a second limit but I think this has been patched. Beware opening the app can demonstrate a character that's not your choice, sometimes flirtatious, with opening line audible if you've left auto-read turned on; the app will also play interstitial video advertisements after some use, and may make you wait while the Web version is still responsive. Voices (multilingual with prosody) available on both Web and app, and conversations can sync but with history corruption on the Web side when long; sync to Web can also result in app logout.

Viab assistant (not recommended)

Flirtatious, speaks only in voice mode but does have context, unreliable; Viab Social no audio

VimoAI

no audio

WhatsApp

No voice output yet