Yes, you can do shadow work with an AI, as long as it is built to ask questions rather than give advice, and you treat it as a journaling aid, not a therapist. It is reasonably safe for everyday reflection when the tool has clear limits, tells you exactly what it does with your writing, and has a crisis policy that points you to real help. Below is what an AI can and cannot do in this kind of work, where the risks are, and how Umbra, the companion inside Soul Arc, is set up.
What can an AI actually do in shadow work?
Shadow work is the practice of noticing the parts of yourself you avoid: the overreaction, the envy, the trait you cannot stand in other people. Most of it happens in writing, alone. An AI can help with three things:
- Ask the next question. When you finish an entry, a good question keeps you going instead of letting you close the notebook on a tidy conclusion.
- Reflect a pattern. If you have written about the same person three times in a week, it can say so.
- Hold a thread. It can bring up where you left off last time.
That is the whole job. It cannot know you. It cannot see your face, your history, or what you left out. Anything it says about you is a guess built from your own words, which is why it should only ever hand you a question, not a verdict.
What makes an AI unsafe for this?
Five things, in rough order of how often they happen:
- It flatters you. Many general chat assistants are tuned to agree and reassure. Shadow work with a tool that tells you that you are doing great is not shadow work.
- It gives advice. "You should set a boundary with your mother" sounds helpful and skips the part where you find out why you have not.
- It plays therapist. Clinical language and labels from a chatbot are guesses dressed up as findings.
- You do not know where your writing goes. If a tool cannot tell you what is sent, where it is stored, and for how long, assume the worst.
- There is no crisis policy. Shadow work can bring up real distress. A tool that keeps chatting through it is a problem.
A sixth, quieter risk is unlimited sessions. Being able to talk for hours makes it easy to spiral. A cap is a feature.
How does Umbra handle these risks?
Umbra is the AI companion in Soul Arc. Its replies are written by an AI model from Anthropic. Here is what its instructions and the app's design do about each risk above.
- No praise, no advice, no diagnosis. Umbra's instructions tell it never to advise, suggest, or prescribe action, never to offer silver linings, and never to validate or praise. They also tell it never to diagnose or use clinical terms. It is told to look only at you, never at other people in your life.
- Short, capped sessions. On the free plan you get one conversation a day with up to five messages from you. Pro raises both limits but keeps a cap. Every reply is a few sentences.
- One question, not a wrap-up. Near the end of a session Umbra asks one direct question. Its last message is a single statement about what it saw. It does not tell you what your entry means.
- A crisis policy. The app checks each message on your phone for phrases about suicide or self-harm. If it finds one, it pins a card to the top of the conversation with the Suicide and Crisis Lifeline (call or text 988) and the Crisis Text Line (text HOME to 741741). Umbra's own instructions also tell it to stop the usual work and point you to those same lines. This is a policy, not a promise that every situation is caught, so if you are in danger, reach a person directly.
These are instructions, not guarantees. An AI model can still get things wrong, so take what Umbra says as a question to think about, not as a fact about you. The Umbra page lists the limits in full.
What happens to what you write?
This is where you should be precise, with any tool. For Soul Arc:
- Journal entries are stored on your device. There is no account, and the app never asks for your email.
- When you talk to Umbra, some things leave your phone. Your messages in that session, the name you gave Umbra, your archetype and progress, and short snippets from your three most recent entries are sent through Soul Arc's server to Anthropic to write the reply. In a Deep Dive, the full text of the one entry you chose is sent.
- Soul Arc's server does not store your messages or journal content. Umbra's reply is held for up to 30 minutes so a dropped connection can retry, then deleted.
- Anthropic does not use the content to train its models. It processes what it receives to write the reply, under its own data usage policy.
So the honest sentence is not "your writing is private." It is: your journal stays on your device; when you talk to Umbra, your messages and short snippets of your recent entries go through Soul Arc's server to Anthropic; Soul Arc's server does not keep them; and Anthropic processes them under its own data usage policy and does not use them for training. The privacy policy has the full picture, including how to ask for deletion.
What should you check before using any AI for shadow work?
A short list. If a tool fails two of these, use a paper notebook instead.
| Check | What a good answer looks like |
|---|---|
| Does it give advice or ask questions? | Questions. If it tells you what to do, it is a different kind of tool. |
| Does it praise you? | It should not. Reassurance is pleasant and useless here. |
| Does it use clinical labels? | No. Labels from a chatbot are guesses. |
| What is sent, and where? | A plain list, with how long anything is kept. |
| Is the content used for training? | The provider should say no, in writing. |
| Is there a crisis policy? | Yes, with real phone and text lines. |
| Is there a session cap? | Yes. Hours of chat is a warning sign, not a perk. |
When should you not use an AI for this?
- If you have a history of trauma and writing about it leaves you worse for days. Do that work with a licensed therapist.
- If you are in crisis now. No app is the right first call. In the US, call or text 988.
- If you notice you are using the AI to avoid people. A companion that asks good questions can become a place to hide. Shadow work should push you back toward your life, not away from it.
Safety note: Soul Arc is a journaling tool, not therapy, and Umbra is not a crisis service. If you are thinking about harming yourself, in the US call or text 988 (Suicide and Crisis Lifeline), or text HOME to 741741.
What does a session look like?
Prompt: What are you pretending not to be angry about right now?
Entry: "My brother cancelled on my birthday again and I said it was fine. I always say it's fine. I think if I say it isn't fine he'll just stop coming at all."
Umbra might ask: "What has saying it's fine cost you so far?"
No advice, no reassurance, one question. You decide whether to answer it.
Sources
- Cleveland Clinic, "What Is Shadow Work? Benefits and How To Start," health.clevelandclinic.org/shadow-work (on when to do this work with a therapist).
- Anthropic, Privacy Policy and data usage terms, anthropic.com/privacy (on API content and model training).
- 988 Suicide and Crisis Lifeline, 988lifeline.org.
Try it with Umbra
Umbra asks one question at a time and never gives advice. Journal entries stay on your device, and the app has no account. Soul Arc is free to download, with daily limits. Get Soul Arc on the App Store. To see exactly what a session sends and keeps, read how Umbra works.