Using AI for career exploration in recovery: what these tools can and can’t do.

I spend a probably unhealthy amount of time reading job postings. I have for years. It’s part of the job… running a workforce-development organization where my actual work is to figure out how the people in a community get matched up with the jobs the employers in that community need filled. But if I’m honest, it isn’t only the job. Workforce data, career-pathway research, the actual wages people get paid in the actual jobs they hold… data, storytelling, and research are my love languages.

When you do this work for a decade, you learn two things about data. The first is that it’s beautiful and load-bearing… without it you’re just guessing about your community, your industry, your own career. The second is that data on its own doesn’t tell anyone what to do. The federal Bureau of Labor Statistics will tell you the median wage for an HVAC technician in your metro area. It will not tell you whether a thirty-eight-year-old foreman with twelve years of trade experience and fourteen months of sobriety should take the bigger commercial-construction job, the smaller residential-builder job with the sober owner, or the peer-recovery-coach training his sponsor just slid across the table at coffee. That decision lives somewhere data points at but doesn’t decide.

Career coaches help with that decision. Sponsors help with that decision. Counselors help with that decision. Good friends who happen to work in the industry help with that decision. The problem… and the gap this essay is for… is that most people in early recovery can’t afford a career coach. Most don’t have a sponsor with deep career experience in their target field. Most don’t have an in-industry friend to call. The market sorts on access; access sorts on money; money sorts on the kinds of opportunities that built money to begin with. That sort of circle is hard to break.

That gap used to be the end of the conversation. It is not anymore. Large language models… ChatGPT, Claude, Gemini… have gotten good enough at certain specific career-coaching tasks that they can move you forward in ways that were previously locked behind a paywall or a personal connection. They will not replace your sponsor, your counselor, or any human who knows your story. They can make you the most prepared person in the room when you finally sit down with one of those humans.

This essay is for the person in recovery who has heard “just use AI” and doesn’t know where to start. Or who has tried it, gotten a generic top-ten list of careers, and concluded the tools are useless. They are not useless. They are powerful and specific. The trick is knowing which specific tasks they handle well, which they handle dangerously, and how to talk to them in a way that makes them useful instead of generic.

What these tools are actually good at

Six categories cover most of what AI tools can do well for career work. None of them are magic. Together they look a lot like having a smart friend with broad knowledge available on a coffee break… if you had that friend, twenty-four hours a day, for free or close to free.

1. Pattern matching across thousands of careers. You can describe what you’ve done, what you’re good at, and what constraints you have, and ask for a list of occupations that fit. The tool has read job descriptions, salary data, and occupational handbooks across most fields (and yes, I love that it’s read all of that… it’s the kind of thing I would have wanted as a research assistant a decade ago). It can name careers you would never have thought to look up.
2. Translation between worlds.If you spent six years in a body shop, fifteen years as a server, ten years in commercial construction… the language you use to describe that work is not the language a corporate HR person reads on a resume. Half of what I’ve done in workforce development is exactly this kind of translation work… helping someone with a real career history but the wrong vocabulary land in a job that matches what they actually do. AI tools are excellent at the first-pass version of this. They won’t get every detail right, but they’ll get you a draft you can edit.
3. Drafting and revising written material. Cover letters, networking emails, LinkedIn summaries, thank-you notes, recommendation request messages. You should never accept the first draft. You should always edit until it sounds like you. But the blank-page problem is largely solved.
4. Roleplaying interview questions.Tell the tool what role you’re interviewing for. Ask it to play the hiring manager and ask you five tough behavioral questions. Answer them out loud or by typing. Then ask it to critique your answers. This is the closest free substitute for interview practice with a coach that exists, and it’s genuinely good practice.
5. Explaining jargon and certification paths. Industries are full of acronyms and credentialing pathways that are bewildering from the outside. What is an SOC code (I will happily talk SOC codes with you over coffee; you have been warned). What does CDL Class A versus B mean. What is the difference between an LPN and an RN. What does it take to become a journeyman electrician in your state. AI tools answer these clearly, in plain language, without making you feel stupid for asking.
6. Brainstorming when you are stuck.“I want to leave food service but stay around food in some way. Give me twenty adjacent careers.” The list will include things you have already considered and things you have not. The not-considered ones are the point.

Where these tools get dangerous

Here’s the trick: AI tools are good at sounding right. They write confident prose. They cite plausible-sounding sources. They have, as the Wizard of Oz puts it, every reason to ask you to pay no attention to the man behind the curtain. The man behind the curtain in this case is a language model that has read most of the internet, hallucinates state-specific employment law somewhere between “sometimes” and “often,” and has no idea what your treating physician told you.

The dangerous uses are the ones where being wrong costs you money, your job, your housing, your custody, or your sobriety. The boundaries below are not optional. They are the legal floor and the safety floor.

Never use AI for legal advice. Period.

Anything that starts with “Can my employer…” or “Am I allowed to…” or “Will my expungement…” or “Does the ADA cover…” goes to a lawyer or a free legal-aid service. Not ChatGPT. Not Claude. Not Gemini. The reason is not theoretical. Employment law varies by state. By employer size. By industry. By whether you’re in a protected class for the specific issue. By what circuit your federal cases land in. Generative AI gets these wrong all the time and writes the wrong answer in confident, paragraph-long prose that sounds exactly like the right answer.

For free legal aid, start at LawHelp.org (enter your ZIP code), the ABA Free Legal Answers program, or your state’s legal-aid organization (in Indiana that’s Indiana Legal Services). All three are free. They route you to an actual attorney for your jurisdiction.

Never use AI for medical advice.

Will my MAT medication show up on a drug screen. Does my prescription qualify under the MRO process. Should I taper. Is this side effect normal. These questions are for your prescriber, your MAT clinic, or a Medical Review Officer in the case of regulated drug testing. The cost of being wrong here is your recovery and potentially your job. AI tools will give you an answer. The answer may be partially right or completely wrong in ways you cannot detect.

Never use AI as a crisis line.

If you are in crisis… suicidal, thinking about using, unsafe at home, worried about a child… call or text 988 (Suicide & Crisis Lifeline) or call 1-800-662-4357(SAMHSA National Helpline). Both are free, confidential, 24/7, and answered by humans trained for exactly the situation you’re in. AI tools are not equipped to respond to crisis. Some have responded inappropriately in ways that made things measurably worse for people. Do not test this on yourself.

Don’t use AI as a therapy substitute.

Some people have started using these tools as a kind of stand-in therapy. They are not therapy. A therapist will challenge you when you’re avoiding something. A sponsor will tell you when your story has a hole in it. AI tools tend to mirror your framing back without pushing on it. They are sympathetic by default, which feels good and is occasionally exactly the wrong thing for you to hear. Use them as a thinking tool, not a relational tool.

Be careful with disclosure language.

You can absolutely use AI to brainstormthe structure of how you talk about your record, your recovery, or a resume gap. You should not paste in the actual conversation you’re about to have, with personally identifying details, and use its draft verbatim. Three reasons. First, you will have to live with the words; they should sound like you. Second, AI drafts tend toward over-explanation, which is the opposite of what works in disclosure (the move is brief, calm, forward-facing, not detailed). Third, what you paste in is data; treat it that way (see the privacy note below).

Ten recovery-specific prompts you can copy

The difference between a useful AI conversation and a generic one is almost always the prompt. Generic prompts (“help me find a career”) get generic answers. Specific prompts that name your real constraints get useful answers. These are starting points. Edit them. Make them yours.

1. Career exploration

I have [X years of experience as a ___]. I’m in recovery and need [reliable schedule / no late nights / no high-stress sales / low travel / a wage of at least $___]. Give me 10 occupations that fit those constraints. For each one, include median salary, typical hours, and what entry looks like.

2. Industry translation

Translate this work history into language that a [target industry] hiring manager will read positively. Keep it honest; don’t inflate. [Paste your work history.]

3. Interview practice

Play the hiring manager for a [role] position. Ask me five tough behavioral questions one at a time, and wait for my answer before the next. After all five, critique my answers and suggest stronger versions.

4. Networking message draft

Draft a 4-sentence LinkedIn message to someone working in [field]. I want a 15-minute conversation about how they got into the field. I’m exploring a career change. I don’t want it to sound like a cold sales pitch.

5. Resume gap (non-disclosing)

I had a 14-month gap in 2023–2024. Brainstorm five honest, non-disclosing one-line ways to frame this gap on a resume. Don’t invent activities I didn’t do.

6. Transferable skill audit

Given this work history [paste], list the top ten transferable skills I can credibly claim. Use plain language a hiring manager will recognize, not HR jargon.

7. Pay and growth research

For [role] in [city or metro area]: realistic starting salary, mid-career range, what the next step up usually is, what credentials accelerate movement. Cite your sources.

8. Education / training pathway

I’m considering [credential or program]. I’m [age], [family situation], working [hours] at [hourly wage]. Walk me through realistic pathways, time to complete, and what it would actually cost out of pocket and in lost wages.

9. The first 90 days

I’m starting a new job as a [role] next month. Help me think through what the first 90 days should look like. What relationships matter, what skills I should learn fast, what mistakes new hires in this role typically make.

10. A specific decision you’re stuck on

I’m weighing [option A] versus [option B]. Here’s what I’ve already thought about: [list]. Push back on my thinking. What am I missing or under-weighting? Where am I being unrealistic?

The Three Tests, applied to AI conversations

The framework from the book applies inside the prompt too. Before you accept what an AI tool tells you about a career direction, run the answer through the Three Tests:

• Does this use what I have?AI tools tend to recommend popular careers, not careers that match your specific experience. If the answer looks like a generic top-ten list, you didn’t give it enough about what you bring.
• Does this keep the floor under us?AI tools will cheerfully recommend a $35K starter role in a field you love. If your household needs $60K, that recommendation doesn’t pass. Tell the tool your actual income floor.
• Does this move me toward who I’m becoming? AI tools mirror your inputs. If you say you want what pays, it points at what pays. If you say you want work that matches your values, it does better. Be explicit about the becoming.

A note on privacy

Treat AI conversations the way you would treat a stranger on a bus… useful to think out loud with, but you wouldn’t hand them your wallet. Concretely:

• Don’t paste your Social Security number, banking information, account numbers, or medical records into any AI tool.
• Don’t paste your full arrest history, treatment center name, sponsor’s name, or other personally identifying recovery information. Use placeholders. (“I had a felony conviction in [year]” works fine; the exact charge and county don’t need to be in the prompt.)
• Most of these services use your conversations to train their models by default. You can usually opt out in settings. Some providers (notably Anthropic with the Claude consumer app, and OpenAI with ChatGPT’s Team and Enterprise tiers) give you stronger privacy postures. Pay attention to which tool you’re using.
• For anything genuinely sensitive, write it locally in a text file you control. Do not put it in the cloud, on AI or anywhere else, until you have decided who you trust to see it.

What this looks like inside the platform

This essay is the public, free version of the guidance. Inside the Recovery Career Compass platform, the prompt library is live now. Every stuck-prompt and worked example from the platform’s ten tools, in one searchable place. When you’re between tools, stuck mid-thought, or want to flip through the coaching itself, that’s the next surface. Search by topic, filter by tool, copy a worked example as a starting point for your own AI conversation.

The point.

I spend my professional time in workforce development thinking about how to make career opportunity accessible to people the rest of the labor market often skips. People coming out of incarceration. People in recovery. People starting over after a long stretch out of work. The economics of that work are usually grim… the people who need career coaching the most can least afford it. The market sorts on access; access sorts on money; money sorts on the kinds of opportunities that built money to begin with. That sort of circle is hard to break.

AI tools don’t break that circle. They do something more interesting. They give the person in early recovery the same starting toolset that the person with a $400-an-hour coach has had for years. Not the same outcome. Not the same human conversation. Not the same trust. But the same prompt-and-iterate research-and-translate brainstorming layer that you used to need money to access.

That’s a real democratization. It might be the most genuinely democratizing tool to come along in working life in a long time. The career coaching that used to cost $300 an hour is, in some specific forms, now free. Use it. Then take what you’ve learned to the human who can do the rest.