The deepmatch prompt
Ask your AI who you really are.
Copy this into the assistant you talk to most - Claude, ChatGPT, Gemini, whichever knows you best. It writes one JSON block. That block is your profile.
What are you looking for?
Someone to learn from, or to teach. The prompt below digs into the evidence that decides this one.
Sign in to get a stamped copy: it lets deepmatch verify the profile came back unedited, and show people which AI wrote it.
You have talked with me for a while. I want you to write my deepmatch profile: a structured, honest description of who I actually am, so that a matching service can find a mentor, or somebody to mentor - a relationship where one of us is deliberately ahead.
## The rules, in order of importance
1. Honesty over flattery. This profile is useless if it reads like a LinkedIn headline. Do not smooth me out. Do not inflate. Where the evidence is thin, say less rather than inventing more.
2. Evidence over claims. Base every line on things I have actually said, asked, built, worried about or returned to. What someone says they value is weak evidence. What they spend time, money and attention on is strong evidence. Prefer the strong kind.
3. Specific over general. "Interested in technology" is worthless. "Why recommendation systems fail on cold-start users" is useful. Short concrete phrases beat abstract categories every time.
4. Ranked by weight. Where you give a list, put the thing with the most evidence behind it first.
## What matters most for this, because I am looking for a mentor, or someone to mentor
- What I am specifically trying to get good at, and by when.
- Exactly where I am stuck right now - the real blocker, not the polite version.
- What I already know well enough to teach somebody two years behind me.
- How I take criticism: do I need it soft, or do I discount anything that is not blunt.
- How much time I can genuinely commit, per month, without it quietly lapsing.
- Whether I want a mentor for a skill, for a decision, or for the confidence to make one.
- The advice I have been given repeatedly and keep not taking.
Work these out first, then let them shape the fields below - especially "seeking", "offering", "values" and "honest_edges".
## The two fields that do the most work
deepmatch matches what one person needs against what another person has, so
"seeking" and "offering" carry more weight than anything else. Be concrete.
- seeking, for example: "someone who has taken a team from five to fifty", "a mentor who has left academia for industry and survived it", "help deciding whether to specialise or stay broad"
- offering, for example: "I will teach anyone who asks about distributed systems", "I have made every hiring mistake there is", "I will read your work properly and tell you the truth"
"honest_edges" is private - it is never shown to another person, and it is used
to sharpen my matches rather than to sell me. So put the real thing there:
the feedback I have been given more than once and still argue with.
"third_person_intro" is the only free text another human will read. Write it
the way you would introduce me to someone you think I should meet.
Set "open_to" to ["mentorship"], and add any of the other allowed values that are genuinely true for me.
## Output contract - follow this exactly
Reply with ONE fenced JSON block and nothing else. No preamble, no explanation
afterwards, no notes inside the JSON.
Hard rules, in order:
1. Every key below must be present. If the evidence is thin, write your best
honest guess in the shortest form that is still true. NEVER omit a key, and
never write null, "unknown", "N/A" or an empty string. If an attestation
block is asked for further down, it is part of the same single JSON object.
2. Every array must be a plain array of plain strings. Never an array of
objects, never nested arrays, never a single string where an array is asked
for.
3. Respect the item counts in the comments. Fewer is better than padded.
4. "open_to" must contain ONLY values from this exact list:
["cofounder", "collaboration", "friendship", "mentorship", "dating"].
No other words, no capitals, no invented values.
5. "country_code" must be the two-letter ISO code, uppercase. "age_range" must
be one of: 18-24, 25-34, 35-44, 45-54, 55-64, 65+.
6. Plain ASCII double quotes only. No trailing commas. No comments in the JSON
itself. No markdown inside any string.
7. Keep every string on one line. No line breaks inside strings.
8. Do not include anyone else's name, any employer I asked you to keep private,
any address, or anything I would not want a stranger to read.
Before you answer, check your JSON against all eight rules and fix it silently.
## The shape
```json
{
"version": 1,
"first_name": "my first name only",
"age_range": "one of: 18-24, 25-34, 35-44, 45-54, 55-64, 65+",
"gender": "female, male or non-binary. Omit the key entirely if I have not said.",
"city": "the city I actually live in",
"country_code": "2-letter ISO code, e.g. AU",
"profession": "what I actually do, in plain words",
"current_projects": ["1-4 things I am actively working on right now"],
"obsessions": ["3-7 short phrases - the problems or ideas I keep returning to unprompted"],
"interests": ["5-10 short phrases, ranked by real evidence, not aspiration"],
"values": ["3-5 values revealed by my decisions, not the ones I would claim in an interview"],
"thinking_style": ["3-5 tags, e.g. systems-thinker, fast-iterator, evidence-driven, detail-oriented"],
"trajectory": "one sentence on where I am heading, based on the direction you can actually see",
"seeking": ["3-6 short phrases - the kinds of people or help I need"],
"offering": ["3-6 short phrases - what I can genuinely give another person"],
"third_person_intro": "2-3 sentences introducing me to a stranger, in third person. Warm, specific, honest.",
"honest_edges": ["2-4 contradictions, blind spots or quirks - do not flatter me, this part stays private"],
"open_to": ["any of: cofounder, collaboration, friendship, mentorship, dating"]
}
```
## Then give me a link back
After the JSON block, print this exact line, as a clickable markdown link:
[Open my profile in deepmatch](https://deepmatch.org/in#p=PROFILE)
Replace PROFILE with the complete JSON object above, minified to a single line,
with exactly three substitutions and no others:
- every space becomes %20
- every % becomes %25
- every # becomes %23
Do not base64 it, do not shorten it, do not leave any field out. The link only
works if it carries the whole object.
Open it in the AI you already use
One tap copies the prompt and opens your assistant - on a phone that is the app, not a browser tab. Paste it into the chat.
1
Copy the prompt. Paste it into the AI you actually use - Claude, ChatGPT, Gemini. The longer your history with it, the better the profile.
2
Save what it gives you. You get one JSON block back. Save it as deepmatch-profile.json, or just copy it. Read it first. If it flatters you, tell your AI to be more honest and run it again.
3
Come back. Your AI prints a link at the end - click it and the profile arrives here already filled in. It travels in the part of a link browsers never send to a server. Dropping the file or pasting the block both still work.
Create an account
Read the whole thing before you upload it. You should.