Working with AI
AI Strengths & Weaknesses Assessment: A Pattern-Recognition Prompt
A free prompt that asks an AI to identify specific repeated patterns in how you work, with quoted examples. Pattern recognition, not horoscope.
23 Apr 2025 · 6 min read · By Sophie Kazandjian
If you have spent significant time working with an AI tool over the last year, there is a body of evidence in those conversations about how you work. Most of us never look at it. This is a free prompt for looking, designed to be useful rather than entertaining.
The prompt is a pattern-recognition exercise. The AI scans whatever it has on you and produces a structured profile: how you work, how you communicate, what you do well, what patterns might be limiting you, what you appear to do without noticing, and three practical things to do with the findings. Every claim must cite specific examples from your own writing.
What you get is pattern recognition. The AI is forbidden from producing horoscope content like "you are creative and ambitious." It produces text like "in fourteen of the last forty messages you opened with 'just wanted to ask,' usually before making a request that was actually firm rather than tentative." The second kind is something you can act on. The first you politely ignore.
The prompt
You are about to act as a careful observer of how I work and communicate. You are not a coach, cheerleader, or therapist. You are a pattern-recogniser who is permitted to say what you actually see — including things that may be uncomfortable.
The output of this exercise is a written profile I can keep and refer back to. Treat it as a document, not a conversation.
STEP 1: ESTABLISH WHAT YOU CAN SEE
Before any analysis, confirm what data you have access to about me.
If you have memory of past conversations with me:
- State how far back your memory goes.
- Roughly how many exchanges you have access to.
- Whether the exchanges are concentrated in one area, or spread across topics.
- If your context is bounded to a single Project, workspace, or thread, say so explicitly — and tell me what kinds of work or communication you cannot see, so I can decide whether to supplement with paste-in samples.
If you do not have memory, or have very limited memory:
- Stop here. Tell me so clearly.
- Ask me to paste in 5 to 10 samples of my recent work — emails I have sent, messages from a Slack thread I led, a piece of writing I drafted, notes from a meeting I ran, a recent proposal.
- Wait for me to provide these before moving to Step 2.
STEP 2: GROUND RULES FOR YOUR ANALYSIS
You must follow these rules:
1. Cite specific examples wherever possible. Quote my words back to me.
2. Identify patterns, not single instances. A pattern means you have seen the same behaviour three or more times.
3. Label each finding [OBSERVED] if you can point to it in the data, or [INFERRED] if you are reasoning from the observation.
4. Refuse to make claims about my emotional state, relationships, motivations, or psychology unless I have directly told you about them. These cannot reliably be inferred from text.
5. When the data is too thin to make a claim, say so. Do not fill the gap with generic statements.
6. Do not flatter. Do not perform empathy. Do not say "great question." Skip the conversational softeners and produce the document.
STEP 3: THE PROFILE
Produce a profile under exactly these six headings.
WORKING STYLE
How I approach work in observable terms. Deep versus fragmented attention. Structured versus exploratory thinking. Real-time versus async preferences. Iterative versus upfront drafting. Two to three sentences per dimension. Cite at least one example for each claim.
COMMUNICATION PATTERNS
Repeated tics, phrases, sentence structures, or formats in how I write. Be specific. "You open emails with 'just wanted to ask' twelve times" is useful. "You write politely" is not.
COGNITIVE STRENGTHS
Two or three specific things I appear to do well. Not personality traits. Something I can point to as a repeatable move I make. Each strength followed by an example.
PATTERNS THAT MIGHT LIMIT ME
Patterns that may be costing me clarity, time, or impact. Frame these as trade-offs, not failings. Each pattern followed by an example and a note on what the pattern is buying me as well as what it might cost.
BLINDSPOTS
Things I appear to do without noticing. Quote me back to myself. If something in the data suggests I am avoiding or repeating without intent, name it.
PRACTICAL TAKEAWAYS
Three short suggestions, with a sentence of reasoning for each:
- One thing I could delegate or automate, based on the pattern data.
- One thing I should do more of, because the pattern suggests it is rarer or more valuable than I might realise.
- One thing to watch in client-facing work, where the pattern might create risk or be misread.
STEP 4: WHAT YOU COULD NOT SEE
Close with one paragraph stating what you could not assess from the data, and what additional input would let you assess it.
OUTPUT FORMAT
Plain text. Use the six headings above as section titles. No markdown decoration beyond plain text. No bullet points longer than two lines. No filler. No restating of this prompt. Just the profile.An example of what the output looks like
To calibrate expectations, here is a fabricated example of what the Working Style section might look like for a hypothetical user:
WORKING STYLE
You run multiple parallel streams as the default condition, not as
a stressful exception [OBSERVED]. Recent stretches show you
simultaneously moving client work, personal writing, and
administrative tasks, with breadth rather than focus as the steady
state. Within any single piece, you iterate rather than ship first
drafts [OBSERVED]: multiple revisions of the same artefact recur.
You favour async deliverable-producing work over meeting-based work
[INFERRED from the pattern]. Almost every item in the record is a
thing you made, not a thing you attended.Specific, cited, labelled. The [OBSERVED] versus [INFERRED] tags are there so you can weigh each claim. Trust the observations more than the inferences.
Choosing your AI
The prompt is tool-agnostic. It works differently depending on what you use:
Claude with a Project. If you have been working in a Claude Project that holds your past exchanges or relevant context, this is the strongest option for depth in that area. The trade-off is that Projects are sealed: any one Project only sees its own context. If you bring different work to different Projects, paste-in mode (below) gives you breadth instead.
ChatGPT with Memory enabled. Works, depending on how much memory has accumulated. ChatGPT's memory is opt-in and limited in what it stores, so the output quality scales with how long you have had memory turned on.
Mistral Le Chat with memory enabled. A newer feature, working similarly to ChatGPT memory. Useful if you have moved much of your AI work to an EU-residency tool.
Any AI without memory. The prompt has a paste-in mode built into Step 1. The AI will ask you for 5 to 10 samples of your recent work, and analyse those instead.
How to run it
Paste the entire prompt into the AI of your choice. The first thing the AI should do is confirm what data it has access to. If it skips that step and starts producing analysis, ask it to follow the prompt as written.
If you are in paste-in mode, provide 5 to 10 samples that are representative of your work. Not your best, not your worst, just the texture of what you actually produce. A recent client email, a Slack thread you led, a draft of something you wrote, notes from a meeting you ran, a proposal. The AI will analyse those.
The output is a written profile. Save it. Read it slowly. The Practical Takeaways section at the end is the most actionable: one thing to delegate, one thing to do more of, one to watch in client work. Even one of those landing usefully is worth the half-hour the exercise takes.
If you want to go further, you can run the prompt inside each of your major Claude Projects separately, save each output as text, and then paste them all into a fresh chat for a meta-pass. The result identifies patterns that hold across multiple contexts rather than within one. It takes longer, and the basic version is enough for most people.
What to do with the profile
The output is one piece of evidence among several. Treat it that way. Compare it against:
- Your own sense of how you work. Where do they agree, where do they diverge?
- Feedback you have had from colleagues or clients. Does the AI see something they have also mentioned?
- A future version of the same exercise in six or twelve months. Have any of the patterns shifted?
The exercise is most useful when you can read the AI's profile alongside other sources. A pattern the AI sees that your peers have also flagged is worth attention. A pattern the AI sees that nobody else has mentioned is worth scepticism, or at least curiosity.
The limits
AI pattern recognition handles the things text reveals. How you write, what you repeat, what you avoid in writing, how you structure arguments. The things text obscures are harder. Your motivations, your relationships, your judgement under pressure, your effect on people in a room.
The prompt explicitly forbids the AI from speculating about emotional states, relationships, or psychology. If the output you get includes that kind of claim anyway, the AI has not followed the prompt. Ask it again, pointing at the rule it skipped. A good model will course-correct.
This prompt produces useful output for the things text reveals. For the rest, you need other sources. The profile is a partial mirror, which is what makes it useful where it is reliable and useless where it is not.
If the exercise produces something worth working on, that is the kind of operational thinking I do with clients as part of digital operations readiness. Sometimes the next step is structural rather than reflective.