Ask ChatGPT for a Diagnosis Before a Patch

The natural instinct when something breaks is to ask for the fix right away. Paste the error, get a patch, apply it, move on. The problem is that a patch without a diagnosis is a guess. It might work by accident, it might hide the real problem, or it might break something else you cannot see yet.

A better flow is to ask ChatGPT for the cause first. Once you read the diagnosis and it matches what you see in the code, you can ask for a fix with confidence. If the diagnosis sounds wrong, you saved yourself from applying a bad patch. Either way, you come out ahead.

What diagnosis before patch means

You split the question into two steps. Step one: what is probably causing this. Step two: given that cause, what should I change. The split matters because it lets you check the reasoning before you trust the conclusion. A diagnosis you can verify is worth more than a patch you cannot explain.

When to ask for a diagnosis first

• The error could have several root causes and you are not sure which one
• The codebase is unfamiliar and a wrong patch could cascade
• You need to explain the fix to a teammate or in a pull request
• The bug appears in production and you cannot afford a second broken deploy
• You want to learn from the bug, not just silence it

Prompt for diagnosing a slow database query

Problem: this query takes 12 seconds on a table with 500,000 rows.

SELECT * FROM orders
WHERE customer_id = 'abc123'
AND status = 'pending'
ORDER BY created_at DESC;

Table has an index on customer_id only.

Task: explain the most likely cause of the slowness before suggesting any fix.
Walk through how the database would execute this query given the current index.

Why this prompt works

It asks for the cause, not the fix. ChatGPT will explain that the index on the customer column narrows the rows but the status filter and the sort still require a scan. You can verify that by checking your table indexes. Only then do you ask for the fix, which might be a composite index or a query rewrite.

Prompt for diagnosing unwanted component renders

Problem: this React component re renders every time the parent updates, even when its own props have not changed.

function PriceTag({ amount, currency }) {
  console.log("render PriceTag");
  return <span>{currency} {amount}</span>;
}

Parent passes amount and currency from a larger state object.

Task: explain the likely cause of the extra renders.
Do not suggest a fix yet. Just describe what is happening and why.

Prompt for diagnosing a missing email

Problem: the confirmation email is not sent after a user signs up, but no error appears in the logs.

Flow:
1. User submits form.
2. Server creates user record (confirmed working).
3. Server calls sendEmail(user.email, "welcome").
4. sendEmail uses a third-party API.

The API key is valid and other emails send fine from the same key.

Task: list the three most likely causes, ranked by probability.
Explain each one in two sentences.
Do not include a fix for any of them yet.

What to include when asking for a diagnosis

• The exact symptom, including what you expected versus what happened
• The relevant code or query
• Any context that rules out obvious causes, such as the API key works for other emails
• A clear note that you want the cause, not the fix, in this message
• Enough detail that ChatGPT does not have to invent the architecture

How to move from diagnosis to fix

Once the diagnosis matches what you see, send a short follow up. Say something like that diagnosis makes sense, now suggest the smallest change that fixes it without affecting the other queries. Anchoring the fix to a confirmed cause keeps the patch tight and explainable.

Common mistakes

• Asking for the fix and the cause in the same message, which blends them together
• Accepting a diagnosis without checking it against the actual code or logs
• Skipping the diagnosis for simple bugs when the cause is actually not obvious
• Giving so little context that ChatGPT has to guess the architecture
• Applying the fix from a diagnosis that felt wrong just to save time

How to verify the diagnosis

Check the single claim at the center of the diagnosis. If ChatGPT says the index is missing, run a query plan. If it says the email function swallows errors, add a log line and trigger the flow. One targeted check is faster than reading the whole diagnosis twice.

Takeaway

A patch is only as good as the diagnosis behind it. Ask for the cause first, verify it once, and then let the fix follow naturally.