Report an AI feature problem with the details support needs first time

TL;DR — When an AI feature gives the wrong answer, exposes sensitive data, or fails to respond, send one report with five things: what you did, the exact prompt/input, the exact result, when it happened, and a screenshot or request ID if available. The single biggest fix for slow support is to include a reproducible example with personal or confidential data removed. Reading time: ~5 min

Goal

When you finish, your software agency will have a complete, safe-to-share problem report for the AI feature, with enough detail to reproduce the issue and investigate it without asking you for basic follow-up information.

Prerequisites

• Your product account login.
• Access to the page in the product where the AI feature failed.
• Permission to contact support through your agency's normal channel: support portal, ticket form, or email.
• A screenshot tool on your device.
• The following information ready:
  • The date and time the problem happened, including your time zone.
  • The exact text you entered into the AI feature, if you can safely share it.
  • The exact output or error message you received.
  • The browser and version you used. In Chrome, Edge, Firefox, or Safari, open the browser menu → Help / About and copy the version number.
  • Your operating system and version.
  • If shown in the product: conversation ID, request ID, trace ID, job ID, or session ID (an ID is a unique reference number support can search for).

Steps

Step 1: Confirm this is the kind of issue that should be reported

Report the problem if it matches one of these cases:

• The AI output is clearly wrong for the same input every time.
• The AI output includes private, confidential, or unrelated customer data.
• The AI feature refuses valid input, times out, or shows an error.
• The AI feature produces unsafe, harmful, or policy-violating output.
• The AI feature behaves differently for different users when it should not.

What you should see when this step succeeds: you can describe the issue in one sentence, such as "The AI summary included another customer's name" or "The AI chat returns an error after I click Send."

Step 2: Capture the exact input and output

In the product, go to the page where the AI feature failed and copy the exact text you entered and the exact result you received.

Use this template in a note app or document:

Issue summary:
Expected result:
Actual result:
Exact input sent to the AI:
Exact output returned by the AI:
Date/time with time zone:
How often it happens:

Fill it with literal values. Example:

Issue summary: AI summary included data from the wrong account.
Expected result: Summary should only mention the current ticket.
Actual result: Summary included another person's email address.
Exact input sent to the AI: Summarize this support ticket in 3 bullet points.
Exact output returned by the AI: Customer Jane Doe at jane@example.com asked for a refund...
Date/time with time zone: 2026-10-30 14:25 America/New_York
How often it happens: 2 out of 2 times with this ticket

What you should see when this step succeeds: you have a note containing the exact input, output, and expected result.

Step 3: Remove sensitive data before you send anything

⚠️ Do not send passwords, API keys, full payment card numbers, government IDs, private health information, or production secrets in a support ticket. Replace them before you attach screenshots or paste text.

Replace sensitive values with placeholders like these:

[email redacted]
[phone redacted]
[customer name redacted]
[api key redacted]
[internal URL redacted]

If the issue is specifically that the AI exposed sensitive data, keep enough evidence to show the problem, but redact most of the value. Example:

Wrong output shown by AI: Customer email is ja***@example.com

What you should see when this step succeeds: your notes and screenshots still show the problem, but no longer expose secrets or unnecessary personal data.

Step 4: Collect IDs, timestamps, and environment details

In the product page where the issue happened, look for labels such as "Request ID", "Trace ID", "Conversation ID", "Message ID", or "Job ID". If your product has an activity log or audit log, open it and copy the matching entry.

Use your provider's dashboard path if available, for example:

Your product dashboard → Activity / History / Logs → open the failed AI action → copy the ID and timestamp

Also record your environment in this exact format:

Browser: Chrome 141.0.0.0
OS: Windows 11
User time zone: America/New_York
Account/workspace: Acme Support Team
Affected record/page URL: https://app.example.com/tickets/12345

What you should see when this step succeeds: you have at least one timestamp and one unique ID, plus browser and OS details.

Step 5: Take screenshots that show the full context

Take 2-3 screenshots:

1. The full page with the AI feature visible.
2. The exact error message or wrong output.
3. If available, the page section that shows the request ID or activity log entry.

If your browser shows the page URL, keep it visible unless the URL contains secrets. If it contains secrets, crop or blur that part before sending.

What you should see when this step succeeds: you have screenshots that show where the issue happened and what the product displayed.

Step 6: Reproduce the problem once in a clean browser session

Open a private/incognito window (a temporary browser session without normal cookies), sign in, and repeat the same action one time.

Use the browser menu path:

Chrome / Edge: Menu → New Incognito/InPrivate Window
Firefox: Menu → New Private Window
Safari: File → New Private Window

Repeat the same input exactly. Add the result to your note:

Retest in private window: Yes
Result: Same problem happened / Problem did not happen

What you should see when this step succeeds: you know whether the issue is reproducible or tied to cached browser state (saved local browser data).

Step 7: Send the report in one complete message

Use your normal support channel. If your agency has a ticket form, use that first. If it uses email, paste this template exactly and replace the values:

Subject: AI feature issue report: [short summary]

Issue summary:
[one sentence]

Impact:
[who is affected and whether work is blocked]

Expected result:
[what should have happened]

Actual result:
[what happened instead]

Steps to reproduce:
1. [exact page or record opened]
2. [exact text entered]
3. [button clicked]
4. [result shown]

Exact input sent to the AI:
[paste redacted text]

Exact output returned by the AI:
[paste redacted text]

Date/time with time zone:
[YYYY-MM-DD HH:MM Area/City]

Frequency:
[every time / intermittent / once]

IDs:
[request ID, conversation ID, trace ID, job ID]

Environment:
Browser: [name and version]
OS: [name and version]
Account/workspace: [name]
Affected URL: [URL]

Attachments:
[screenshot 1, screenshot 2, screenshot 3]

What you should see when this step succeeds: the ticket or email contains all details in one message, with attachments included.

Verify it works

Check that your sent report includes all of the following before you close the page:

If you used a support portal, the ticket page should show a ticket number after submission. If you used email, your sent mail should show the subject line and attachments.

[You] ---- 1. capture input/output ----> [Issue Note]
  |
  | 2. redact secrets
  v
[Safe Report] ---- 3. attach screenshots ----> [Support Ticket]
  |
  | 4. include IDs and time
  v
[Reproducible Case]

Common pitfalls

Sending only a screenshot

Mistake: You send an image of the error but not the exact input text.

Symptom: Support replies asking what you typed or which action triggered the issue.

Fix: Paste the exact redacted input and the exact output into the ticket body.

Omitting the time zone

Mistake: You write "happened at 2:25" with no time zone.

Symptom: Support cannot match your report to logs, especially across regions.

Fix: Send the time in this format: 2026-10-30 14:25 America/New_York.

Redacting too much

Mistake: You replace the entire prompt and output with [redacted].

Symptom: Support cannot reproduce or understand the issue.

Fix: Keep the structure and enough text to show the problem, and redact only the sensitive parts, for example ja***@example.com.

Not testing in a private window

Mistake: You report the issue without checking whether it is caused by cached browser state.

Symptom: Support cannot reproduce it, or it only happens in your normal browser session.

Fix: Repeat the same steps once in a private/incognito window and include that result in the ticket.

Leaving out IDs shown in the product

Mistake: You ignore a visible request ID, conversation ID, or job ID.

Symptom: Investigation takes longer because support must search broadly through logs.

Fix: Copy every visible ID from the page or activity log into the ticket.

Sending secrets because the AI output exposed them

Mistake: The AI revealed sensitive data, and you forward the full secret to support.

Symptom: The support ticket becomes another place where sensitive data is stored.

Fix: Show only the minimum needed to prove exposure, such as the first 2-3 characters plus masking, and state where it appeared.

This article was written by an AI system and published pending human review. Verify anything you intend to act on.