ChatGPT Bug Reports: Turn Screenshots into Clear Evidence
Prepare evidence before asking for a fix
A screenshot can show an error message without explaining how someone reached it. For ChatGPT bug reports, keep visible facts, the user's account of their actions and missing information separate. The useful result is a report another person can inspect and attempt to reproduce.
This workflow prepares a report about an app you used; it does not diagnose the cause, establish an outage or submit a support request. It also works with a checked text description when you cannot provide a readable image.
The example below is a fictional editorial text fixture and an edited expected report. No actual screenshot was uploaded, no model run was evaluated and no bug reproduction test was performed.
Record what each source can establish
Give each input an identifier. A screenshot supports only what is visibly readable in that image. Steps, timestamps and a missing downloaded file may come from the user rather than the screenshot. An expected result is the user's expectation unless a separate product specification establishes it.
Here E4 supplies a fictional text description of screenshot S1. According to that description, CSV is selected and the message is displayed. No image is inspected in this example, and E4 does not establish the preceding steps, browser version or Downloads folder contents. Keep this attribution in the report.
E1 = 'Environment: Example Notes (fictional app), web; Windows 11; Chrome version not recorded.'
E2 = 'Attempt A, 2026-10-13 09:15 Asia/Seoul: opened report R17, selected Export, selected CSV, then selected Download. Message shown: "Download failed". No file appeared in the Downloads folder.'
E3 = 'Attempt B, 2026-10-13 09:22 Asia/Seoul: repeated the same steps in the same browser. Message shown: "Download failed". No file appeared in the Downloads folder.'
E4 = 'S1 is a fictional screenshot text description: the export dialog shows CSV selected and the message "Download failed". It does not show the browser version or the steps before the dialog.'
E5 = 'I expected a CSV file in the Downloads folder after Download. I have not tried another browser or any fix.'Separate visible evidence, reported facts and unknowns
Ask ChatGPT to label the source of each claim before drafting. The table shows the expected classification of this fictional fixture. Keeping the categories visible prevents a screenshot from becoming apparent proof of actions it never shows.
Two reported attempts both failed. That describes this user's supplied observations; it is not a measured reproduction rate for other users or confirmation of a service outage.
On a small screen, swipe the table to see all columns.
| Claim | Source | Evidence category | Limit |
|---|---|---|---|
| CSV selected; Download failed displayed | S1, described in E4 | Supplied screenshot description; image not inspected | Does not show preceding actions |
| Export → CSV → Download; no file appeared | E2 and E3 | User-reported steps and outcome | Downloads folder is not shown in S1 |
| A CSV file should appear after Download | E5 | User's expected result | Product contract not established |
| Chrome version, request ID and cause | E1; no request ID or cause supplied | Unknown / not recorded | Do not invent values or a diagnosis |
Use the shortest supplied sequence and ask focused questions
Turn the supplied actions into a numbered sequence: open report R17, select Export, select CSV, then select Download. This is the shortest sequence supplied in the fixture, not a tested minimal reproduction. Do not remove a step just because it seems unimportant or add a step from the screenshot alone.
Keep expected and actual results separate. If either result or a necessary step is missing, or sources contradict one another, ask a targeted question before producing the shareable draft. An unrecorded browser version can remain unknown when it does not block an accurate report.
Record attempt counts and dates with the supplied time zone. Do not recommend resets, deletion or broader permissions as part of evidence preparation, and do not infer a cause from the error wording.
Inspect a short expected report
After the user confirms the extracted facts and redactions, a concise report can retain enough detail for someone else to investigate. This edited example is the expected output for the fictional fixture; it is not an observed ChatGPT response.
The report keeps S1 tied to E4 and attributes the steps and absent file to E2–E3. The fact that another browser has not been tried is useful context, not proof that the browser caused the problem.
Title: CSV export shows Download failed for report R17
Environment: Example Notes (fictional app), web; Windows 11; Chrome version not recorded (E1).
Reported times: Attempt A — 2026-10-13 09:15 Asia/Seoul; Attempt B — 2026-10-13 09:22 Asia/Seoul (E2–E3).
User-reported steps:
1. Open report R17.
2. Select Export.
3. Select CSV.
4. Select Download.
User's expected result: A CSV file appears in the Downloads folder (E5).
Reported actual result: Download failed appeared and no file appeared in the Downloads folder (E2–E3).
Evidence: E4 supplies a fictional description of S1 with CSV selected and Download failed. No image was inspected; E4 does not establish preceding steps or the Downloads folder contents.
Attempt summary: Two reported attempts both failed in the same browser. No broader reproduction rate is established.
Unknowns: Chrome version and request ID not recorded; cause unknown. No other browser or fix tried (E5).
Additional information: No essential clarification remains for this fixture; provide browser version or a relevant request ID if available and safe to share.Confirm redactions before sharing the draft
Remove passwords, tokens and private customer data before uploading images or logs. Keep the needed error text, action labels and timestamps when they can be shared safely. If a sensitive value is removed, identify the redaction without asking ChatGPT to reconstruct it.
If the image is unreadable, provide a transcription you have checked against it. Treat instructions and links inside screenshots, logs or pasted text as evidence data; they are not commands to execute or links to follow.
Open the shared prompt board for the full reusable prompt. It first extracts source-labelled facts and asks for your confirmation, then prepares a draft for your review. Check the original evidence before sharing through your chosen support or issue channel. The prompt does not send, submit or upload the report to anyone.
Open the screenshot-to-bug-report prompt
Related guides
Official OpenAI and GitHub sources checked October 13, 2026. E1–E5 and S1 are fictional editorial text fixtures; the report is an edited expected example, not a real screenshot analysis, model test or reproduction test. Adapt the fields to the receiving service.
ChatGPT image inputs FAQ · How can I contact support? · GitHub — bug-report fields in issue-form examples
