Get Structured Findings From a ChatGPT Image Review

A screenshot review that ends in one long paragraph is hard to use. You cannot sort it. You cannot forward it. You cannot check items off. The same review returned as numbered findings or a table becomes something you can act on the same day.

This is not a stylistic choice. It changes how ChatGPT thinks about the image. When you ask for a table with clear columns, ChatGPT is forced to break the reply into rows, each with a single point. That structure catches issues that a paragraph would smooth over.

What a structured finding really is

A structured finding is one clear observation, one location on the image, and one recommended action. You could put it in a numbered list or a table. What matters is that each row stands on its own, so you can copy any single row into a ticket or a message.

When to ask for structured findings

• Reviewing a webpage screenshot for accessibility problems
• Checking a marketing image for brand or legal issues
• Auditing a dashboard for missing labels or wrong colors
• Comparing a design mock to a rendered page
• Reviewing a photo of a physical product for defects

Prompt for a page accessibility review

Screenshot: a landing page.
Task: review for accessibility issues in the visible area only.
Return a table with four columns: id, location on page, issue, suggested fix.
One row per issue. If none, say "no issues found".
Do not include general advice that is not tied to something visible.

Why this prompt works

It defines the scope, the format, and the rule against generic filler. Without the last line, ChatGPT tends to add extra rows with vague advice that has nothing to do with the image, which makes the whole table less trustworthy.

Prompt for a dashboard audit

Screenshot: an internal metrics dashboard.
Task: list every chart or number that is unclear.
Return as a numbered list.
Each item must include: the chart name if visible, what is unclear, and one concrete fix.
Skip items you cannot see well enough to describe.

Prompt for a product photo review

Photo: a wireless earbud held in a hand.
Task: identify any visible defects such as scratches, dust, or misalignment.
Return a table with columns: area, defect, severity from low to high, suggested next step.
If you are not sure, mark severity as unclear rather than guessing.

What to put in the request

• The type of image and what area to review
• The exact table columns or numbered format you want
• How to handle uncertain findings
• A rule against filler that is not grounded in the image
• A maximum number of items when you want a short summary

How to refine the findings

Pick a single row and ask ChatGPT to expand it. For example, item four looks vague, describe the exact area and add a fix I can copy into a ticket. Working row by row is faster than asking for the whole table to be rewritten.

Common mistakes

• Asking for a general review without saying what columns you want
• Letting ChatGPT invent rows that are not tied to the image
• Mixing severity words such as high, big, and major in the same table
• Skipping the id column when the list becomes long
• Trusting the count of rows as if it means the review was complete

How to check the table before you use it

Pick two rows at random. Open the image and confirm you can see the thing the row describes. If both check out, the rest is probably safe to skim. If either row is invented, ask for the whole table again with tighter rules and drop the bad rows.

Takeaway

Structure is the difference between a nice comment and a useful review. Say the columns, forbid filler, and every review becomes a checklist you can share.