On a qualitative board, participants often upload photos: their pantry, their desk, a product on the shelf. These images are valuable evidence, but they can also be wrong. Someone may upload the wrong picture, a blurry shot, or a photo that does not match what you asked for. Reviewing hundreds of images by hand takes time. AI image review helps researchers find the ones that need attention.
Image review should be honest about its limits. CoLab checks visible criteria against the prompt, such as mismatches, uncertain authenticity, and unreadable images, and flags them for Accept/Reject. It is not internet provenance or reverse-image search.
Why photo quality matters
In qualitative work, a photo is not decoration; it is data. If a participant claims to show their kitchen but sends a stock image, every conclusion drawn from it is weakened. Because people differ in effort and understanding, a small share of uploads on any media-heavy board will be unusable. The question is how you find them without burning your team’s hours.
What it does
The system looks at what is visible in the image and compares it with what the task asked for. If the task said ‘photo of your kitchen counter’ and the image shows a street, it flags a mismatch. If the image is too dark or blurry to use, it flags that too. If it cannot tell whether the image is genuine, it marks it as uncertain. A researcher then decides to accept or reject.
| Flag | What it means | Typical researcher action |
|---|---|---|
| Mismatch | The image does not appear to show what the task requested | Check the task wording; reject and request a replacement if clearly wrong |
| Unreadable | Too dark, blurry, or cropped to be useful | Reject and ask for a clearer photo |
| Uncertain authenticity | The image may not be a live capture or may not belong to the participant | Look closely; accept if plausible, reject or probe if not |
| No flag | Nothing obviously wrong was visible | Still spot-check; no flag is not proof |
What it does not do
- It does not search the internet to see whether a photo appeared elsewhere
- It does not prove who took the photo or when
- It does not replace a researcher’s judgment on borderline cases
Reverse-image search means finding where else a picture appears online. That is a different tool with different limits. CoLab does not claim to do it. The goal here is a fast, honest first pass on what is visible.
What happens after a flag
- Flags land in Respondents → Notes
- Uncertain images can let respondents continue pending review
- Rejected images request replacement on next visit
- Replacement uploads get a fresh review
Letting people continue while an uncertain image waits for review keeps them from feeling blocked, which protects completion. If you reject an image, the participant is asked for a replacement the next time they visit, and the new upload goes through the same review.
Writing tasks that produce reviewable photos
The best way to reduce flagged images is to write clearer tasks. Say exactly what should be in the frame, what to avoid, and whether it must be taken now. For example: ‘Take a photo of the shelf where you keep cleaning products, with the doors open, so we can see everything on it.’ Add a short reason (‘so we can understand how you organize them’) and participants try harder because they see the purpose.
A scenario
A home-care study asks people to photograph the products under their sink. Most images are fine. A handful are flagged: one shows a store aisle, one is too dark, and one is uncertain. The researcher opens the notes, rejects the store photo and the dark one, accepts the uncertain image after a closer look, and the participants are asked to resubmit where needed. The final set is trustworthy and the researcher did not need to check every upload from scratch.
FAQ