UX feedback that survives past the Notion doc

This page is for UX designers and researchers running heuristic evaluations and usability passes. The workflow fits the rest of the use cases for visual product feedback hub, but UX feedback is where the cost of a bad artifact is highest, the finding is subtle, the rationale is everything, and the dev fixing it is two handoffs away from the person who saw the problem.

The problem

UX feedback lives in the wrong places. The heuristic eval goes into a Notion doc that gets read once and then goes stale. The usability findings get pinned as Figma comments that someone marks resolved before the implementation actually changes. The video walkthrough sits in a Loom library that nobody rewatches. By the time a dev opens the issue, the artifact has been summarized, compressed, or skipped entirely, and the dev is working from a fragmented version of what the UX reviewer actually saw.

The mental model mismatch you spotted on screen 4 is now one bullet on slide 7. The reason it mattered, the affordance, the user's expectation, the moment of hesitation, is gone.

The CobaltCapture workflow

Open the product. Walk the flow the way a user would. When you hit a UX issue, a mental model mismatch, a confusing label, a friction step, a primary action that doesn't read like one, switch to cobaltcapture.com and hit Capture screen. Drag a box around the exact element that's wrong. Hit Dictate and talk through the finding the way you'd say it out loud in a review:

"The 'Apply' button looks like a system action but the affordance reads more like a filter, users miss it because they're looking for a primary CTA below the form, not at the top. Heuristic 6, recognition rather than recall, but really it's a visual-hierarchy problem. The form has three styled buttons of similar weight and the actual submit is the least prominent."

Repeat for each finding. Every screenshot gets stamped with the source URL of the page you captured from. Hit Publish. The link is the UX review.

What the output looks like

The output is markdown with embedded cropped UI and your dictated rationale as paragraphs, the heuristic call, the user behavior you'd expect, the fix direction:

# Settings flow, UX review

Source: https://app.example.com/settings/notifications

![Apply button at top of form](https://cobaltcapture.com/r/uxr44218/img/1.png)

The 'Apply' button sits at the top right of the form, styled the same
as the filter chips next to it. Users scan top-to-bottom and expect a
primary CTA at the bottom. In the moderated test three of five
participants filled the form and then scrolled looking for a submit
button before noticing 'Apply' at the top. Move to bottom-right of
the form, or restyle as the only primary button on the page.

![Toggle label ambiguity](https://cobaltcapture.com/r/uxr44218/img/2.png)

The label 'Quiet mode' on the master toggle doesn't say what it
silences. Two participants assumed it muted in-app sounds; it actually
suppresses email and push. Heuristic 2, match between system and the
real world. Either rename to 'Pause notifications' or add helper text.

![Step indicator on review screen](https://cobaltcapture.com/r/uxr44218/img/3.png)

The review screen shows 'Step 3 of 3' but the next action submits and
exits. Users hesitated here because the indicator implies one more
screen. Either change to 'Review and submit' or add a confirmation
step so the indicator matches behavior.

Why this beats Notion docs

Notion docs are fine for synthesis. They are bad as the canonical record of a UX review, for two reasons.

First, the screenshots lose their source. The embedded image in the Notion page doesn't carry the URL of the screen it came from. Three weeks later, when the dev asks "which page was that toggle on again?" the trail is broken. CobaltCapture stamps every screenshot with the source URL of the page it was captured from, so the finding is always traceable to a live screen.

Second, Notion docs compress the rationale. The pressure to keep a doc scannable pushes the reviewer toward one-line findings. Dictation removes that pressure, you talk for thirty seconds per finding and get a paragraph of nuance the dev can actually use. The published CobaltCapture URL also works as a Loom alternative: the same audience reach as a video walkthrough, but in a format a dev (or the dev's coding agent) can ingest directly instead of rewatch.

Who this is for

UX designers and researchers running heuristic evaluations against Nielsen's 10 or a team-specific heuristic set. Product designers reviewing dev's implementation for affordance and labeling issues. UX leads doing pre-launch passes on a finished flow. Anyone whose feedback is "users won't understand this" and whose feedback then needs to inform an actual code change, often by a dev steering an AI coding agent.