User research synthesis the rest of the team will actually read
Pause the recording on the moment, crop the screen, talk through what happened, publish a single markdown URL the PM, the designer, and the engineer all open.
This page is for UX researchers who run interviews, watch the recordings back, and have to turn six hours of footage into something the rest of the company will act on. The pattern fits anywhere on the use cases for visual product feedback hub, but research synthesis, many small moments across many sessions, each one visual, each one needing context, is where the markdown-with-screenshots format pays off the most.
The problem
Research synthesis lands in one of two shapes. The first is a 40-slide deck with a title card, an agenda, four "key themes" slides, and 35 screenshots floating in a Miro that nobody else has access to. The second is a 12,000-word doc with a table of contents and section headers like "Observation 4.2.1" that the PM scrolls past on the way to a recommendation that isn't there.
Either way, the team that needs the findings, the PM scoping the next sprint, the designer mocking the fix, the engineer estimating the rebuild, skims, misses the key moments, and builds off intuition. Six hours of recordings, eight participants, a week of synthesis time, and the artifact gets read for thirty seconds.
The CobaltCapture workflow
After the sessions wrap, queue up the first recording. When a participant hits the moment that matters, the dead-end in onboarding, the confused stare at the empty state, the rage-click on a disabled button, pause the playback. Open CobaltCapture in a new tab and hit Capture screen, picking the recording window. Drag a box around exactly what the participant saw at that moment. Hit Dictate and talk through the observation: "Participant 4, two minutes into the trial flow. She read the headline three times before realizing the CTA was below the fold. This is the third session out of five where the hierarchy on this screen lost the participant."
Repeat for every moment across every recording. The page you captured from, the recording playback URL, the Dovetail tag, the timestamped Loom, gets stamped on every item as a source so a teammate can jump back to the raw footage.
Hit Publish. You get a single URL like cobaltcapture.com/r/<slug> with one insight per section. Drop it in the readout doc, the sprint planning agenda, or a prompt to an AI assistant helping you draft interventions.
What the output looks like
The synthesis is a structured markdown document. Every finding has a heading, an embedded session screen, the source URL or timestamp, and your dictated observation as a paragraph that ties participant behavior to a product implication:
# Onboarding research synthesis, trial-to-activation flow
Source: https://recordings.example.com/p4/session-2?t=2m14s
Participant 4 read the headline three times before scrolling. The CTA
is below the fold on a 13" MacBook, which was her hardware. Three of
five participants did the same scroll-search behavior on this screen.
Implication: the primary action needs to be in the initial viewport
for the laptop sizes that dominate our analytics.
Source: https://recordings.example.com/p6/session-1?t=5m02s
Participant 6 clicked the disabled Continue button four times in eight
seconds before realizing the email field was the blocker. The disabled
state reads as "loading" to her. Implication: error state on the
field, not a disabled state on the button.
Source: https://recordings.example.com/p2/session-3?t=11m47s
Participant 2 sat on the empty workspace for forty-three seconds. She
said out loud "I guess I just... make one?" Implication: the empty
state needs a guided first-action, not a marketing illustration.
Why this beats a 40-slide deck
Decks bury insights inside a structure designed to be presented, not read. Nobody on the engineering team is sitting through your synthesis presentation. They are skimming the readout doc the morning of sprint planning, looking for the three things that change what they were going to build.
CobaltCapture is one URL per insight, scannable in two minutes, embeddable as a link in the PRD, and quotable by paragraph in a Slack thread. The PM can pull a single observation into a sprint ticket. The designer can paste the cropped frame straight into Figma. The engineer can read the markdown source directly.
The format also works for the next thing the PM is going to do, which is increasingly to paste the synthesis into ChatGPT or Claude and brainstorm interventions. Markdown with structured headings and embedded images is exactly the input shape large language models read best, see structured feedback for LLMs for why structure beats prose for any LLM-mediated workflow.
Who this is for
UX researchers turning interview recordings into team-readable artifacts. ResearchOps leads who own the synthesis template and want a format every research project ships in. PMs running their own customer-discovery sessions who do not have a research team to lean on and need to get findings into the next sprint without writing a deck nobody will read.
Capture your first review.
About a minute from open tab to a shareable URL your agent can ingest.
Start capturing