LinkedIn Post Ideas for UX Researchers

Six engineers loved this feature. Watched it die in a single 45-minute usability session. The feature: a bulk-edit tool the engineering team had pushed hard for, convinced it would save power users hours. Clever technically — batch operations, keyboard shortcuts, the works. We recruited five power users for a study before the wider rollout. All five completed the primary task. All five, unprompted, said some version of "I don't think I'd use this." The reveal came in the debrief. Power users didn't want batch efficiency on this particular workflow — they wanted control, one item at a time, because the cost of a bulk mistake was too high in their context. The feature solved a problem engineering assumed existed. It didn't, not for this segment. We killed it before the wider build. Six engineers who'd spent three weeks on it were, understandably, not thrilled watching a 45-minute session override their instinct. But the roadmap saved probably eight more weeks of building the full version, plus the support cost of an unused feature nobody could explain internally later. Nothing argues for research like watching a confident internal assumption meet a real user in real time. The saved roadmap is the whole pitch.

Someone quoted "five users finds 85% of usability issues" at me in a planning meeting, as justification for skipping research entirely on a redesign. I had to walk through why that number doesn't mean what people think it means. The original research (Nielsen, on relatively homogeneous user groups and simple, well-scoped tasks) found five users surfaces most usability issues — for that specific context. It says nothing about: — Complex or highly variable user populations. Five enterprise admins across five different industries are not five instances of the same person; they're five different contexts. — Preference or value questions, as opposed to pure usability breakage. Five people can't tell you whether a pricing model resonates broadly. — Statistical claims about frequency or prevalence. Five sessions tell you a problem exists. They don't tell you if it affects 5% or 60% of users. Where five genuinely works: a single, well-defined task, a relatively homogeneous user group, early-stage usability testing where you're hunting for breakage, not measuring prevalence. Where it badly doesn't: anything you're about to make a resourcing bet on, anything spanning meaningfully different user segments, anything where you need a number for a business case. The heuristic became a shortcut people use to avoid a harder conversation about what question they're actually trying to answer. Ask that question first. The sample size follows from it, not the other way around.

Nobody attends research sessions because you told them to. I stopped mandating it two years ago. Attendance went up. What actually works: — Highlight reels under two minutes. Not the full 45-minute recording — a tight edit of the three most damning or delightful moments. Sent in Slack the same day, while it's still relevant to whatever's in flight. — Watch parties, with actual pizza. Scheduled around a specific high-stakes study, framed as "come see what almost broke the onboarding flow" rather than "research readout." Framing matters more than the food, but the food doesn't hurt. — One devastating clip in sprint review. A single 20-second moment of a user struggling with a flow the team just shipped, played live, no commentary needed. This is the single highest-leverage move I have. It does more for research buy-in than any deck. — Personal invites tied to specific tickets. "The person in session four hit exactly the bug you're working on" gets an engineer to a session mandates never will. None of this is about compliance. It's about making the moment more interesting than whatever else is competing for their attention that hour. Exposure, not obligation, is the whole strategy.

Ran the same twelve interview transcripts through an AI synthesis tool and through our own manual analysis. Compared the outputs side by side. The AI summary was fast — under two minutes for all twelve transcripts — and it wasn't wrong, exactly. It correctly identified the top three most-mentioned pain points, cleanly categorized, well-organized. What it missed: a pattern one participant described almost in passing, mentioned by only two of twelve people, about a workaround they'd built because a specific integration silently failed under a rare condition. Low frequency. High severity. The AI's frequency-weighted synthesis under-surfaced it because it wasn't common language across the set — it was a specific, technical, easy-to-miss detail buried in otherwise unrelated sentences. That single finding, when we dug into it manually, turned out to affect a disproportionately valuable customer segment and became the top engineering priority the following sprint. The AI is genuinely good at "what did most people say." It's weak at "what did almost nobody say, but does it matter enormously." Frequency and importance are not the same axis, and most synthesis tools currently optimize for the wrong one. I still use AI synthesis for a fast first pass. I don't skip the manual read anymore. This is exactly the finding I'd have missed if I had.

A participant started crying twenty minutes into a session about a workplace tool tied to a layoff she'd recently experienced with a previous employer. Here's exactly what I did, because I didn't have a script for it going in. First: paused the recording immediately, before saying anything else. Consent covers research participation, not an unplanned emotional moment — that needed a separate, explicit check-in. Second: asked directly if she wanted to continue, take a break, or stop entirely. Made clear all three were fully fine and she'd still be compensated regardless. She chose to continue after a few minutes, and the session finished normally. But the debrief afterward mattered more than the protocol did. What I changed after that session: every consent form now includes an explicit line that participants can pause or stop at any point with no compensation penalty, stated verbally at the start of every session, not just buried in the form. I also now build in a genuine mid-session check-in for any topic with even mild emotional risk, not just ones that seem obviously sensitive. Research ethics training covers consent forms. It covers almost nothing about the actual moment something goes off-script in front of you. Handling it with care, not panic, is the real craft — and it's the part nobody teaches you until it happens.

Three research reports I wrote last year: 38 pages, 44 pages, 51 pages. Combined views after the initial share: I genuinely don't know, because nobody could tell me they'd read past the executive summary. The fourth report: one page. A single recommendation, three supporting data points, one quote, one next step. Sent to four stakeholders. All four responded within the day. Two forwarded it to their own teams. One turned it into an actual roadmap ticket that week. What changed wasn't the rigor behind the research — the one-pager represented the same depth of work as the 40-page reports, just compressed to the decision that mattered. What changed was respecting that stakeholders don't want the research. They want the answer, with enough evidence to trust it and act on it. The 40-page deck format exists because it feels thorough to produce. It's optimized for the researcher's sense of completeness, not the reader's actual need. I still do the full analysis — I just stopped shipping all of it as the deliverable. Every study I run now produces two things: the full analysis, archived for anyone who wants depth, and a one-page decision brief, which is the only thing anyone actually reads. What format finally got your research read?

We democratized research eighteen months ago — trained PMs and designers to run their own studies, gave everyone access to our panel and recording tools. Study volume roughly tripled. Decision quality, honestly, got worse for a stretch. Here's the nuanced version, not the hot take. What improved: coverage. Small questions that never would've made it onto a research team's queue — a single flow tweak, a quick concept check — now got tested instead of shipped on instinct. That's a real win. What got worse: rigor on anything that mattered. Leading questions in scripts nobody reviewed. Sample sizes treated as statistically meaningful when they weren't. Confirmation bias running unchecked, because the person who wanted a feature was also the person interpreting whether users wanted it. The guardrails that fixed it, mostly: a lightweight script review from a trained researcher before any self-serve study launches. A tiering system — small, low-stakes questions go fully self-serve; anything touching a resourcing decision above a certain size requires a researcher partner. Training on leading-question detection, which turned out to be the single highest-leverage two-hour workshop we ran all year. Democratization isn't the wrong move. Democratization without any rigor guardrails is. The nuance matters more than either side of this debate usually admits.

Needed eight compliance officers at mid-market financial firms for a study. Had zero in our existing panel. Here's the actual week it took, unglamorous version. Monday: posted in three relevant LinkedIn groups. Zero responses by end of day. Tuesday: cold DM'd 40 people with "compliance officer" in their title, personalized first line referencing their actual company. Six responses, two maybes. Wednesday: increased incentive from $75 to $150 for this specific recruit, since the segment was clearly harder to reach than our standard panel assumptions. Three more responses. Thursday: asked our two confirmed participants for a referral, explicitly framed as "know anyone else in a similar role." Got two warm intros — both converted, no incentive negotiation needed because trust was already established. Friday: one recruiting agency contacted as backup, quoted $200/participant plus a placement fee, used for the final slot when organic outreach stalled with two days left before the study window closed. Final count: eight participants, five days, mix of cold outreach, incentive escalation, and warm referral. No single channel got us there alone. The lesson: niche B2B recruiting isn't a tool problem, it's a persistence-plus-referral problem. Budget the time, not just the incentive line item. What's your best niche recruiting tactic?

Seven signs, from experience, that a research finding will actually change the roadmap — versus get nodded at and forgotten. 1. A named decision is already waiting on it. If nobody's blocked on an answer, even a great finding has nowhere to land. 2. A specific stakeholder watched the session live. Secondhand findings get debated. Witnessed findings get acted on. 3. The finding has a cost attached. "Users are confused" is ignorable. "Users are confused, and it's costing us an estimated 200 support tickets a month" is not. 4. It contradicts a stated assumption from someone senior, directly and specifically. Vague findings get filed. Findings that name the wrong assumption get addressed. 5. There's a clip, not just a quote. Video of a real person struggling moves rooms that a bullet point never will. 6. The fix is scoped small enough to start this sprint. Findings requiring a six-month roadmap shift get deprioritized by default, regardless of how important they are. 7. Someone besides you is willing to champion it internally. If the finding lives only in your report, it dies with your report. The craft researchers aren't taught in grad school: findings don't create change. Findings paired with the right conditions do. Learn to engineer those conditions, not just the finding itself.

Researchers — what's the finding stakeholders dismissed, right up until it cost them something real? Mine: I flagged, eight months before launch, that our onboarding flow's core assumption — that users would arrive already knowing our product category — was wrong for over a third of our actual signups. Presented it with session clips, quotes, a clear recommendation to add a category-explainer step. The response: "that's not our target user," repeated in three separate meetings. We shipped without the change. Six months post-launch, support tickets and early churn data showed exactly the pattern I'd flagged — a third of churned users citing confusion about what the product even was, in their own words, in exit surveys. We added the explainer step eleven months later than the research had recommended it. Churn in that segment dropped measurably within two months of shipping it. Nobody apologized, exactly. But the finding got treated with real weight the next time I brought something similar, because it had already been proven right the expensive way. Vindication doesn't feel great in the moment. It's also the best argument you'll ever have for listening to research before the number confirms it. What's yours?