# Comparing AI Skill Focus Area · Comparing Block · Decision Map · Oct 23, 2026 (v1.1) --- ## 1. What the Skill Does The Comparing skill helps teams repeat the same measurement while one thing changes, then read the difference. It is the third block in the Focus area of Glare's Decision Map, after Initiatives and Variables and before Bets. Measure designs one test. Comparing runs that test again across versions, competitors, segments, time periods, or another variable. A single score only means something next to what it is compared with. Is 72% task success good? Only a comparison can say. Every comparison measures two things for each option, the same way: - **Use:** which option works better, such as first-click success, task success, comprehension, or errors - **Prefer:** which option people would choose, asked as a side-by-side choice, likely to choose (top-two box), trust, or pull-away Without a Prefer measure, a comparison can't lead to a bet. Desirability words alone, such as "it looks nicer," don't count as Prefer. The skill helps the team pick a comparison design. Each design says what its result can prove. | Design | How it works | What the result can prove | |---|---|---| | Randomized A/B | Two versions shown to randomly split audiences | The design caused the difference, when significant | | Randomized multivariate | Several combinations of elements tested at once | Which combination works best, when significant and the sample is large | | Within subjects | The same people see every option | A real difference for the same people; vary the order | | Separate samples | Different people see each option | A real difference between options; cause not shown | | Before and after | The same measure taken again after a change | An observed change, not proof of cause | | Benchmark | Your experience measured next to others or a standard | Relative standing only | | Segment split | One study, with results split by group | Groups differ; observational | ### The Same-Measurement Rule Only one thing changes. The UX metric, the technique, the kind of audience, and the Prefer question stay the same on every side. If two things change at once, split the comparison or name the confound in the readout. A comparison that changes the audience and the design together can't say which one moved the number. ## 2. Business Benefit Decisions stall when teams look at results one at a time. A concept scores well, but is it better than what is live today? Comparing answers that, so the team has a reason to choose. This helps teams: - see what got stronger or weaker between versions - know whether a difference is real or noise - name the tradeoff instead of chasing the highest score - see where one audience responds differently from another - bring Bets a readout it can commit to ## 3. Skill Output When used correctly, the skill produces a comparison readout. The readout shows: - what was compared, and the one variable that changed - a **Use line**: which option worked better, with every option's number side by side - a **Prefer line**: which option people would choose, and how it was asked - the tradeoff that showed up - what the design can and can't prove - the next step Every signal in the readout is tagged Need, Use, Prefer, or Adopt. The example below shows how this works for a mobile banking dashboard. | Field | Example Output (Mobile Banking Dashboard) | |---|---| | Decision It Supports | Which home screen to take into the next release | | Variable | Iteration: current home screen vs. redesigned home screen | | Held Fixed | Same task (find your balance), same audience (100 habitual mobile banking users per version, Helio), same Prefer question | | Design | Separate samples: shows a real difference between versions, not what caused it | | Use Line | First-click success on balance: Redesign 79%, Current 61% | | Prefer Line | Likely to choose this home screen (top-two box): Redesign 68%, Current 49% | | Tradeoff | The redesign hides shortcuts that power users rely on | | Baseline and Cadence | Current version is the baseline; repeat after launch and again at +90 days | | Failure Mode to Watch | Changing the layout and the copy between versions in the same round. If two things change, the readout can't say which one moved the number. | | Next Step Handoff | → glare-focus-decisions (Bets) to commit to the redesign and set its predicted outcome | The output connects to the other Focus blocks: - Initiatives names the need, goal, and metric the comparison serves - Variables names the one thing that changes - Bets uses the Use line and the Prefer line to call a winner, or names a tie When Use and Prefer disagree, the skill names it and does not force a winner. Bets applies the winner rule. ## 4. Prompt Strategies The prompts below show different ways to use this skill. Each example uses a mobile banking dashboard update. ### Prompt 1: Setup Entry: Design a fair comparison *"We want to compare our current mobile banking home screen with a redesign before the next release. Using the glare-focus-comparing skill, name the variable, list what must stay fixed, pick the comparison design, and tell us how to measure both Use and Prefer the same way for each version."* **Why this works:** It sets up the repeat before any data comes in, so the result can actually be compared. It also makes sure Prefer is measured, so the comparison can lead to a bet. **Best for:** - planning a test round with more than one version - A/B test design - making sure a comparison can support a decision ### Prompt 2: Reading Entry: Make sense of a mixed result *"Our mobile banking comparison came back. The redesign won first-click success, 79% vs 61%. But likely to choose was 52% for the redesign and 55% for the current version. Using glare-focus-comparing, write the readout with a Use line and a Prefer line, name the tradeoff, say what this design can and can't prove, and tell us whether this is ready for a bet."* **Why this works:** When Use and Prefer point different ways, teams tend to quote the number they like. This prompt puts both lines side by side and names the tie instead of forcing a winner. **Best for:** - results where two metrics disagree - preparing for a review where someone will push back - deciding whether another round is needed ### Prompt 3: Repeat Entry: Track change over time *"We launched the new mobile banking home screen in March. Using glare-focus-comparing, set up a before-and-after comparison against our pre-launch baseline: what to measure, when to repeat it, what size of change counts as real, and what this design can and can't prove about the launch."* **Why this works:** Before-and-after shows an observed change, not proof of cause. This prompt sets the baseline and cadence up front and keeps the readout honest about what it can claim. **Best for:** - post-launch checks - seasonal or campaign changes - building a benchmark the team repeats every quarter --- *Glare Framework · glare-focus-comparing · Focus Area. Handoffs: glare-focus-initiatives · glare-focus-methods (Variables) · glare-focus-decisions (Bets) · glare-measure-findings (single test, no options)*