A Workflow for Running a UX Benchmark

A UX benchmark is really two jobs in one.

The first asks for the judgment of a senior designer: what to look for in this market, what a user of this brand actually cares about, what a competitor's homepage is quietly saying to the people it was built for.

The second asks mostly for hours. Open ten sites. Fill the same twelve cells for each one. Keep the notes clean enough to come back to later.

On her latest mandate, an online store for teenage girls aged 15 to 17, Haydée, designer at Lake House Group, used Claude to take on the second job. She kept the first.

The deliverable came together faster, and her hours went to the decisions that actually shape the recommendation. Here is her workflow, for you to use, step by step.

Step 1. Haydée builds the grid

Haydée started with the mandate. The brand talks to Gen Alpha girls, an audience that discovers brands through TikTok, trusts peers more than polished influencers, and often decides to buy while screenshotting a product to a group chat.

Each of those facts changes what a competitor's site needs to do well. From there, she wrote twelve criteria grouped into five dimensions: visual appeal, mobile experience, UGC integration, social proof, alignment with Gen Alpha codes.

Each criterion had to answer a concrete question tied to the mandate. "Is UGC present on the product page" is the checklist version. What Haydée wrote was closer to: does the site show photos that look like they came from a phone, in outfits a 16-year-old would actually wear, with captions that sound like a friend and not a brand.

A generic checklist flattens that distinction. A specific grid protects it.

This is where the value of a benchmark is decided. Knowing that UGC integration matters is easy. Knowing which version of UGC integration matters for this audience is the part that takes ten years of practice. Claude does not intervene here.

Step 2. Claude Code pre-fills the grid

With the grid set, Haydée gave Claude Code a list of direct and indirect competitors and asked it to work through each one, site by site, and populate the table.

Claude Code visited each site, read what was there, and returned a filled grid. Haydée went back over every cell.

Some entries were accurate and saved her a full site walkthrough. Some were close but missed the nuance: the model flagged UGC as present, for example, when what the site actually had was a block of stock photography labelled "community". A few were wrong in ways only a designer would catch.

The UX eye stayed hers. What changed was that she was now correcting a first draft instead of starting from a blank sheet.

This is where the time savings are concentrated. Visiting ten competitors and completing a twelve-criteria grid takes several hours of low-judgment work. Those hours came back. They went into the parts of the mandate where her expertise actually matters: interpreting the results and shaping the recommendation.

Step 3. Structuring the interview guide

The benchmark told Haydée what competitors were doing. It could not tell her why any of it worked. A site might have textbook UGC integration and still feel off to the audience. The grid would not catch that. Only the users could.

She used Claude to draft a first version of the interview guide around the themes she wanted to explore: how these users find brands, what makes them trust one, the role TikTok plays in the funnel, how the website fits into a journey that mostly happens somewhere else.

She then rewrote the questions. She cut the ones too abstract for a 15-year-old. She added questions that picked up on patterns she had already spotted in the benchmark, so the interviews could confirm or break her early reads.

The gain here is not time. It is clarity. Writing a first draft forces you to name what you are actually trying to learn, and in what order. By the time she sat down with a user, the guide had earned its shape.

Step 4. Synthesizing the findings

Once the interviews were done, Haydée passed her raw notes to Claude to pull out patterns.

The interviews carried the weight. A few things came through clearly. TikTok is the entry point for these customers, not the website. Photos of real people do more work than professional shoots. The site is not where the brand is discovered. It is where trust gets confirmed before a purchase.

Claude organized the material. Haydée ranked the findings, pushed back on the ones that felt too neat, and translated them into design recommendations.

This last step is where senior expertise earns its keep, and it is the step that cannot be handed off. "The site validates trust" can become ten different decisions. It can mean redesigning the product page to lead with reviews. It can mean adding a UGC feed above the fold. It can mean rewriting the FAQ in a voice that sounds like the brand. It can mean shortening the checkout so doubt has no time to set in.

Which version to defend depends on the brand, the mandate, the team that will implement it, and what the client can realistically ship in the next two quarters. That choice is a judgment, not a pattern.

What AI changes in the work

The split shifted where Haydée's hours went. The value of a benchmark was never in the grid. It was in the criteria picked, the questions asked, and the recommendations defended.

AI did not take any of that over. It absorbed the part of the work that was always the least focused on actual UX design and, by doing so, made her expertise come through.