User interview transcripts and competitor research pile up in a shared folder faster than anyone can make sense of them. In design research, the bottleneck was never gathering the material — it’s pulling patterns out of it afterward. Used as an assistant rather than a shortcut, ChatGPT can shave real time off that synthesis stage.
1. What Actually Changes When AI Joins the Synthesis
Reading ten interview transcripts by hand to find the patterns that repeat across them can eat a full day. Paste the transcripts into an AI tool and ask for a structured summary, and a rough draft appears within minutes. But that draft is exactly that — a fast first pass, not a finished insight. The real work of research synthesis still comes down to a person going back through the source material and filling in the nuance and context the model inevitably strips out.
2. In Practice — Summarizing Interviews and Benchmarking Competitors
Getting Cleaner Interview Summaries
Give the model an explicit role — “act as a research assistant” — and spell out the output format you want: how many direct quotes, how often a point should recur before it counts as a pattern, and what falls outside scope. The more specific the request, the more usable the result. Something like: “From this interview, list every complaint that came up more than once, each with a supporting quote, and keep it under five items” works far better than a vague ask for a summary.
Structuring Competitive Benchmark Data
When you’re comparing several competing products, asking the model to lay the material out as a table makes side-by-side comparison legible at a glance. Asking it to separate “patterns common across the field” from “what’s distinct about this one product” gets you most of the backbone of a benchmarking report in one pass. That synthesized output feeds directly into design briefs and design-system direction afterward — for the broader question of folding AI tools into a design system, see our piece on bringing AI into a design system.
Interview Summaries
Assign the model a “research assistant” role and specify the output format precisely — quote count, recurring themes, anything outside the intended scope.
Competitive Benchmarking
Ask for a table, then request “shared patterns” and “product-specific differences” as separate columns for a benchmark report backbone in one pass.
💡 Pro tip — ChatGPT will confidently invent details that aren’t in the source material, and this gets worse when you ask it to summarize several interviews at once — it can blend quotes from different respondents into what reads like a single, coherent statement. Before any summarized insight goes into a report or a presentation deck, go back to the original transcript and confirm every quote and figure. If you want to turn synthesized findings into something more systematic, it’s worth looking at how design tokens already structure that kind of information for reference.
3. Why AI Synthesis Fits the Affinity-Mapping Stage So Well
Writing everyone’s observations and interview notes on small notes, sticking them to a wall, and clustering the similar ones to surface patterns is a technique researchers have called affinity diagramming for decades. It used to mean physically moving sticky notes around by hand. Feed the raw interview notes to a model and ask it to “cluster similar statements and title each group,” and you get a first-pass draft of that clustering almost instantly. The judgment call of where the boundaries between groups actually belong, though, is still something a researcher has to check.
A Common Misreading
Some teams take the groups and titles the AI produces and ship them as-is. That’s risky, because the model tends to cluster by surface-level wording — sentences that sound similar can get grouped together even when they mean something entirely different in context. Treat AI-generated clusters as a rough draft for a human to rearrange, never as the final call.
💡 Pro tip — Don’t let the AI name the groups. Have a person actually read the quotes inside each cluster and rename it in the team’s own language — the result reads far more persuasively in the final report.
This approach traces back to an analysis method called the KJ technique, developed by the Japanese cultural anthropologist Jiro Kawakita, and it later spread widely as a way to organize qualitative data across many fields. From sticky notes on a wall to AI-assisted synthesis today, the core idea hasn’t changed — it’s not about summarizing material without judgment, but about structuring it into a form a person can meaningfully revisit.
4. Closing thoughts
Specify the output format and item count precisely when requesting a summary
Always cross-check quotes against the original transcript
Verify that statements from different respondents haven’t been blended together
For table-format requests, define the columns first, then have the model fill them in
ChatGPT’s capabilities and usage policies keep evolving, so it’s worth checking the official OpenAI site periodically. As an assistant that cuts down the time spent organizing design research, it’s genuinely effective. But the final call on which insight actually matters still belongs to the researcher — that’s not a role worth handing off.
Design Daily Life · Notes on design, daily