Can the 91% figure be trusted?
The direction is credible; the exact proportion deserves a discount. The sample is 900-plus designers, but the pages I read did not give fieldwork dates or sampling method. People who answer a design-community survey may be more invested in AI than the average designer. That is my inference, not something the report tests.
Anthropic being a report partner is worth knowing, though a year-over-year shift like 54% to 91% does not hinge on any single tool's ranking.
Are State of AI Design's 62% and UXtools' 52.2% the same measurement?
No, and they can't be compared directly. The samples differ and so does the question wording. What one can say is that both put output quality among the main obstacles.
In UXtools the top blocker was actually time to learn tools (55.7%), with output quality third. Treating 62% as "everyone is stuck on quality" would overstate it.
Does only 1.4% giving full trust mean 98.6% distrust AI?
No. The UXtools options are graded: 1.4% chose full trust with no oversight and 2.4% don't use AI output, while everyone else sits somewhere in the "needs some checking" range.
The more accurate reading is that almost nobody uses AI output with zero checking. That supports the earlier point: checking is a real workload that adoption figures never count.
How do I fold verification cost into tool evaluation?
Run a one-week experiment. For each tool, record two numbers: time saved, and time spent fixing, redoing and checking its output. The difference is the tool's net benefit.
Add one more column: if the output is wrong, who downstream finds out? For tasks where errors surface only in production (code, accessibility, design systems), allow a wider checking budget.
The second annual State of AI Design report (the 2026 edition), published by Designer Fund with Foundation Capital, draws on more than 900 designers at startups, enterprises and agencies across 60-plus countries, plus over 20 leadership interviews. The headline number: designers using AI in their work at least weekly rose from 54% in 2025 to 91%. Of those, 75% use it every day and 16% several times a week. The average number of off-the-shelf AI tools per designer went from 3 to 7.
It is easy to dismiss 91% as a survey that only asked whether people use AI, but 75% using it daily makes "just trying it out" hard to defend. The fastest-growing uses are worth a look too: code generation and front-end implementation up 31 percentage points on last year, wireframing +27, design systems and components +24, Design QA and accessibility +16, developer handoff +12. None of these is a low-stakes "suggest me a headline" task. When the output is wrong, someone downstream pays.
A related figure: 50% of respondents have pushed AI-generated code to production. By company stage that is 68% at early-stage companies, 50% at growth stage and 33% at public companies. The report does not explain the gap. Looser review processes are a plausible guess, but that is speculation, not a finding.
In the same report, 80% say reliable, high-quality output is what makes a tool stick, and 62% name inconsistent or unreliable output as their biggest challenge. Adoption is climbing, and the problem most people care about remains unsolved.
The UXtools Spring 2026 survey (1,478 designers and builders across 18 regions, fielded March 14 to April 6) gives a consistent picture from another angle. Only 1.4% chose "full trust, no oversight," and another 2.4% don't use AI output at all; everyone in between applies some level of checking. Their top three blockers were time to learn tools (55.7%), too many tools (53.0%) and AI output quality (52.2%).
First, disclosure. State of AI Design lists Anthropic as a report partner, and its general-purpose AI tool numbers put Claude at 78%, ChatGPT at 65% and Gemini at 48%, with Claude first. That doesn't make the numbers wrong, but readers should know the relationship exists. Second, the two surveys use different samples and wording, so 62% and 52.2% can't be compared directly; all one can say is that both put output quality among the main obstacles. Third, the pages I read did not give State of AI Design's fieldwork dates or sampling method, so 91% describes designers who answered this survey, not necessarily the whole industry.
Going from 3 tools to 7 means subscription costs pile up, and each extra tool is one more place where you have to check the output. When evaluating a tool, don't just record how fast the first draft appears; record how long it takes to reach something accepted. Over one week, note the hours each tool saves and the hours you spend fixing and redoing its output. If the difference is negative, that subscription is the one to cut.
If your work sends AI output into production, make sure you have a checking gate first and speed things up second, not the reverse. The report shows half of respondents have shipped AI-generated code, but there is no data on how thoroughly that output was verified.