The usual takes on AI in UX are that it is magic or that it is a waste of time. Neither one helps you get work done. I have burned plenty of hours pushing LLMs at tasks they cannot do, and I have also found the places where they genuinely speed up client projects.
If slow research cycles are your problem, the bottleneck is rarely the thinking. It is the grunt work around the thinking. Two years of daily use taught me that AI in UX has nothing to do with finding a perfect prompt. Treat the model like an enthusiastic intern with no life experience. You would not let them ship code unsupervised, but they are great at sorting the mail.
Why AI in UX is like managing an intern
An intern has the qualifications and none of the context. So I work to a fixed pattern: define the role (“act as a senior copywriter”), split the task into atomic steps, and say what a good result looks like as a rubric. I also make the model reason about its approach before it writes a word, which is what keeps the output from collapsing into generic filler.
Synthesizing user research is the case I lean on most. I used to skip open-ended survey questions because reading hundreds of answers was miserable. Now the transcripts go into a project space and the recurring themes and objections come back in minutes. The gotcha is trust, so I require a direct quote behind every theme and check them myself.
On that note, I recently wrote about how to master robust virtual personas with this simple hack. Functional personas built from real data beat the marketing fiction most of us used to pass around.
Design audits and predictive visuals
AI is not ready to design your final UI. Its choices still land slightly off. Auditing is another story. Tools like Attention Insight are trained on thousands of eye-tracking studies and predict where users will look with 90-96% accuracy. That is a handy thing to have in the room when a stakeholder wants five CTAs crammed into the header.
For e-commerce I reach for Baymard’s UX Ray, which checks a site against their database of more than 200,000 hours of usability research. It is no substitute for a human audit, but it surfaces the easy wins, checkout friction especially, that would otherwise take me days to document.
Automating the small stuff
On the development side, I use AI for the throwaway tools I used to sink hours into. A client recently wanted a “Last Evaluated” date on their blog posts so readers could see the content was still maintained. Rather than hunt for a plugin that drags in half a framework, I had AI help me draft a small function.
<?php
/**
* Automatically append 'Last Evaluated' date to UX articles.
* Prefixing with bbioon_ for safety.
*/
function bbioon_append_last_evaluated_date( $content ) {
if ( is_single() && in_the_loop() && is_main_query() ) {
$updated_date = get_the_modified_date();
$evaluated_html = '<p class="ux-meta">Last UX Review: ' . esc_html( $updated_date ) . '</p>';
return $evaluated_html . $content;
}
return $content;
}
add_filter( 'the_content', 'bbioon_append_last_evaluated_date' );
Nothing clever about it, but it shipped in 30 seconds. Same story with ROI calculators and scripts for top task analysis, which used to get cut from the budget as nice to have.
Fixing the terrible copy bottleneck
Bad copy does more damage to a WordPress site than almost anything else on the page. I have stopped asking clients to write web copy. I ask for bulleted answers to specific user questions instead, then draft from those with AI against a strict readability target, usually a ninth-grade level along the lines of the Hemingway Editor.
It also takes away the blank page problem for stakeholders and keeps the site close to the UX research patterns that are known to convert. The project moves faster and I spend less time untangling jargon.
If AI in UX work is eating your dev hours, I can take it off your plate. I have been working with WordPress since the 4.x days and I know where these tools help and where they fall over.
About a third faster
My output is up somewhere between 25% and 33%. A report that used to take five days takes a few hours. I am not working less because of it. I am spending the recovered time on strategic thinking instead of reading survey transcripts line by line.
What you get paid for is the questions you ask and the leaps you make between them. A model recognizes patterns; it does not invent. Start with small tasks, keep treating it like that intern, and save your own attention for the problems that need it.
