AI has made it fast and cheap to produce a first draft of almost anything: a social post, a quote, a product description, a client email. New research suggests that speed has created a different problem. Someone still has to check the work, and in a lot of businesses, nobody is.
A study commissioned by Figma and conducted by YouGov surveyed more than 1,000 Australian working professionals and 200 senior decision-makers at organisations with 20 or more employees that actively use generative AI. Among the decision-makers, 78% reported a rise in output since adopting AI. But 59% said fixing AI-generated output is now one of the most time-consuming parts of their job, and professionals reported spending a third of their working time creating or reviewing AI-generated material.
The part that should worry any business owner is what happens when that review doesn't take place. More than one in four respondents, 27% of working professionals and 28% of senior decision-makers, said their organisation shares AI-generated work with customers and internal stakeholders without checking it first. That's despite roughly nine in ten workers (89%) and executives (92%) agreeing that human judgement is the essential foundation of customer trust.
"Speed without direction and differentiation is just noise," said Alex Train, Country Manager at Figma Australia. "If every enterprise relies on the same generic AI outputs, digital products quickly descend into a sea of sameness. The winners in Australia's digital economy will be the teams that collaborate closely, and preserve the space for human craft and accountability to ship high quality products."
A small business doesn't need an enterprise process to close this gap. Most of what goes wrong with unchecked AI work comes down to a handful of recurring problems, and a short review habit catches most of them before they reach a customer.
A basic framework for checking AI-generated work
Before anything AI-drafted goes out the door, it helps to run it past a short set of questions rather than a quick skim.
Check the facts first. AI tools confidently state things that aren't true, wrong prices, wrong dates, wrong details about a product or policy. Anything with a number, a name or a claim needs a human to verify it against a real source before it's sent.
Read it for voice, not just errors. AI-generated writing tends to have a recognisable rhythm: lots of short punchy contrasts, tidy threes, a dramatic one-line closer at the end of every section. If a customer email or a social post sounds like a press release, it probably needs a plainer rewrite in the business's own voice.
Have someone outside the task read it. The person who generated the draft is the worst person to catch its mistakes, because they already believe the output. A second set of eyes, even a quick five-minute read from a colleague, catches more than another pass by the same person.
Treat anything customer-facing as a higher bar. A draft used internally for a brainstorm carries little risk if it's rough. Anything a customer will actually read or rely on, a quote, a contract summary, advice, should get a proper check every time, not just when there's time.
Keep a record of what gets caught. If the same kind of mistake keeps turning up (wrong figures, awkward phrasing, outdated details) that's a sign the process needs a fix, not just the individual draft.
None of this requires new software or a big process overhaul. It requires treating an AI draft the same way a business would treat a draft from a junior staff member: useful, often good, but not ready to go out the door without someone checking it first.
The research also points to a gap in how confident different levels of a business feel about this. Nearly all leaders surveyed, 94%, said their teams now move faster from concept to final product, compared with 80% of the professionals actually doing the work. Leaders were also more likely to feel excited about AI's impact on their work, 78% compared with 58% of professionals. That gap is worth noticing on its own. The people closest to the work are less convinced things are moving as smoothly as the people managing it think.
Whatever a business's size, the lesson from the data is the same. AI has solved the speed problem. It hasn't solved the judgement problem, and for now, that part is still a job for a person.