AI use in the Digivante testing community
AI tools are now part of daily working life for most professionals in digital and technology roles, and testers are no exception. They research with it, learn with it, write code with it, and increasingly use it to support the QA work itself. What hasn’t kept pace is a straightforward question: who’s checking what it produces?
We put that question to our own testing community, because they’re a useful group to ask. Testing is a profession built on scepticism — on assuming something is broken until it’s proven otherwise. If verification habits are slipping even among people trained to verify things for a living, that tells you something about where the rest of the working world probably stands.
As part of our AI Community Pulse, we surveyed 43 testers on how often they use AI, what they use it for, how much they trust what it gives them, and critically, how often they actually check it before acting on it. The findings below are drawn from that survey. It’s the first wave of a research series we’ll be running at regular intervals, so this is a baseline rather than a trend.
How professional testers are actually using AI
Before getting to the problem, it’s worth establishing how deeply embedded these tools already are. This isn’t a group experimenting cautiously with AI on the side. It’s a group that has adopted it as core working infrastructure.
Daily, and across several tools at once
84% of respondents use AI tools daily, and 98% use them at least weekly. Almost nobody in the sample is an occasional user. Usage is also rarely limited to a single tool: 88% used two or more distinct AI tools in the past 30 days, with the average respondent using 3.3 different tools across the month.
Real, measurable time savings
The median respondent reported saving two hours a week through AI use; the average was 4.7 hours, pulled upward by a sizeable group saving considerably more. Respondents rated AI’s impact on their productivity at 7.8 out of 10, and their confidence in getting good results from it at 7.5 out of 10.
What people primarily use it for tracks closely with their day job: research (selected by 24 of 43), learning and development (24), and testing or QA itself (22) were the three leading use cases, ahead of coding, data analysis, and content creation.
Survey results: where trust and checking come apart
Our survey indicates that the top concern isn’t whether people trust AI too much. Average trust in AI-generated output sits at a measured 6.7 out of 10 — hardly blind faith. The concern is what people do, or don’t do, despite that measured trust.
Three figures from the survey tell that story on their own:
found incorrect or misleading AI output in the past month
always verify AI output before using it
always verify, among the heaviest AI users
Put plainly: almost everyone has caught AI getting something wrong recently. Fewer than four in ten make a habit of checking before they use it. And the group saving the most time by using AI heavily is the group checking it least of all.
How often testers encounter AI mistakes
Asked whether they’d identified incorrect or misleading AI-generated information in the past month, only two respondents out of 43 said no.
- 18% said it happened frequently
- 53% said it happened occasionally
- 23% said it happened once or twice
- Just 5% said it hadn’t happened at all
How often testers actually check
Set against that near-universal awareness of AI error, verification habits are considerably patchier.
- 35% always verify before using AI output
- 35% verify sometimes
- 28% verify often, short of always
- 2% verify rarely
Verification falls as reliance rises
The intuitive assumption is that heavier AI users would check more carefully, more exposure should mean more caution. Our data points the other way, and it does so consistently across three separate cuts of the sample.
By time saved
Respondents saving 8.5 or more hours a week through AI use verified at a rate of just 8%, against 45% among everyone saving less. That’s a 37-point gap between the group getting the most value from AI and everyone else.
By self-rated skill
Self-described expert users verified always just 17% of the time. Self-described basic users verified always 50% of the time. Three times as often.
By training received
Respondents who’d had formal AI training verified always just 14% of the time, compared with 50% among those with no training at all.
Always-verify rate by segment (base: 43, sub-group size shown)
| Segment | n | Always verify |
|---|---|---|
| Saving under 8.5 hrs/week with AI | 31 | 45% |
| Saving 8.5+ hrs/week with AI | 12 | 8% |
| Self-rated basic users | 8 | 50% |
| Self-rated expert users | 6 | 17% |
| No AI training received | 8 | 50% |
| Formal AI training received | 7 | 14% |
These sub-groups are small (some as few as six or seven respondents) and we’re not presenting them as statistically significant on their own. What makes the pattern worth reporting is that it repeats identically across three independent ways of slicing the same data. Time saved, self-rated skill, and training route are all measuring different things, and all three point the same direction.
The likely explanation isn’t that expertise makes people careless. It’s that fluency brings speed, speed is the entire reason people adopt these tools, and verification is friction that sits directly against that speed. The more value someone extracts from AI, the more that friction costs them — and the more likely it is to quietly get dropped.
Bias: widely seen, rarely checked for
A near-identical pattern shows up around fairness and bias.
65%
had personally observed a biased AI output in the past month
23%
often or always consider whether an output could unfairly disadvantage someone
6.5/10
average self-rated confidence in spotting bias
Nearly two-thirds of respondents have seen what they believed was biased output firsthand — 30% of them more than once. But only 23% often or always stop to consider whether an output could unfairly disadvantage someone, and a quarter said they rarely or never do.
Asked who should be primarily responsible for making AI systems fair, the largest group — 35% — pointed outward, to AI providers. Regulators came second at 19%. Only 12% put the responsibility on individual users.
Respondents rated the importance of transparency in AI systems at 8.5 out of 10, the single highest-scoring item anywhere in the survey. People clearly want to understand how these systems reach their conclusions. Far fewer are auditing the conclusions themselves.
Where testers think the risk sits
Asked which areas are most at risk from AI bias, one category stood out well ahead of the rest:
selected by 27 of 43 respondents
13 each
11
AI-generated content was flagged roughly twice as often as any other single category, notable, given content creation, research, and learning are among this same group’s most common uses. They are most concerned about the category they personally touch most often, and it’s also the category where an error travels furthest: generated content gets published, forwarded, and built on before anyone traces it back to its source.
One further figure worth noting: 26% of respondents had avoided using AI at least once in the past month because of ethical, legal, or privacy concerns. The other 74% hadn’t.
What is the AI Community Pulse?
The AI Community Pulse is a recurring survey run by Digivante among members of our global testing community, tracking how AI tools are actually used in professional and testing work — adoption, trust, verification habits, and perceptions of fairness and bias. This report covers the July 2026 wave.
A note on this sample
Respondents are professional and semi-professional software testers, and are substantially more AI-literate than a general working population. Findings shouldn't be generalised beyond that group.
No trend or month-on-month comparison is presented here — that begins once a second comparable wave has run.
A question like "how often do you verify?" tends to flatter the respondent, so the true verification rate is more likely to sit below 35% than above it.
This survey measured professional AI use at work. It did not measure consumer or shopping behaviour — that's the subject of separate Digivante research now under way.
What this means if you run a digital or ecommerce team
It’s tempting to treat AI verification as a temporary cost of early adoption — something teams do heavily now and taper off as the tools mature. This data suggests any tapering that happens is unlikely to track the error rate. It happens because checking is slow and the tools feel reliable, not because the tools have become more reliable.
For any team where AI is now involved in producing customer-facing content, code, product data, or test coverage — which by now is most digital and ecommerce teams — the practical implication is that the check on that output can’t be assumed to be happening inside the workflow by default. Our data suggests the people closest to the tools are, if anything, the least likely to be the ones doing it.
The answer isn’t to slow AI adoption down. Our own testing community is getting real, measurable value from these tools, and nothing here suggests that should change. The answer is to stop treating verification as something individuals do when they remember to, and start treating it as a deliberate, resourced step in the process — carried out by someone other than the person, or system, that produced the output in the first place.
Build independent verification into your AI workflow
Digivante’s testing community provides the independent, human check on what your team, and increasingly your tools, are actually producing, across real devices, real markets, and real customer journeys.
