When AI Assists and When AI Decides Are Very Different Questions

The conversation about AI in the enterprise has been dominated for the past two years by questions around how fast teams can adopt it. Most organizations aren’t dragging their feet. AI tools have moved into workflows at a pace that would’ve seemed implausible in 2022. But speed isn’t the only way to understand where enterprise AI actually stands. To uncover that, you need to know what organizations are willing to let AI do once it’s in the room.

New research from Liferay, based on a survey of 500 U.S. content managers and digital publishing professionals, helps answer that question. The findings reveal that, despite widespread adoption, enterprise users don’t trust AI to operate independently. 

What the Numbers Actually Show

The headline figure from the 2026 Liferay Digital Content Management Survey is that 86% of content teams already use AI features in their content tools, with 41% using them extensively. More than half say AI assists in at least a quarter of their content output. By any reasonable measure, AI adoption in content functions is mature.

Whether teams trust their AI tools to operate independently is another story altogether. Only 14% of respondents say they completely trust AI to publish content without human review. Another 26% mostly trust it, while 31% trust it somewhat. The remaining 28% don’t trust AI for autonomous publishing much or at all. In a function where AI assistance is now nearly universal, fewer than one in seven teams are willing to hand final publishing decisions to the model.

This reflects a meaningful distinction between using AI as a productivity tool within a human-governed workflow and granting AI the authority to make final decisions. Those are genuinely different things, and most organizations are drawing a clear line between them.

Why Enterprise Users Don’t Trust AI

The survey asked respondents directly about what holds them back from trusting AI more fully. Concerns about accuracy and quality topped the list at 34%, followed closely by the desire for human review and oversight at 30%, and security or privacy concerns also at 30%. 

All of these anxieties are rational. AI-generated content can be confidently wrong. Hallucinations remain a problem across current-generation models, particularly for specific factual claims, regulatory details, or product specifications. A content manager who allows AI to publish autonomously is accepting liability for whatever errors the model introduces. In regulated industries, those errors can come with compliance implications that extend well beyond brand reputation.

The preference for human review also reflects the risk that content teams carry. Publishing is a public-facing act. Errors in published content are externally visible in ways that errors in a draft or internal document are not. The asymmetry between the cost of a publishing error and the efficiency gain from removing review is, for many teams, not yet favorable enough to change behavior.

Adoption Without Integration

One of the more counterintuitive findings in the research concerns tool fragmentation. The survey found that 78% of content managers switch between multiple tools to complete a single content task, with 22% doing so very often. The relationship between AI intensity and switching frequency is  even more striking. Heavy AI users are the most likely to switch tools very often, at 31%, compared to 17% of AI-limited users and 10% of those not yet using AI or still piloting it.

This challenges a common premise about AI’s role in workflow simplification. Adding AI tools to a content workflow can, and apparently often does, add steps rather than remove them. A drafting assistant lives in one system, localization capabilities in another, compliance review in a third, and the publishing CMS as a fourth. Each AI capability may accelerate its particular step while increasing the number of handoffs a team has to manage.

This idea is reinforced by the data about team stress. Cross-team coordination and tight deadlines tied as the top sources of day-to-day stress for content managers at 32% each, with keeping up with new technology close behind at 30%. Tool proliferation was cited by 24% as a primary stressor, and too many manual processes by 25%. The teams adding the most AI to their workflows are also experiencing more friction, which points to integration, not just adoption, as a variable that matters.

The Drafting Room Problem

The research revealed another structural issue that speaks to how content platforms have historically been designed. Only 7% of content managers say their primary content platform is where they draft content. The other 93% begin their work in Microsoft Word (33%), Google Docs (21%), a digital asset or content marketing platform (19%), or a project management tool (18%).

This is significant because every handoff between a drafting environment and a publishing environment creates an opportunity for metadata, formatting, brand standards, governance requirements, and any AI-assisted work done during drafting to get lost or degraded. For teams using AI to assist with SEO optimization, tone consistency, or accessibility requirements during the drafting phase, those improvements may not survive the migration into the CMS intact. 

What Heavy AI Users Reveal

Heavy AI users in the survey are doing qualitatively different work than their peers. Fifty-seven percent of heavy AI users manage multilingual content, compared to 20% of non-AI users, a nearly threefold difference. They are also more willing to trust AI for autonomous publishing, with 31% expressing complete trust compared to just 1% of non-AI or piloting users.

The lack of trust is partly a function of accumulated experience. Teams that have processed significant volumes of AI-assisted content develop calibrated judgment about where the model is reliable and where it requires closer scrutiny. Knowing when to extend trust and when to maintain review is a capability that takes time and volume to build.

The implication for organizations is that the path to higher AI trust runs through deliberate exposure and governance. Teams that develop strong review practices while working with AI are building the organizational knowledge to identify, over time, which categories of publishing decisions are genuinely safe to automate.

Security Leads, Innovation Follows

The survey’s data on technology evaluation priorities rounds out the picture. When content managers assess new business technology, security and trust rank first at 27%, followed by ease of use at 20% and cost at 17%. Integration with existing systems ranks fourth at 15%. Innovation and AI capabilities rank fifth at just 14%.

Content teams are already using AI at scale. What they are resistant to is AI that introduces new security exposure, complicates existing integrations, or adds to the coordination overhead they already find difficult. Speed isn’t the factor that determines which teams are furthest along in AI adoption. The most mature teams are those that built a governance model capable of sustaining trust over time.

The Operational Reality Underneath Adoption Metrics

The picture that emerges from this research is of a function that has adopted AI rapidly and responsibly, in the sense that adoption has not run ahead of the organization’s ability to govern it. That is a reasonable outcome, particularly in content functions where publishing errors are externally visible and potentially consequential.

Tracking adoption rates has value, but the more important question is whether the platforms and workflows supporting that usage are designed to reduce friction, maintain governance, and allow trust to develop at a pace grounded in evidence. Where those conditions exist, trust follows adoption. Where they don’t, adoption metrics can look healthy while the underlying workflow remains fragmented and the lack of trust persists.