What 150 Knowledge Workers Told Us About AI and Cognitive Redistribution

There is a tension at the heart of the AI productivity narrative. On one side, the claims are compelling: AI tools make knowledge workers faster, more capable, more productive. On the other side, the practical evidence is often anecdotal — case studies of specific successes, testimonials from early adopters, and surveys that ask people whether they feel AI has made them more productive without checking whether their work has actually changed.

At GrowthBoxx, we wanted to understand what was actually happening rather than what people reported was happening. The result was a survey-based study of 150 knowledge workers, published on SSRN as a working paper on cognitive redistribution in AI-enabled knowledge work. This piece draws on that research to explore what the findings mean specifically for marketing and growth functions — including where the data challenges assumptions that are widely held in the AI marketing community.

The Research Question

The study started with a specific conceptual framework: cognitive redistribution. Rather than asking whether AI makes workers more productive overall, we asked how AI tools are redistributing cognitive load — where in the workflow human cognition is still being applied, and where it has been redirected or freed.

The distinction matters because productivity is ambiguous. If AI tools save two hours of drafting time per week, that is a productivity gain relative to the time input. But if those two hours are redirected to low-value tasks that were previously deprioritised rather than to high-value strategic work, the net benefit is smaller than the headline time saving suggests.

Cognitive redistribution is a more precise frame: it asks not just whether cognitive load has been reduced, but where the freed capacity is actually going and whether that redirection is being managed intentionally.

Key Finding One: Time Saved Is Not Reliably Redirected to Higher-Value Work

The most practically significant finding from the study is that AI-enabled time savings do not automatically flow to higher-value work. In a significant proportion of the responses, participants described using time freed by AI assistance for tasks that were also relatively low-value — email management, administrative tasks, additional meetings — rather than for the strategic or creative work they identified as their highest-value contribution.

This finding challenges a common assumption in AI marketing: that AI tools free up marketers to focus on strategy and creativity, with execution handled by the AI layer. This can be true, but it requires intentional design. Without a deliberate decision about what the freed cognitive capacity is for, it is absorbed by the gravitational pull of the inbox and the calendar.

For growth leaders implementing AI tools in their marketing function, the implication is direct: the value of AI-enabled time savings depends on what the team does with the time. This is a management decision, not a technology decision. If the freed time is not explicitly allocated to the strategic and creative functions that AI cannot handle, the productivity gain from AI adoption may be significantly lower than the time saving suggests.

The practical fix is simple but often overlooked: when AI tools are deployed in a marketing function, the workflow redesign should include explicit allocation of the freed capacity. What specific work should team members be doing with the time the AI is saving them? This should be defined before the tools are deployed, not after.

Key Finding Two: AI Literacy Is the Primary Performance Differentiator

The study found a significant gap in outcomes between participants with higher AI literacy — those who reported more comfort with prompting, system design, and evaluating AI outputs — and those with lower AI literacy, even when using the same tools.

This finding is more nuanced than it first appears. The high-literacy group was not simply using more sophisticated tools or spending more time with AI. They were getting qualitatively different outputs from the same interactions, because they understood how to structure their inputs, how to evaluate the quality of the output, and when to accept an output versus when to iterate.

For marketing and growth functions, this suggests that the ROI of AI tool adoption is gated by the AI literacy of the team using the tools. A marketing team with high AI literacy will extract significantly more value from Claude, from a well-designed prompt architecture, and from an integrated workflow than a team using the same tools with lower literacy.

The investment implication is direct: AI literacy training should precede or accompany AI tool deployment, not follow it. And the definition of AI literacy should be practical rather than conceptual — not "understands what LLMs are" but "can structure an effective prompt for a specific task, evaluate the output against defined criteria, and identify when to escalate to human judgment."

Key Finding Three: Self-Reported Quality Improvements Do Not Reliably Match Objective Quality

Participants in the study broadly reported that AI assistance improved the quality of their work. When the study attempted to cross-reference self-reported quality with objective indicators — word count, time spent on revision, peer assessment of output quality — the correlation was weaker than expected.

In some cases, AI-assisted outputs that participants rated as higher quality were rated lower by peer evaluators. In others, participants reported minimal quality improvement while peer evaluators saw significant improvement. The self-assessment of AI-assisted quality appears to be influenced by factors beyond the actual quality of the output, including the novelty of the tool, the reduced effort required, and the plausible-sounding nature of AI-generated text.

For marketing leaders, this finding has a direct implication for quality management. Relying on team members' self-assessment of AI output quality is insufficient. The quality gate in an AI-integrated workflow needs to be external to the person who prompted the output — either an automated review prompt, a peer review step, or both.

This is the finding that most directly informs how we design quality gates in the GrowthBoxx AI growth stacks: the quality evaluation is always separate from the output generation, and the criteria are defined in advance rather than applied judgmentally after the fact.

Key Finding Four: The Functions Where AI Adds the Least Relative Value Are Often the Most Automated

Counter-intuitively, the research found that many participants had automated the functions where human contribution is actually most valuable — strategic synthesis, creative direction, nuanced client communication — while retaining manual processes in the functions where AI assistance would add the most efficiency — data processing, standard content formats, reporting.

This inversion appears to be driven by two factors: the perceived ease of using AI for "creative" tasks (because the output looks impressive), and the perceived risk of using AI for operational tasks that touch external systems or stakeholders (because the consequences of errors are more visible).

For marketing and growth functions, the practical implication is an audit question: where in the current workflow is AI being applied versus where could it be applied most usefully? The most productive AI deployments are not necessarily the most visible ones. A background analytics pipeline that saves an analyst three hours per week of data processing may create more value than a headline AI-generated content programme, even if it is less visible as a change.

Implications for Marketing and Growth Teams

Drawing from all four findings, the implications for marketing and growth functions specifically are:

Design for redistribution, not just savings. When AI tools are deployed, define explicitly what the freed cognitive capacity is allocated to. Strategic work, relationship management, creative direction, and the specific human judgment functions that the AI system cannot handle should be the destination for the time AI frees.

Invest in literacy before tooling. The gap between low and high AI literacy users is large enough to make literacy investment the highest-ROI action before deploying tools widely. A short, practical AI literacy programme — focused on prompting, output evaluation, and judgment about when to escalate — pays off across every subsequent tool deployment.

External quality gates. Self-assessment of AI output quality is unreliable. Every AI-integrated workflow that produces external-facing content should include quality evaluation that is separate from the person who prompted the output.

Audit before automating. Before building an AI workflow for a function, audit where in the function human contribution is actually most valuable. Automate the low-value-human-contribution components first, and be cautious about automating the functions where the human contribution is the key differentiator.

What the Research Did Not Answer

Intellectual honesty requires acknowledging what the study does not tell us. It is a cross-sectional survey study of 150 participants, which limits the causal conclusions that can be drawn. The sample, while geographically diverse, is not representative of the full population of knowledge workers. And the study was conducted at a specific moment in time — the AI landscape is changing rapidly, and findings from a study conducted in one period may not translate directly to a later period.

What the study does provide is a more structured empirical perspective on the AI productivity claims that are ubiquitous in the market, and a set of findings that are practically actionable for leaders thinking about how to deploy AI in their organisations.

The full working paper is available on SSRN for those who want to engage with the methodology and findings in detail. The implications described here are our own interpretation of what the data means for marketing and growth functions specifically.

GrowthBoxx's AI Adoption Benchmarking service applies evidence-based frameworks to evaluate AI adoption in your specific growth function and identify the highest-priority improvements. Book a discovery call to understand where your current adoption stands.