From Content to Capability. A Playbook for Educational Content Producers in the AI Era.

Generative AI has made information abundant, while reliable knowledge still depends on evidence, expertise, context, and trust. The more consequential question for education concerns the cognitive work learners still need to perform themselves in order to develop understanding.

Cognitive work splits into two kinds. Mechanical cognition — retrieval, summarization, classification — is what AI now performs fluently, and it’s fast becoming commoditized. Non-mechanical cognition — discovery, discernment, judgment — is what a learner can only build by doing the work, not by watching AI do it. Educational value is moving accordingly: from content to trusted knowledge infrastructure, from information delivery to learning experience design, and from grading outputs to evidencing capability.

The article uses two categories as a strategic lens rather than a settled taxonomy of cognition. Mechanical cognition covers tasks such as information lookup, summarization, classification, and pattern recognition that AI can increasingly perform with fluency. Non-mechanical cognition covers discovery, discernment, and judgment developed through direct engagement with evidence, uncertainty, and consequences.

The boundary between the categories will move as AI develops. The OECD Digital Education Outlook 2026 draws a related distinction between stronger task performance with GenAI and learning gains that depend on pedagogical intent and preserved cognitive effort. Educational value is therefore likely to concentrate increasingly in trusted knowledge infrastructure, learning experiences that preserve meaningful cognitive work, and evidence that a learner can exercise capability independently.

That argument, laid out in more depth in Where Educational Value Moves in the Age of AI, is clear enough in the abstract. The harder question for teams producing educational content, including textbook publishers, curriculum teams, assessment developers, and EdTech product teams, is what it means for the work on their desks this quarter.

The operational implications are more concrete.

1. Content volume is not the primary differentiator

If a chatbot can generate a passable explanation of photosynthesis in seconds, higher content volume or faster drafting is becoming easier for competitors to match. Word counts, chapter counts, and item-bank size still matter when they represent curriculum coverage, progression, accessibility, localisation, or breadth of practice. On their own, however, they are weaker indicators of differentiated educational value.

A more defensible source of value is trustworthy explanation tied to a real curriculum standard, checked against a discipline’s evidence base, accessible to the learners who will use it, and traceable to its authorship and review history. That requires investment in editorial provenance alongside editorial expertise. Metadata can show how content was developed, reviewed, aligned, and updated so that trust is inspectable rather than merely asserted. Emerging work on trusted portable learning context also points toward provenance as part of the infrastructure around AI-enabled learning.

2. Redesign around what the exercise develops

The calculator analogy in the previous article works as a practical design test. For each exercise in a course or program, content teams can ask whether doing the work changes the learner in a way that matters for later understanding, transfer, or judgment. The answer may justify retaining a task even when AI can produce the same immediate output.

This requires a recurring audit rather than a one-time judgment. Three distinctions help:

• Redesign exercises where the learner is only looking up or reformatting information and where the activity has no additional learning purpose. Retain retrieval when recall itself is the learning mechanism. Retrieval practice remains a well-established strategy for strengthening learning, so the distinction should be between information lookup and purposeful recall.
• Preserve and strengthen exercises that require learners to form a position under uncertainty, defend it under challenge, or reconcile conflicting evidence. These tasks can develop discernment and judgment through practice.
• Review tasks that appear open-ended but rely on formulaic pattern matching, such as a comparison that can be completed through a fixed template. As AI handles more of this work, curriculum teams will need to reassess what the task actually develops.

This is a genuine content-development task, not just an assessment-policy one. It touches how practice sets, worked examples, and discussion prompts get written.

3. Build scaffolding into the product

The source article treats scaffolding as a condition that shapes how AI affects learning. Evidence points to variable outcomes. The learning objective, learner population, prior knowledge, access to support, and the way the tool is integrated all matter. For content producers, the practical implication is substantial. AI integration needs explicit pedagogical sequencing and support.

Products and materials can reflect that obligation in several ways

• Structure when a learner is prompted to attempt something independently versus when AI support is offered, rather than leaving that sequencing to the learner.
• Build in visible checkpoints where a learner has to show reasoning before receiving an AI-generated answer or hint.
• Make scaffolding portable so that a teacher, parent, or learner can use it across settings, especially when informal mentoring is limited.

AI features without clear scaffolding or sequencing can amplify existing differences in prior knowledge, access, and learner support. They can also provide useful support where one-to-one help is scarce. Jisc’s 2025 student research documents both uneven access to AI tools and demand for clear institutional support, which makes equity an issue of design and implementation.

4. Treat assessment as an evidence-design problem, not a scoring problem

A single polished output can no longer provide sufficient evidence of independent learning in many contexts. HEPI’s 2026 Student Generative AI Survey found that 94% of UK undergraduates use generative AI to help with assessed work. Assessment development therefore needs stronger evidence of what a learner can understand, explain, apply, and defend.

A wholesale redesign is rarely practical given institutional time and budget constraints. Content teams can sequence the work.

• Identify the small number of assessment points that carry the most weight, and prioritize process-visible design there first (oral defense, drafts with tracked revision, in-context application).
• For lower-stakes practice, AI assistance can be appropriate when the goal is exposure, guided practice, or faster feedback rather than proof of independent capability.
• Build rubrics and tooling that can evaluate reasoning artifacts (question quality, revision history, defense under questioning) alongside or instead of final answers. This is new production work most content teams don’t yet have a workflow for.

5. Reposition “AI-powered” as a claim that needs evidence

The existence of a simulation, tutor, or dashboard does not demonstrate growth in judgment. Efficacy claims need a defined learner population, practice that targets the claimed capability, interpretable evidence, and evaluation over time. Product marketing should follow the same evidentiary discipline as the underlying learning claim.

Any team publishing claims such as “improves critical thinking” or “builds discernment” should be able to state what evidence would support the claim, even before a formal study is complete. TEQSA’s 2026 assurance-of-learning guidance emphasises evaluative judgement, critical thinking, ethical reasoning, and trustworthy evidence of learning. In August 2026, the U.S. Department of Education also asked EdTech providers to specify the learning problem a product addresses, when and for whom it should be used, and what evidence demonstrates improved student learning. Institutional buyers are likely to apply more of this scrutiny to vendor claims.

Guiding principles

The five moves above rest on a smaller set of operating principles.

1. Abundance increases the value of judgment. AI can lower the cost of generating generic material. Editorial judgment, pedagogical design, verification, and accountable decisions therefore carry more of the differentiation.
2. Every exercise should earn its place. Keep, remove, or redesign tasks according to the learning they produce, including tasks such as retrieval practice where the cognitive act itself is valuable.
3. Scaffolding belongs in the product design. AI support should be sequenced around independent attempt, feedback, and the learner’s level of knowledge. Treat scaffolding design as core instructional IP.
4. Evidence claims should survive scrutiny before publication. Capability claims such as “builds critical thinking” or “improves discernment” need a defined learner population, interpretable evidence, and a credible method of evaluation before they appear in a spec sheet or curriculum guide.
5. Design for the institution’s mission. K-12 programs, universities, and corporate upskilling platforms operate under different expectations for rigor, equity, regulation, and evidence. Product strategy should reflect those differences.

Practical steps by producer type

For education publishers:

• Audit existing content and item banks against the “does the learner change” test described above. Flag exercises that mainly proxy learning through information lookup or formulaic output, while protecting tasks where recall, practice, or reasoning itself develops capability.
• Make content provenance metadata a standard production output, including authorship, review history, curriculum alignment, and evidence base. The same requirement can be applied to vendors.
• Shift editorial investment toward materials that require strong judgment to design well, including authentic assessment tasks, discussion prompts that resist formulaic answers, and worked examples that model reasoning as well as procedure.
• Start a small, well-scoped pilot on process-visible assessment (drafts with revision history, oral defense, applied tasks) in one subject or program before attempting a full-catalog redesign.

For EdTech companies:

• Instrument products to capture reasoning artifacts — question quality, revision paths, how a learner responds to a challenge — not just final answers or completion rates; this is the raw material for evidencing capability later.
• Make scaffolding configurable and visible in the product, so educators can see and adjust when AI assistance is offered relative to independent attempt, rather than leaving this as a hidden default.
• Before marketing an “AI-powered” feature as building a capability, define what evidence would be needed to support that claim, and build the instrumentation to eventually collect it.
• Design differently for different institutional contexts (K-12 versus higher ed versus corporate) instead of shipping one AI layer everywhere; the regulatory, equity, and rigor constraints genuinely differ.

For both:

• Treat this as a recurring design review. The boundary between work AI can perform cheaply and work learners still need to practise will continue to move as AI capability advances, so the audit should repeat.
• Avoid two predictable responses. Blanket bans forfeit useful efficiency and access gains. AI features added to unchanged content or pedagogy can also reproduce weak learning design and existing inequities. The stronger response is selective use tied to a defined learning purpose, evidence, and review.

The shift in one line

For content producers, the practical direction is to invest less in undifferentiated explanation and more in verifiable trust, purposeful cognitive effort, built-in scaffolding, and evidence of reasoning. AI can generate content, hints, apparent reasoning traces, and even provenance metadata. Educational value depends on whether those outputs are trustworthy, whether the learner has developed the intended capability, and whether the evidence can support that conclusion.

Publishers and EdTech companies do not need to redesign an entire catalogue or platform at once. A scoped audit of one subject, product, or workflow can identify where explanation has become easier to generate, where learner effort still matters, and where stronger evidence or scaffolding is needed.

---

About the Author

Ashutosh Ghildiyal is Vice President – Growth & Strategy at Integra, where he leads enterprise-wide initiatives spanning growth, strategy, marketing, innovation, and AI-driven solutions for scholarly publishing. With nearly 20 years of international leadership experience, he works at the intersection of scholarly publishing, artificial intelligence, and knowledge infrastructure. His interests include AI governance, research integrity, peer review, publishing innovation, and the future of human knowledge systems. Ashutosh hosts Upstream by Integra, a podcast featuring conversations with global leaders shaping the future of scholarly communication. He serves on the Board of Directors of the International Society of Managing and Technical Editors (ISMTE) and is a Chef at The Scholarly Kitchen, where he writes about scholarly communication, AI governance, research integrity, publishing strategy, and the future of knowledge.