Courses That Count: An Evidence-Based Guide to Professional Upskilling

Most working professionals now treat short courses as routine infrastructure for a career: a certificate in cloud security here, a project-management specialisation there, a generative-AI primer because the team is expected to have one. The supply side has responded at scale. Credential Engine's national count identified more than one million distinct credentials on offer in the United States alone across degrees, certificates, certifications, licences, apprenticeships and digital badges (Credential Engine, 2022). Coursera reports 191 million registered learners and 41.8 million enrolments in the year to 30 September 2025 (Coursera, 2025b) — a company-reported figure, not an audited one, but indicative of the volume involved.

Abundance is not the same as quality, and enrolment is not the same as learning. The uncomfortable finding running through three decades of research is that the format of a course — online, in-person, synchronous, self-paced — predicts outcomes far less reliably than what the learner is asked to do, and whether the workplace afterwards supports the new behaviour. Meanwhile, the credentialing infrastructure that would let an employer verify what a certificate actually represents has matured considerably since 2022, and most professionals are not yet using it.

This article is for professionals who spend their own money or their employer's on courses and want a defensible way to choose. It is also useful for managers who approve training budgets, and for L&D and HR teams designing internal programmes. You will find: what the causal evidence says about learning from courses; which study techniques have the strongest empirical support; what employers actually do with credentials; the standards that make a credential portable; a seven-point evaluation framework you can apply before you pay; and the regulatory changes landing in 2026 and 2027 that will affect AI-driven learning platforms.

---

Executive summary

• Format is a weak predictor of outcomes. Rigorous studies find online delivery can help some learners and harm others; the effect depends on prior attainment and course design, not the medium (Bettinger et al., 2017; Cacault et al., 2021).
• Two study techniques stand out. A landmark review rated practice testing and distributed (spaced) practice as high-utility, while rereading, highlighting and summarising were rated low-utility (Dunlosky et al., 2013).
• Course ratings do not measure learning. Across 34 studies, the average correlation between trainee reactions and immediate learning was approximately .08 (Alliger et al., 1997). A five-star rating tells you about satisfaction, not competence.
• Completion is the exception, not the rule. In MIT and Harvard courses on edX, 3.13% of participants completed in 2017–18, down from roughly 6% in 2014–15; around half of registrants never opened the course at all (Reich & Ruipérez-Valiente, 2019).
• Transfer to the job depends heavily on the workplace. A meta-analysis of 89 studies found motivation and a supportive work environment among the significant predictors of whether training changes on-the-job behaviour (Blume et al., 2010).
• Employers say more than they do. Analysis of 11,300 roles found that removing degree requirements raised non-degree hiring by only 3.5 percentage points — fewer than one in 700 hires — with real change concentrated in about 37% of firms (Fuller et al., 2024).
• The verification layer is now standardised. The W3C Verifiable Credentials Data Model 2.0 became a Recommendation on 15 May 2025, and Open Badges 3.0 is built on it, making credentials cryptographically verifiable without contacting the issuer (W3C, 2025; 1EdTech, 2024).
• Europe has a common description standard. The Council Recommendation of 16 June 2022 defines micro-credentials and sets out mandatory descriptive elements, giving buyers a checklist of what a credible provider should disclose (Council of the European Union, 2022).
• Regulation is shifting under AI-driven platforms. The EU AI Act's transparency obligations applied from 2 August 2026, while high-risk obligations covering education and vocational training moved to 2 December 2027 under the Digital Omnibus adopted in June 2026.
• Adult learning participation is not rising. OECD data published on 10 December 2024 show participation in adult learning stagnating or declining across most participating countries, alongside falling literacy in many of them (OECD, 2024a; OECD, 2025).

---

How the professional course market took its present shape

The current landscape is the residue of a failed prediction. When massive open online courses arrived around 2012, the expectation was that free, high-quality instruction from elite institutions would substitute for conventional higher education at global scale. That is not what happened. Reich and Ruipérez-Valiente (2019), analysing every MIT and Harvard course delivered on edX between 2012 and 2018, documented three patterns: participation growth concentrated in wealthier countries, most learners never returning after their first year, and completion rates that declined rather than improved. Among all participants, 3.13% completed in 2017–18, against roughly 6% in 2014–15. The platforms subsequently pivoted towards paid certificates, enterprise contracts and online degrees for professionals — a more conventional business, and the one professionals now buy from.

Two consequences followed. First, the sheer proliferation of credentials: Credential Engine's counts rose from 334,114 unique credentials in 2018 to more than one million by 2022 (Credential Engine, 2022), a market in which quality signals are genuinely hard to read. Second, a policy response. The European Union's Council Recommendation of 16 June 2022 established a common definition of a micro-credential and a list of standard elements every micro-credential should describe — identification of the learner, learning outcomes, workload, assessment method, level, and quality assurance (Council of the European Union, 2022). Whatever its uptake, it is a useful artefact for buyers: it is, in effect, a disclosure checklist.

Meanwhile the underlying demand signal is not what the marketing suggests. The OECD's 2023 Survey of Adult Skills, published on 10 December 2024, covered around 160,000 adults aged 16–65 across 31 countries and economies. Average literacy proficiency improved significantly only in Finland and Denmark; on average, 18% of adults did not reach the most basic proficiency level in any assessed domain (OECD, 2024a). The OECD's accompanying analysis of adult learning found participation stagnating or declining in most participating countries (OECD, 2025). A booming course market and a stagnant skills base can coexist, which is itself a reason for scepticism about enrolment figures as a measure of anything.

---

What the evidence actually says about learning from courses

Delivery format is a weak and conditional predictor

The most-cited causal estimate comes from Bettinger, Fox, Loeb and Taylor (2017), who used an instrumental-variables design at a large for-profit US university where the same courses ran online and in person with identical content and grading rubrics. Taking the online version reduced grades in that course, reduced grades in later courses for which it was a prerequisite, and reduced the likelihood of remaining enrolled. The authors are careful about scope: these are local average treatment effects for students who had both options available. For learners with no in-person alternative, the comparison is not online versus classroom but online versus nothing.

A randomised experiment at the University of Geneva sharpens the picture. Cacault, Hildebrand, Laurent-Lucchetti and Pellizzari (2021) randomly offered first-year students access to live-streamed lectures. Live streaming lowered achievement for lower-ability students and raised it for higher-ability ones. This heterogeneity is the practically important result: remote, self-directed formats reward learners who already have strong foundations and self-regulation, and penalise those who do not. A professional who is confident in a domain may learn efficiently from a self-paced course; the same person entering an unfamiliar field may need structure, deadlines and human contact.

Two techniques with unusually strong support

Dunlosky, Rawson, Marsh, Nathan and Willingham (2013) reviewed ten study techniques against the empirical literature and rated only two as high-utility: practice testing (retrieving information from memory, in low-stakes conditions) and distributed practice (spreading study across time rather than massing it). Elaborative interrogation, self-explanation and interleaved practice were rated moderate. Rereading, highlighting and summarising — the default behaviours of most professionals watching course videos — were rated low-utility.

This gives a blunt but reliable screening question for any course: does it make me retrieve, repeatedly, over time? A course consisting of 40 videos and a final multiple-choice quiz fails it. A course with weekly problem sets, spaced review and graded artefacts passes.

Satisfaction ratings measure something else

Alliger, Tannenbaum, Bennett, Traver and Shotland (1997) meta-analysed 34 studies of training criteria and found the average correlation between trainee reactions of any type and immediate learning was about .08. Purely affective reactions ("I enjoyed it") correlated near zero; "utility" reactions — judgements that the content would be useful on the job — did somewhat better, at around .26, and were stronger correlates of transfer than immediate learning measures were.

The practical implication is specific: when reading course reviews, discount enthusiasm and weight statements about applicability. A review saying "the instructor was engaging" carries little information. A review saying "I used the week-four framework in a client engagement the following month" carries considerably more.

The transfer problem sits outside the course

Blume, Ford, Baldwin and Huang (2010) meta-analysed 89 studies of training transfer and found cognitive ability, conscientiousness, motivation and a supportive work environment all positively related to whether trained behaviour appeared on the job. Notably, work-environment and motivational variables mattered more when training targeted open skills — judgement-based capabilities such as leadership or negotiation — than closed, procedural ones.

This reframes the buying decision. For a procedural skill (a specific tool, a compliance requirement), a good course may be sufficient. For an open skill, the course is at most a third of the intervention; the remainder is opportunity to practise, feedback, and a manager who expects the new behaviour. Buying a leadership course into an environment with none of these is predictably low-yield.

AI tutoring: promising, early, and narrow

Kestin, Miller, Klales, Milbourne and Ponti (2025) ran a randomised controlled trial in Harvard's largest introductory physics course (N = 194) during autumn 2023, comparing a purpose-built AI tutor against an active-learning classroom using identical material. Students learned more, in less time, with the AI tutor. The result is real and was published in a peer-reviewed venue.

It should be read narrowly. The tutor was extensively engineered with expert-authored pedagogical scaffolds and guardrails; the comparison was over two weeks in one subject at one institution; outcomes were immediate post-tests, not durable retention or transfer. It is evidence that well-designed AI tutoring can work, not that AI features on a commercial platform confer the same benefit. Treat vendor claims citing this study as claims about a research prototype until they present their own evidence.

---

Do credentials signal anything to employers?

Less than the marketing implies, and unevenly. Fuller and colleagues at Harvard Business School's Project on Managing the Future of Work, with the Burning Glass Institute, examined 11,300 roles at large US employers before and after degree requirements were removed. On average, hiring of workers without a bachelor's degree rose by 3.5 percentage points — an effect the authors scale to fewer than one in 700 hires. The firms split into three groups: roughly 37% that changed hiring mechanics and saw meaningful increases, about 45% that announced changes without altering practice, and around 18% that made short-lived gains and reverted (Fuller et al., 2024).

Two readings follow, and both are defensible. Pessimistically, credential-based signalling is weak because most employers have not rebuilt screening around skills. Optimistically, a substantial minority of employers have, and the study reports better outcomes at those firms for non-degree hires. What is not supportable is the common claim that degrees have been broadly displaced by certificates.

For an individual professional, the operational conclusion is that a credential rarely functions as an entry ticket on its own. It functions as corroboration attached to demonstrable work. A cloud certification alongside a deployed project is a different proposition from the certification alone.

---

The verification layer: standards worth knowing

A credential is only portable if a third party can check it cheaply. Three developments matter.

W3C Verifiable Credentials Data Model 2.0 became a W3C Recommendation on 15 May 2025, alongside six companion specifications covering data integrity, cryptographic suites and status lists (W3C, 2025). It defines a three-party model — issuer, holder, verifier — in which claims are cryptographically signed and machine-verifiable.

Open Badges 3.0, maintained by 1EdTech, is aligned to that data model and was released for public certification in May 2024 (1EdTech, 2024). The substantive change from version 2.0 is where trust lives: a 3.0 credential is a signed document held by the learner, verifiable from the signature, rather than a file hosted on the issuer's server that disappears when the issuer does.

The EU Council Recommendation on micro-credentials (2022/C 243/02) supplies the semantic layer — what a credential must say about itself to be trustworthy across borders (Council of the European Union, 2022).

Together these mean a well-issued credential in 2026 should be independently verifiable and self-describing. If a provider issues a PDF certificate with a verification page on its own website and nothing else, that is a legitimate reason to discount it — not because the teaching is poor, but because the artefact will not survive the provider.

---

Latest developments

Dated items relevant as of 4 August 2026. Regulatory positions should be confirmed against the Official Journal and primary sources before being relied upon.

• 15 May 2025 — Verifiable Credentials 2.0 published as W3C Recommendation. Seven specifications reached Recommendation status simultaneously, stabilising the technical foundation for portable digital credentials (W3C, 2025).
• 10 December 2024 — OECD publishes 2023 Survey of Adult Skills. Literacy improved significantly in only two of 31 participating countries and economies; numeracy improved in eight (OECD, 2024a).
• 16 December 2025 — Coursera reports 2025 platform figures. 191 million registered learners, 41.8 million enrolments (up 14% year-on-year) and 5.4 million generative-AI enrolments, with data as of 30 September 2025 (Coursera, 2025b). These are company-reported and not independently audited.
• 19 November 2025 to July 2026 — EU AI Act amended by the Digital Omnibus on AI. The European Commission proposed the Omnibus on 19 November 2025; political agreement was reached in May 2026 and the text was adopted by the European Parliament on 16 June 2026 and the Council on 29 June 2026. The effect for the learning sector: obligations for stand-alone high-risk AI systems under Annex III — which includes AI used in education and vocational training — move from 2 August 2026 to 2 December 2027, while AI embedded in regulated products under Annex I moves to 2 August 2028.
• 2 August 2026 — AI Act transparency obligations apply. Article 50 requirements, including disclosure that a user is interacting with an AI system and marking of AI-generated content, took effect as originally scheduled, with a transitional arrangement to 2 December 2026 for certain systems already on the market. For professionals, this is the practical change: learning platforms operating in the EU should now be disclosing where AI is generating content or interacting with learners.

Status note: the deferral of high-risk obligations is established in adopted legislation, but implementation detail, harmonised standards and national enforcement arrangements remain in progress. Claims that the AI Act has been "delayed" are imprecise — parts were deferred and parts were not.

---

A seven-point evaluation framework

The following framework is original to this article. It is designed to be completed in about fifteen minutes before purchase, using only publicly available information.

Point 7 is deliberately about the buyer rather than the course, and it follows directly from the transfer literature (Blume et al., 2010). If a professional cannot answer it, the course is unlikely to change practice regardless of quality.

---

Comparing the main course types

Original comparison; maturity ratings reflect the state of evidence and market practice as assessed for this article, not a formal index.

---

Diagram briefs for the design team

Figure 1 — "The Course-to-Capability Pipeline" Purpose: Show that a course is one stage in a longer chain, and that most value is lost after the course ends. Components (left to right, horizontal flow with arrows): (1) Need identification — box labelled "Skill gap defined against a real task"; (2) Selection — box labelled "Seven-point evaluation"; (3) Learning — box labelled "Retrieval + spaced practice"; (4) Assessment — box labelled "Graded artefact"; (5) Credential — box labelled "Verifiable credential issued"; (6) Transfer — box labelled "Applied at work within 30 days"; (7) Reinforcement — box labelled "Manager feedback and repeated use". Additional elements: A downward "leakage" arrow beneath stages 3, 4 and 6, each labelled with the dominant failure mode — "passive consumption", "unassessed completion", "no application opportunity". A dashed feedback arrow from stage 7 back to stage 1. Visual hierarchy: Stages 1–5 in a neutral tone; stages 6–7 emphasised in the accent colour to signal where value is realised. Caption: "Most course spending fails after the certificate is issued, not before."

Figure 2 — "Credential Verification: Hosted vs Verifiable" Purpose: Explain in one image why credential format affects long-term portability. Layout: Two stacked panels. Upper panel ("Hosted certificate"): Three nodes — Issuer server, Learner (holding a PDF), Verifier. Solid arrow from Verifier to Issuer server labelled "must query issuer"; the Issuer server node marked with a warning glyph labelled "single point of failure". Lower panel ("Verifiable credential"): Same three nodes. Arrow from Issuer to Learner labelled "signed credential"; arrow from Learner to Verifier labelled "presents credential"; a small key/lock icon on the Verifier labelled "checks signature — no issuer contact required". Caption: "Under the W3C Verifiable Credentials Data Model 2.0, trust travels with the credential rather than depending on the issuer remaining online."

---

Worked examples

Example A — Documented: a regulated profession's structured CPD. In regulated fields such as medicine, accountancy and engineering, professional bodies define continuing professional development requirements, maintain member registers and specify assessment. The credential's value derives from the register, not the course provider, and verification is a lookup. This is the maturity level other segments of the course market are slowly moving towards, and it is the reason vendor certifications with proctored examinations and issuer registries generally carry more weight than open course certificates.

Example B — Documented: European policy alignment. The Council Recommendation of 16 June 2022 asks member states and providers to describe micro-credentials using a standard element set and, where possible, to develop public registers (Council of the European Union, 2022). A professional in the EU can therefore reasonably ask a provider which of the standard elements its credential publishes. A provider unable to answer is disclosing something useful.

Example C — Illustrative (hypothetical). A 38-year-old operations manager at a mid-sized logistics firm in South Asia wants to move into data-informed planning. Applying the framework: she rejects a heavily marketed twelve-hour "data science masterclass" on tests 1, 2 and 3; selects a credit-bearing university micro-credential in analytics with weekly graded assignments spread over ten weeks (tests 1–3 pass); confirms it issues an Open Badges 3.0 credential (test 4); and — critically — agrees with her manager in advance that she will rebuild the depot utilisation report using the new methods in the month after completion (test 7). The course cost is the smaller part of the intervention; the agreement with her manager is the larger.

---

Frequently misunderstood ideas

"Low completion rates mean MOOCs failed." Partly. Many registrants never intended to complete, and completion is a poor proxy for value for browsing learners. But Reich and Ruipérez-Valiente (2019) found completion falling even among learners who stated an intention to complete and among those who had paid, which is harder to explain away.

"Online learning is as good as in-person — the research says so." The research says the comparison is conditional. Bettinger et al. (2017) found negative effects; Cacault et al. (2021) found effects in both directions depending on prior attainment. Anyone citing a single global verdict is overstating the literature.

"AI will personalise learning and solve this." One well-designed RCT showed strong short-run gains (Kestin et al., 2025). Generalising from a purpose-built prototype in undergraduate physics to commercial platforms and workplace skills is not warranted by current evidence, and durable retention was not measured.

"Employers now hire on skills, not degrees." Announced far more often than implemented (Fuller et al., 2024).

---

Mistakes to avoid

1. Buying an open-skill course (leadership, influence, negotiation) into a work environment that offers no opportunity to practise it.
2. Treating course ratings as a proxy for learning; weight applicability comments instead (Alliger et al., 1997).
3. Accepting certificates that exist only as PDFs with issuer-hosted verification.
4. Choosing intensive, massed formats for material you need to retain — spacing beats cramming (Dunlosky et al., 2013).
5. Enrolling without a named application task and a date.
6. Assuming a credential recognised in one jurisdiction or sector transfers to another without checking the relevant framework or register.
7. Confusing content volume with rigour: hours of video are a cost, not a benefit.

---

Practical takeaways

For individual professionals

• Define the workplace task before browsing catalogues. Search for the capability, not the credential.
• Apply the seven-point framework; treat failures on tests 2 (assessment) and 7 (transfer plan) as disqualifying.
• Prefer courses with graded artefacts you can show. A portfolio piece outlasts a certificate.
• Schedule spaced review at roughly one week, one month and three months after completion, and use self-testing rather than rereading.
• Check the credential format before paying, and store credentials in a wallet or record that you control.

For managers approving training

• Fund the transfer, not just the course: allocate the practice opportunity and the feedback conversation at the point of approval.
• Ask for the application task in the approval request. It costs nothing and filters weak spend.
• Evaluate at the behavioural level where feasible. Satisfaction surveys measure satisfaction (Alliger et al., 1997).

For L&D and HR teams

• Issue internal credentials to the Open Badges 3.0 / Verifiable Credentials specification so they remain meaningful to employees who leave.
• If operating in or selling into the EU, map any AI-driven learning tools against the AI Act timeline now: transparency obligations already apply, and Annex III high-risk obligations for education and vocational training apply from 2 December 2027.
• Publish the standard descriptive elements for internal programmes — outcomes, workload, assessment, quality assurance — using the EU Recommendation's element set as a template, regardless of jurisdiction.

---

Key insights

1. Format is not the active ingredient; retrieval, spacing and assessment are.
2. Practice testing and distributed practice are the only two techniques rated high-utility in the most comprehensive review of the field.
3. Course satisfaction correlates weakly with learning — around .08 across the meta-analytic evidence.
4. Completion rates in open online courses declined over time even among paying and intending learners.
5. Remote self-paced formats advantage learners who already have strong foundations and disadvantage those who do not.
6. Whether training changes behaviour depends substantially on the work environment, not just the course.
7. Skills-based hiring is announced far more often than practised; credentials corroborate evidence of work rather than replacing it.
8. Verifiable credentials are now a stable standard; certificates that depend on an issuer's website are fragile assets.
9. The EU has both a semantic standard (the 2022 Recommendation) and, from 2027, binding obligations on high-risk AI in education.
10. The single highest-yield decision a buyer makes is naming the task the learning will be applied to, and the date.

---

Frequently asked questions

What is a micro-credential? A certification of the learning outcomes of a short learning experience, assessed against transparent standards. The EU's Council Recommendation of 16 June 2022 provides the most widely referenced definition and specifies the elements a micro-credential should describe, including learning outcomes, workload, assessment and quality assurance.

Are online courses less effective than in-person courses? Not universally. Causal evidence is mixed and conditional: one large study found online delivery reduced grades and persistence, while a randomised experiment found live-streamed lectures helped high-attaining students and hurt lower-attaining ones. Design and learner readiness matter more than medium.

Why are online course completion rates so low? Because enrolment is nearly costless and completion is not. Research on MIT and Harvard courses found roughly half of registrants never started, and completion among all participants was 3.13% in 2017–18. Intent, cost and course design all contribute.

Do employers value certificates from online platforms? Variably. Vendor certifications with proctored examinations carry real weight in relevant technical roles. General course certificates carry much less on their own. Research on skills-based hiring found that most employers announcing skills-first policies did not change hiring practice.

What is the single best way to make a course stick? Repeated retrieval spread over time, applied to a real task at work within about a month of finishing.

Is spaced repetition genuinely evidence-based? Yes. Distributed practice was one of only two techniques rated high-utility in Dunlosky et al.'s 2013 review of ten common study strategies.

Should I trust course reviews and star ratings? Treat them as evidence about experience, not learning. Weight comments describing later application on the job more heavily than comments praising the instructor.

What is Open Badges 3.0 and why does it matter? It is the current version of the digital badge standard, aligned to the W3C Verifiable Credentials Data Model 2.0. It matters because the credential is cryptographically signed and held by the learner, so it can be verified without depending on the issuer's servers.

What are Verifiable Credentials? A W3C standard, at Recommendation status since 15 May 2025, for expressing tamper-evident, machine-verifiable claims exchanged between an issuer, a holder and a verifier.

Are AI tutors better than human instruction? One randomised trial in a Harvard physics course found greater learning in less time with a purpose-built AI tutor than with active-learning classroom instruction. That result is narrow — short duration, one subject, immediate post-tests — and should not be generalised to commercial platforms without their own evidence.

Does the EU AI Act apply to online learning platforms? Yes, in parts. Transparency obligations under Article 50 have applied since 2 August 2026. Obligations for stand-alone high-risk systems under Annex III, which covers education and vocational training, apply from 2 December 2027 following the Digital Omnibus adopted in June 2026. Organisations should confirm current requirements against primary sources.

How much should a professional spend on courses each year? There is no evidence-based universal figure, and any specific number would be invented. A more defensible approach is to budget against identified capability gaps tied to current or next-role responsibilities, and to allocate time for practice at least equal to the course hours.

Is a bootcamp worth it for a career change? It depends on the local labour market and the portfolio produced, both of which vary considerably. Evaluate on assessment rigour, portfolio output and verifiable placement data rather than advertised outcome percentages.

How do I check whether a provider is legitimate? Look for named instructors with traceable records, external quality assurance or accreditation, published assessment methods, disclosed workload, and credentials issued to an open standard. Absence of all five is a meaningful warning.

What should I do with a certificate once I have it? Store it in a format you control, link it to the artefact you produced, and reference the applied outcome rather than the credential name in professional contexts.

Do micro-credentials count towards formal qualifications? Sometimes. Credit-bearing university micro-credentials may stack towards a qualification; most platform certificates do not. Confirm credit value and transferability with the awarding institution before enrolling.

---

Glossary

Assessment rigour — The extent to which a course requires demonstrable performance under controlled conditions rather than participation alone.

CPD (Continuing Professional Development) — Structured ongoing learning required or recognised by a professional body, typically tracked against annual requirements.

Distributed practice — Spreading study across separated sessions rather than massing it; rated high-utility by Dunlosky et al. (2013). Also called spaced practice.

ECTS — European Credit Transfer and Accumulation System; the credit unit used to express workload in European higher education.

EU AI Act (Regulation (EU) 2024/1689) — The European Union's horizontal regulation of artificial intelligence, in force since 1 August 2024, applying obligations in phases according to risk category.

Europass — The European Union's platform and framework for presenting qualifications and skills, including digital credentials.

Interleaving — Mixing different problem types within a study session; rated moderate-utility in the Dunlosky review.

Micro-credential — A certification of assessed learning outcomes from a short learning experience, as defined in Council Recommendation 2022/C 243/02.

MOOC (Massive Open Online Course) — A course designed for unlimited open enrolment, typically delivered online at low or no cost for the audit track.

Open Badges 3.0 — 1EdTech's digital credential specification, aligned to the W3C Verifiable Credentials Data Model 2.0.

PIAAC — The OECD Programme for the International Assessment of Adult Competencies, which produces the Survey of Adult Skills.

Practice testing — Low-stakes retrieval of information from memory; the other high-utility technique in the Dunlosky review.

Transfer of training — The extent to which learning is applied and maintained in the work context, as distinct from performance during training.

Verifiable Credential (VC) — A tamper-evident, cryptographically signed claim conforming to the W3C Verifiable Credentials Data Model

---

References

APA 7th edition. All sources were consulted in preparation of this article.

Academic papers

Alliger, G. M., Tannenbaum, S. I., Bennett, W., Jr., Traver, H., & Shotland, A. (1997). A meta-analysis of the relations among training criteria. Personnel Psychology, 50(2), 341–358. https://doi.org/10.1111/j.1744-6570.1997.tb00911.x

Bettinger, E. P., Fox, L., Loeb, S., & Taylor, E. S. (2017). Virtual classrooms: How online college courses affect student success. American Economic Review, 107(9), 2855–2875. https://doi.org/10.1257/aer.20151193

Blume, B. D., Ford, J. K., Baldwin, T. T., & Huang, J. L. (2010). Transfer of training: A meta-analytic review. Journal of Management, 36(4), 1065–1105. https://doi.org/10.1177/0149206309352880

Cacault, M. P., Hildebrand, C., Laurent-Lucchetti, J., & Pellizzari, M. (2021). Distance learning in higher education: Evidence from a randomized experiment. Journal of the European Economic Association, 19(4), 2322–2372. https://doi.org/10.1093/jeea/jvaa060

Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students' learning with effective learning techniques: Promising directions from cognitive and educational psychology. Psychological Science in the Public Interest, 14(1), 4–58. https://doi.org/10.1177/1529100612453266

Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. (2025). AI tutoring outperforms in-class active learning: An RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15, Article 17458. https://doi.org/10.1038/s41598-025-97652-6

Reich, J., & Ruipérez-Valiente, J. A. (2019). The MOOC pivot. Science, 363(6423), 130–131. https://doi.org/10.1126/science.aav7958

Government and policy sources

Council of the European Union. (2022). Council Recommendation of 16 June 2022 on a European approach to micro-credentials for lifelong learning and employability (2022/C 243/02). Official Journal of the European Union, C 243, 10–25. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32022H0627(02)

European Commission. (n.d.). A European approach to micro-credentials. European Education Area. https://education.ec.europa.eu/education-levels/higher-education/micro-credentials

Standards and technical documentation

1EdTech Consortium. (2024). Open Badges specification, version 3.0. https://www.imsglobal.org/spec/ob/v3p0

World Wide Web Consortium. (2025, May 15). The Verifiable Credentials 2.0 family of specifications is now a W3C Recommendation. https://www.w3.org/news/2025/the-verifiable-credentials-2-0-family-of-specifications-is-now-a-w3c-recommendation

World Wide Web Consortium. (2025). Verifiable Credentials Data Model v2.0 (W3C Recommendation, 15 May 2025). https://www.w3.org/TR/2025/REC-vc-data-model-2.0-20250515/

Industry and institutional reports

Coursera. (2025a). Global Skills Report 2025. https://www.coursera.org/skills-reports/global

Coursera. (2025b, December 16). 2026's fastest-growing skills and top learning trends from 2025. Coursera Blog. https://blog.coursera.org/2026s-fastest-growing-skills-and-top-learning-trends-from-2025

Credential Engine. (2022). Counting U.S. postsecondary and secondary credentials. Credential Engine. https://credentialengine.org/toolkit/counting-u-s-postsecondary-and-secondary-credentials-report/

Fuller, J., Langer, C., & Sigelman, M. (2024). Skills-based hiring: The long road from pronouncements to practice. Harvard Business School Project on Managing the Future of Work & The Burning Glass Institute. https://www.burningglassinstitute.org/research/skills-based-hiring-2024

OECD. (2024a). Do adults have the skills they need to thrive in a changing world? Survey of Adult Skills 2023. OECD Publishing. https://www.oecd.org/en/publications/2024/12/do-adults-have-the-skills-they-need-to-thrive-in-a-changing-world_4396f1f1.html

OECD. (2024b, December 10). Adult skills in literacy and numeracy declining or stagnating in most OECD countries [Press release]. https://www.oecd.org/en/about/news/press-releases/2024/12/adult-skills-in-literacy-and-numeracy-declining-or-stagnating-in-most-oecd-countries.html

OECD. (2025). Trends in adult learning. OECD Publishing. https://www.oecd.org/en/publications/trends-in-adult-learning_ec0624a6-en.html

Books

Kirkpatrick, J. D., & Kirkpatrick, W. K. (2016). Kirkpatrick's four levels of training evaluation. ATD Press.

---

Editorial note on sourcing: statements about the EU AI Act's amended timetable reflect legal analyses of the Digital Omnibus on AI adopted by the European Parliament on 16 June 2026 and the Council on 29 June 2026. Because implementing detail continues to develop, readers with compliance obligations should verify current requirements against the consolidated text in the Official Journal of the European Union. Platform statistics attributed to Coursera are company-reported and have not been independently audited. No case study, quotation, organisation or statistic in this article has been invented; where evidence is preliminary or contested, this is stated in the text.