When AI compares training programmes, what can the evidence actually tell you?

Steve New

Imagine someone who has never managed projects in a professional setting. They want eventually to earn the Project Management Professional (PMP) credential and ask an AI assistant to compare online training they could complete around their job.

The assistant could offer a persuasive comparison. Google's Project Management Professional Certificate is designed for beginners, offers flexible study and awards a certificate on completion. The Project Management Institute (PMI) offers self-paced PMP exam preparation that provides the 35 training hours needed for an application. Both descriptions are accurate, but neither tells this buyer whether taking a course will make them eligible for the PMP credential.

The missing distinction is not hidden in an obscure technical document. PMI's own requirements make clear that PMP certification also requires substantial qualifying project-management experience, as well as passing the exam. Completing a preparatory course cannot supply experience the learner does not have.

This is an illustrative example, not an observed AI response or evidence that buyers have been misled. It shows how a comparison can be correct about the courses while leaving the larger decision unresolved. When an AI assistant brings several sources into one fluent answer, the important question is whether it preserves the conditions that limit what the evidence can establish.

A course certificate and a professional credential are not the same thing

Google's programme on Coursera welcomes beginners without requiring a degree or previous experience. Its published information describes a flexible, multi-course certificate and says that it provides more than 100 hours of project-management education that can count towards requirements for certain PMI credentials. That is a meaningful benefit for somebody starting in the field.

PMI's own on-demand PMP exam preparation serves a different immediate purpose: exam-focused study that also supplies the 35 training hours required for an application. Google's certificate can also contribute relevant project-management education towards PMI credentials, so the distinction is not between ‘recognised’ and ‘unrecognised’ training. A buyer might need foundational learning, education hours, professional experience, exam preparation or some combination of them. These are related needs, not interchangeable ones.

A short comparison can give course certificates, accredited programmes, education hours and professional credentials equal visual weight: another tick under ‘recognition’. They are different kinds of evidence. A provider may award a certificate of course completion, an accreditor may recognise an education programme, and a credential-awarding body may impose further requirements on the learner. The buyer needs to know which claim each fact supports before treating it as evidence of suitability.

None of this means that the Google programme is inferior to PMP preparation, or that everyone needs PMP in the first place. For a newcomer, the beginner course could be a more sensible starting point. The mistake would be to let the apparent prestige of a credential settle a decision whose prerequisites have not been checked.

‘Accredited’ establishes something important, but not everything

The distinction becomes particularly clear in coaching education. The Institute of Coaching Studies advertises an online Professional Coaching Skills Certification programme and a route towards the International Coaching Federation's Associate Certified Coach (ACC) credential. ICF's own accreditation directory lists the programme as Level 1 accredited. That is independent evidence of a recognised education programme, rather than merely a provider describing itself as reputable.

ICF's Level 1 pathway covers coaching education, ten hours of mentor coaching and a performance evaluation. This provider includes all three, so the mentoring and evaluation are not separate requirements a graduate needs to arrange after successfully completing the programme. An individual seeking ACC still needs at least 100 qualifying client-coaching hours, including at least 75 paid hours, to pass the ICF ACC exam and have their credential application approved. The programme's accreditation is real, but completing the course does not automatically make the learner an ACC or establish that it suits their circumstances.

There is also a more everyday distinction on the same provider's site. The programme is advertised as ‘100% Online’, but its mix of self-study and live teaching comes with fixed dates that cannot be changed, and participants must attend every class. For somebody who needs to learn wholly on their own timetable, ‘online’ is true but does not establish whether they can take part.

The provider publishes those conditions, and a buyer who reads them may understand them. But a compressed answer that preserves ‘online’ while dropping the compulsory live classes has preserved a true fact and lost part of what the buyer needs to know.

That is why I find these examples more revealing than a hunt for badly written course pages. The important information often exists. The question is whether the comparison retains the qualifications that make it useful.

A useful answer is not always a sufficient answer

AI can make this comparison easier. Instead of reconciling several providers' pages and the separate credential-awarding body's rules unaided, a learner can ask a specific question and receive a joined-up answer. A well-grounded assistant could check the training against the official requirements, ask about the buyer's experience and availability, and distinguish what the course provides from what remains to be done.

In the project-management example, Google's published education hours and PMI's PMP eligibility rules are both relevant. An AI assistant might retrieve each fact accurately and still slip from ‘this offers education hours relevant to some PMI credentials’ to ‘you are eligible for PMP’ without checking the specific training requirement or the applicant's experience. Another assistant could identify those gaps immediately and explain a sensible sequence of steps. The difference is in the inference joining the source facts to a recommendation for a particular person.

Now, course directories and human advisers can also confuse education with credential eligibility. AI did not create that problem. What AI adds is the ability to answer a personal question with a fluent synthesis across providers and credential rules, rather than only a list of facts. That can expose a missing condition or obscure it, depending on what the system checks and what it assumes about the person asking.

Nor is every omitted detail significant. A comparison does not need to reproduce the full handbook, every payment clause or all possible exceptions. It needs to preserve the conditions that could reasonably reverse the recommendation. If the buyer cannot meet compulsory attendance, lacks the experience required for a credential, or needs a form of professional recognition that the course does not provide, that is not background detail. It could decide the purchase.

The important distinction is between a factual description and a justified inference about suitability. The first can be well sourced without the second following from it.

What should a training provider actually do about it?

For a training provider, the useful question is where the interpretation broke down. The course page might be unclear, a summary might lose an important condition, or a buyer might understand the information perfectly and still find the programme unsuitable. Publishing more detail is unlikely to help equally in all three cases.

Take the coaching programme described as ‘100% Online’, which also has compulsory scheduled classes. One learner might find the timetable and sensibly decide not to enrol. Another might miss it because it is difficult to find, or because an AI comparison retained ‘online’ while dropping the attendance condition. Those situations look quite different once you ask what the person actually understood.

In a case worth investigating, I would compare the buyer's actual question, the answer they saw, the original source claims and the decision they believed they were making. I would want to know whether a claim was wrong, a consequential condition went missing or a recommendation went beyond what the facts supported. Where possible, I would also check whether explaining the disputed condition changed the buyer's understanding.

If the provider's own claim is inaccurate, it needs correcting. Conflicting dates or attendance rules need reconciling. If otherwise suitable buyers repeatedly misunderstand an eligibility requirement, a clearer explanation where they compare options or speak to an adviser may help. But when a third-party summary drops a qualification that the source already states plainly, a wholesale rewrite of the website is hard to justify from that alone.

A few AI answers can reveal a mismatch worth examining; they cannot establish how often buyers encounter it or what enrolments have been lost. Nor is every decision not to enrol a loss that better persuasion should reverse. A person who understands the demands of a programme and cannot meet them may be making exactly the right decision. The provider has an interest in helping that person find out before paying.

Sometimes the useful change is small: a clearer sentence about what a certificate confers, a timetable placed where people make the decision, or a human explanation of an individual case. Sometimes no page change is justified. I would rather resolve one consequential misunderstanding than build a longer course page around hypothetical AI answers.

The boundary between a fact and a recommendation

A professional-training decision can involve substantial money and months of work. The practical value of a comparison lies in how well it relates the available evidence to the individual's intended outcome.

Accredited education, a course certificate and eligibility for a professional credential are different claims, just as online access and freedom to study whenever you want are different conditions. Leaving those distinctions out can change what a buyer thinks they are committing to.

AI can help a buyer establish those differences, and sometimes do it better than a conventional course listing. But an accurate answer to ‘What does this programme offer?’ is not necessarily an adequate answer to ‘Will it achieve what I need?’

The important question is not simply whether the facts in a comparison are correct. It is what those facts actually entitle the buyer to conclude.

Selected original sources