Do AI systems actually use third-party course directories?

There is no public evidence that any AI company has a licensing agreement with a major higher education course aggregator, and it would be wrong to claim otherwise. What is documented is that AI developers combine publicly available web content with search, APIs, structured datasets and commercial data partnerships. OpenAI's agreements with Yelp and Reddit are the clearest examples. Course directories hold exactly the kind of structured, comprehensive data those agreements tend to target. That is a reason to keep listings accurate, not proof that any specific one is shaping AI answers today.

Which third-party listings should we prioritise?

Start with the ones holding the most structured data about the most courses, rather than the ones sending the most referral traffic. A directory with thousands of well-formed course records has value as a dataset even if its click-through numbers look modest. Referral traffic remains a sensible measure of direct marketing performance. It is just no longer the only thing worth measuring.

How often should we audit our third-party course listings?

At minimum, whenever fees are set for a new cycle and whenever a course is redesigned, renamed or withdrawn. Those are the changes that create the largest factual discrepancies. An annual full sweep alongside spot checks after major updates is realistic for most teams. The point is having a named owner and a repeatable process, not the frequency itself.

Our listings show different fees to our website. Does that matter?

Yes, and more than it used to. A prospective student comparing an online master's is weighing cost and flexibility above almost everything else, so the difference between a total course fee and an annual one is material. It is also the kind of discrepancy that propagates. Once an inaccurate figure sits in a structured database, anything reading that database inherits the error.

Isn't this just admin rather than marketing?

It looks like admin and it largely is. That is the unglamorous conclusion. But the work determines how accurately your institution is described in places you do not control, which has always been a marketing concern. What has changed is that those descriptions may now feed the systems prospective students use to shortlist and compare, which raises the cost of letting the data drift.

Someone else is describing you to the machines, and they have a contract

Saul Bass-style poster: a black hand releasing a fan of red cards that spread outward and distort as they travel, evoking a single set of course facts degrading as it passes through third-party listings.

In July, details emerged of an agreement under which Yelp licenses its reviews, ratings, photos and business information to OpenAI, giving ChatGPT another source of real-time information when people ask for local recommendations. Yelp first disclosed the deal in its February 2026 results. The scope became clearer in July, when it was reported to cover around 330 million cumulative reviews and more than eight million business listings.

At first glance that has very little to do with higher education marketing. It raises a question we should be paying much more attention to: who describes your university to the machines?

Most of the conversation about AI visibility has focused on what organisations publish themselves. Make your website easier for AI systems to understand. Improve your content structure. Strengthen the relationships between pages and answer the questions people actually ask. All of that matters, but it assumes our own website is the most important version of us. I have made a version of that argument about individual researchers, where the same problem appears at a much smaller scale.

The Yelp agreement is a useful reminder that AI systems have other sources. OpenAI says its models are developed using information that is publicly available online, information it partners with third parties to access, and information provided by users, trainers and researchers. Its agreement with Reddit provides access to Reddit’s real-time structured content through its Data API.

There is no evidence that OpenAI has an equivalent agreement with the major higher education course aggregators. The more interesting point is that an AI system does not have to accept the version of your institution that appears on your website as the definitive one. And universities have spent decades creating alternative versions of themselves.

We already give our information to other people

Third-party course listings are hardly new. Universities provide course information to directories, aggregators, rankings organisations and other platforms because prospective students use them. They generate awareness, referral traffic and enquiries, particularly when someone is still comparing institutions rather than searching for a specific university.

FindAMasters links through to university websites and lets prospective students register their interest directly on the platform. There is already a perfectly good marketing reason to keep those listings accurate, and at LSTM improving our third-party course information is part of our wider postgraduate recruitment work.

AI adds another reason. The audience for a course listing is no longer only the people who visit that website. It may also form part of the information environment from which machines learn about, discover and compare our courses. That gives fairly routine data maintenance a different strategic weight.

One course, two versions: what the listings actually say

I tested this with something I know well: LSTM’s online MSc Global Health.

Our course page describes a 100% online, part-time master’s built to flex around working professionals. Students can complete it over two to five years, and the 2026 tuition fee is £14,939 for the whole course, for UK and international students alike.

FindAMasters reflects the two-to-five-year duration and directs students to LSTM for current fee information. Its record was last updated on 6 July 2026. Mastersportal, another major postgraduate platform, describes the same MSc Global Health as lasting 24 months and lists tuition at £14,490 per year.

Someone, or something, trying to understand this fairly simple course can therefore encounter two quite different versions. One says two to five years and £14,939 for the whole course. The other says 24 months and £14,490 a year. That is not an obscure discrepancy buried in the module information. Cost and flexibility are the two things most likely to decide whether someone can do an online master’s alongside work at all.

This is not evidence that anyone is doing a bad job. University course information changes constantly. Fees change, courses are redesigned, copy is rewritten, and platforms update at different times. Some services pull information automatically. Others depend on supplied data or manual updates. Keeping every representation of an institution synchronised is genuinely difficult.

Universities have lived with that problem for years. What is changing is the audience for those representations.

The intermediary may be more important than its traffic suggests

Traditionally I would judge a third-party course platform by what it sends back. How many referrals did it generate? How many enquiries? How many applications can we attribute to it? Those remain sensible questions. They may no longer be the whole story.

Imagine a directory that sends relatively little direct traffic to your website while holding thousands of well-structured course records covering most of UK higher education. Its direct audience is modest next to Google. Its value as a dataset is something else entirely.

That is the lesson I take from Yelp. Yelp spent years building structured information about millions of businesses. OpenAI does not need to recreate that dataset from scratch when it can strike a deal with the people who already own it.

Higher education has its own information intermediaries. Course directories, student review platforms, league tables, government datasets and institutional profiles all hold structured versions of who we are and what we offer. We do not know which of them will matter to AI systems, and some never will. But judging their importance purely by how many clicks they send us today is starting to look like the wrong measure.

What to check in your third-party course listings

The work itself is unglamorous. Course title, fee, duration, mode of study, start date. Whether the link still goes where it should. Whether programmes that no longer exist have been taken down. And whether the thing that makes the course distinctive has survived the journey from your website into somebody else’s database.

None of that requires a generative AI strategy. It requires ownership, process and time. For marketing teams it also means widening what we count as our digital estate. Our website is the bit we own. It is not the only place our institution exists. The same logic applies at individual level, which is the argument behind my AI visibility checklist for researchers.

At LSTM we are already prioritising the accuracy of our third-party listings because they form part of the prospective student journey. What the changing AI landscape has done is give that work another dimension. We are not only maintaining listings for the people who find them directly. We may also be maintaining the source material from which future discovery systems understand us.

Your website might not get the final word

The answer is not to control every description of a university online. That would be impossible and undesirable. Students should have independent sources. Reviews should reflect what students actually think. Rankings should apply their own methodologies, and journalists should describe institutions independently.

Factual information is different. Whether a master’s costs £14,939 for the entire course or £14,490 every year is not a difference of opinion.

As more prospective students ask AI systems to find courses, shortlist universities and compare options, those discrepancies matter more. The difficulty is that we cannot see the full information supply chain behind any given AI answer. Licensing deals get announced. How individual sources shape particular answers does not. So we should be careful about claiming that a specific third-party listing is determining what ChatGPT or Gemini says.

The direction of travel is clearer. The open web is no longer the only game in town. AI companies combine publicly available material with search, APIs, structured datasets and information accessed through commercial partnerships. OpenAI explicitly acknowledges third-party partnerships as one of the sources used in developing its models.

For years we worried about whether prospective students could find the right information on our websites. The better question now is how many versions of us a machine would find, and whether we would stand behind all of them.