What AI says about your college and where it gets it from

A candidate opens an AI assistant and asks whether your institution is worth attending. The answer arrives in seconds, in the tone of a verdict, and they never visit your site. Knowing what AI says about your college has stopped being a curiosity and become management information.

The discomfort usually comes from the fact that nobody approved that answer. It did not pass through the communications team, it did not come from institutional materials and it does not necessarily cite the official site.

The good news is that the answer does not come from nowhere. It is assembled from public sources, and almost all of those sources can be edited, corrected or contested by someone who knows the route.

How does an AI assistant build an answer about a college?

The assistant builds the answer by retrieving passages from public pages and stitching them into running text. OpenAI's help page on ChatGPT search describes the flow: the system typically rewrites your query into one or more targeted queries, sends them to search providers, and may include citations that open the source when clicked.

Caption: what AI says about your college is stitched together from public sources, and that is where the correction has to happen

It does not consult an official register of institutions or any internal record of yours. Everything on screen came from something that was indexed, accessible and easy to extract.

That detail explains most of the strange answers. If the correct information sits in a heavy PDF, in an image or behind a form, it does not make it into the stitching.

OpenAI's own help page warns that results and citations can be incomplete, outdated or incorrect, and recommends checking that the cited source supports the answer. Worth keeping that line for the next meeting where someone treats an assistant's answer as a verdict.

Google describes how its own AI features work without promising a shortcut. The documentation states that there are no additional technical requirements to appear in AI features, no AI text file and no dedicated schema, and that the page has to be indexed and eligible to be shown with a snippet.

Video: the backdrop to what AI says about your college: how artificial intelligence is changing the routine of marketing teams, on the mkt4edu channel (in Portuguese)

In practice, that dismantles the search for a magic button. The same SEO work that puts a course page in organic results is what feeds the generated answer, because the base is the same.

The starting point is still crawling. If institutional pages are not visited regularly by Google's search crawler, the version of the institution in circulation is the old one.

There is also a surface effect that weighs here. A share of queries ends without a click, and the read on searches that never generate a site visit helps size how much of the conversation about your brand happens away from your domain.

When the assistant cannot find enough to work with, it falls back on whatever is available anywhere. That is where the old forum comment, the outdated course portal and the rating nobody at the institution recognizes come in.

Which sources weigh most in what AI says about your college?

No assistant publishes the weight of each source, so any hierarchy here is a market reading, not an official rule. What is observed often is a recurring set of origins: the official site, the public regulator's register, collaborative encyclopedias, reviews, press and course aggregators.

Those sources do not compete cleanly with each other. They reinforce one another: when three different origins say the same thing, that information tends to show up with more confidence in the answer.

Here is how each origin tends to behave and where you can act:

Table: recurring origins behind what AI says about your college and how much each one can be influenced

Reading the right-hand column is what sets priorities. Where control is total, the error is the institution's responsibility and the fix ships in days.

Where control is low, the intervention changes nature. You do not correct an aggregator by decree, but you do reduce its weight by publishing the right information somewhere more trusted and more cited.

Google has opened a mechanism where the user picks which sites they want to see more content from, and how Google preferred sources work helps explain that move toward personalization.

For institutions with campuses and learning centres, reviews carry extra weight. That is why local SEO for colleges and replying to reviews belong in the same plan as SEO for AI.

How do you audit what AI says about your institution today?

The audit is manual, and that is not a flaw in the method. It consists of running a fixed set of questions across the main assistants, recording the literal answer, the date and the sources cited, and classifying every discrepancy. Without a record, there is nothing to compare the following month.

The first step is building the question list. It has to imitate a real candidate, not the institution's internal vocabulary.

Five families of question usually cover the essentials:

• General reputation: “is [institution] any good?”, “is [institution] worth it?”.
• Regulation: “is [institution] accredited?”, “is the [x] programme at [institution] authorized?”.
• Comparison: “[institution] or [competitor], which is better for [programme]?”.
• Price and admission: “how much does [programme] cost at [institution]?”, “how do you get into [institution]?”.
• Format: “does [institution] offer distance learning?”, “where are [institution]'s learning centres?”.

Run the same list on more than one assistant. Answers diverge between systems, and that divergence is what reveals which source each one favoured.

Record it in a simple spreadsheet: question, assistant, date, the answer pasted in full, the sources cited and a classification field. The classification field is what turns reading into a task.

Three classifications are enough to start. Factual error, outdated information and gap, which is when the assistant does not know the answer and invents a vague workaround.

The gap is the most actionable of the three. It points to a subject the institution has never published anything extractable about, and which therefore depends only on your own production.

On the data side, Search Console help already describes how clicks, impressions and position work in AI Mode and AI Overviews, including the detail that a whole AI Overview occupies a single position. The analysis of the AI report in GSC complements the qualitative audit with numbers.

Teams without dedicated specialists can run this audit. What changes with SEO for LLM work on the table is the next step, deciding what to fix first and what to ignore.

What do you do when AI gets an institutional fact wrong?

The correction happens at the source, never in the assistant. Because the answer is assembled from public pages, correcting means updating the origin of the error and making sure the right version is indexed, visible as text and repeated in more than one trusted place.

Start with the official site, because that is where control is total and the deadline is yours. Price, duration, format, admission route and accreditation have to be in text on the programme page, not only in a PDF handbook.

Video: the technical base behind what AI says about your college: six practices for a site that gets crawled and extracted, on the mkt4edu channel (in Portuguese)

Then check the public regulator's register. A discrepancy between what the site says and what the register shows feeds contradictory answers, and resolving it is regulatory work, not marketing.

On collaborative encyclopedias, the platform's rules govern. An edit made by the institution itself usually runs into conflict of interest, so the predictable route is to supply a verifiable public source and flag the error through the community's channels.

On files that promise to control how AI reads your site, calibrate expectations. Google's documentation on AI features states you do not need to create AI text files, and the analysis of what llms.txt solves and what it does not saves a whole sprint with unproven effect.

There is one correction mistake that holds everyone up. Publishing the right information only in a social media post does not solve it, because permanence and extraction are fragile in that format.

When the error is serious and reputational, press relations come in alongside. A note in an indexed outlet tends to weigh more in the stitching of the answer than ten updates on your own site.

None of this has an immediate effect. Once the origin is corrected, the change still depends on a new crawl and new processing, and that interval varies by assistant.

How do you track what AI says about your college month by month?

Tracking is a short monthly routine, with the same questions, the same record and the same classification. What you measure is the variation: how many answers cite the official site, how many factual errors are left, how many gaps turned into answers and in how many comparisons the institution shows up.

Four indicators do the job at the start:

1. Official domain citation rate: how many of the audited answers show the institution's site as a source.
2. Open factual errors: how many remain uncorrected, with the date they were identified.
3. Gaps covered in the period: how many unanswered questions gained extractable content of your own.
4. Presence in comparative answers: in how many comparisons with direct competitors the institution appears.

Fix the day of the month and the question list. An audit that changes its questions every round produces no time series, only impressions.

Keep the whole answer, not a summary of it. When someone on the board asks what has changed since March, the value is in the literal text compared side by side.

SEO work and SEO for AI share almost the entire technical base, so this panel does not need to live separately. It fits in the same organic search report the institution already receives.

Consulting work that follows this process usually crosses three sources in the same meeting: the qualitative audit, Search Console data and enrollment volume per programme.

What does the AI answer change in student recruitment?

The assistant's answer acts before the site, at the stage where the candidate is still building their shortlist of institutions. If AI describes the institution vaguely, out of date or negatively, the candidate may never reach the programme page, and recruitment loses the lead before the first click.

The impact is bigger in higher-ticket decisions. Postgraduate programmes, MBAs and long-duration courses usually involve more prior research, and more research means more exposure to generated answers.

The market this plays out in is large and concentrated. The 16th Higher Education Map from Instituto Semesp reports 10.2 million higher education enrollments in Brazil in 2024, with 79.8% in the private network.

The same survey puts distance learning at 50.7% of higher education enrollments, against 49.3% on campus. With no geographic barrier, a candidate compares institutions from across the country in the same conversation with an assistant.

There is a quiet effect on the other side of the funnel. Student retention also suffers when the public answer contradicts the experience promised, because expectations miscalibrated at the origin charge their price in the first term.

That is why Educational marketing strategies that work today treat recruitment and retention as the same conversation. What AI says about your college feeds both ends at once.

Frequently asked questions about what AI says about your college

How do you turn what AI says about your college into an advantage?

Start with the audit, because without it any decision is guesswork about public perception. One round of questions across two assistants, recorded in a spreadsheet, reveals in a few hours what the institution has been assuming for months.

Then prioritize what is under total control. A complete programme page in text, up-to-date admission data and coherent regulatory information resolve much of the error without depending on third parties.

Only then invest in covering gaps. That is where your own content pays off most, because competitors have not published anything extractable about that question yet.

Treat the result as an input to educational marketing strategy, not an isolated communications report. Public perception of the brand feeds recruitment and retention in the same cycle.

Auditing what AI answers is the beginning; correcting it runs through how the site describes the institution to the machine. The content on JSON-LD errors covers the markup that feeds what search engines and models understand about your page.

If your institution wants to move past guesswork and build this routine, the path runs through treating search and AI as the same job. That is the scope of mkt4edu's SEO for LLM service.