The Context: Why 2025 Was the Year Everything Changed for Higher Education Search
In January 2025, the institution I was working with had a reasonable organic search presence, healthy traffic, decent rankings, a reliable pipeline of enrolment leads from non-paid channels. By mid-year, that picture had become significantly more complicated.
ChatGPT's weekly active user base doubled from roughly 400 million to 800 million users over the course of 2025. Google's AI Overviews were appearing in approximately 60% of U.S. search queries by late 2025. Third-party benchmark data from Advanced Web Ranking and BrightEdge showed click-through rates on traditional organic results dropping by up to 30% year-over-year once AI summaries entered the picture.
For a higher education institution marketing to international students, a population that heavily relies on search to discover, evaluate, and compare programs, this was not a theoretical future problem. It was a present one.
The core challenge: Prospective international students were increasingly getting answers to their program research questions, "best data analytics programs in Canada," "how long does a master's degree take in Canada," "is Vancouver a good place to study for international students", directly from AI-generated summaries. If the institution was not cited in those summaries, it did not exist in that discovery moment.
The strategic question was not how to recover lost traffic. It was how to redefine success metrics for organic search in a zero-click environment, and how to build the kind of content architecture that AI systems choose to cite, not just rank.
The Strategic Framework: Separating AIO, LLM, and Traditional SEO
One of the most important decisions we made early was to stop treating AIO optimisation, LLM visibility, and traditional SEO as the same channel. They share some foundations, quality content, technical health, authoritative backlinks, but their performance logic and measurement frameworks are fundamentally different.
| Channel | Primary Goal | Key Metric | What AI systems reward |
|---|---|---|---|
| Traditional SEO | Rank in top 10 results | Clicks, organic sessions | Authority, relevance, backlinks |
| AI Overview (AIO) | Be cited in Google's AI summaries | AIO impressions, # of AIO keywords | Structured, answer-ready content |
| LLM Visibility | Be mentioned in ChatGPT, Perplexity, etc. | Brand mention rate in tracked prompts | Third-party credibility, source ecosystem |
This distinction shaped everything: the content we created, the technical work we prioritised, and how we reported results to stakeholders who were used to measuring organic search purely through sessions and leads.
Phase One: Building an AIO-Ready Content Architecture
Restructuring existing content for AI extractability
Google's AI Overview system uses Gemini to synthesise answers from web pages it considers authoritative and structurally readable. Our audit of existing program pages and blog content found several common patterns that were actively reducing AIO extractability:
- Long, promotional paragraphs with no scannable structure
- Missing or thin FAQ sections that didn't address real prospective student questions
- No summary tables on program pages (duration, intake, tuition range, credential type)
- Weak or absent conclusion sections on blog posts
- Keyword-poor introduction paragraphs that buried the topic signal
We restructured content across 16 program pages and 18 new blog posts following a consistent AI-friendly architecture: keyword-rich introduction paragraph, table of contents for longer pages, short and scannable paragraphs (under 80 words), data tables where relevant, FAQ blocks using natural question-phrasing matched to "People Also Ask" patterns, and a clear conclusion that directly restated the main answer.
Non-branded keyword growth: building top-of-funnel AI discoverability
A heavy reliance on branded keyword traffic is a structural vulnerability in the AIO era. Prospective students who already know the institution's name will find it regardless. The high-value opportunity is capturing students who are searching for what they want to study, not where.
We tracked 219 non-branded keywords across the institution's program areas. By November 2025:
- 122 keywords had improved in ranking position
- 95 of those were ranking on page one
- 87 keywords remained stable
- Non-branded keywords grew from 131 to 667, a 409% YoY increase
Comparing YoY non-branded keyword growth against eight regional competitors, the institution showed the highest percentage growth among all tracked peers, outperforming the closest competitor by more than 5× on this metric. Competitors with significantly larger existing footprints grew non-branded keyword sets by 5–69% in the same period. This institution grew by 409%.
The keywords gaining the most significant traction were high-intent, program-adjacent terms: "data analytics courses in Canada," "data analytics masters programs," "master of management programs," "what can you do with a biomedical science degree," and "AI tools for students in Canada." These are exactly the queries prospective international students use when they are evaluating options, not yet searching for a specific institution.
AIO keyword growth: from 519 to 2,952 in 10 months
AIO keywords, terms where the institution's pages appear as cited sources within Google's AI-generated summaries, grew from 519 in January 2025 to 2,952 by November 2025. That is a 465% increase in AI Overview citation coverage over ten months.
Google updated its AIO algorithm in July and August 2025, which caused a temporary decline in AIO keyword counts across all domains, including competitors. The structured content approach enabled a consistent recovery through September to November, ending the year with the strongest AIO visibility the institution had recorded.
Critically, higher AIO visibility did not translate into proportional traffic growth, which was expected. AIO impressions create zero-click satisfaction for the user. The value is in brand presence during the research phase, not in direct traffic attribution. This required a significant reframing of how the marketing leadership understood organic search ROI.
Phase Two: Technical SEO as the Foundation for AI Credibility
AI systems, both Google's Gemini for AIO and large language models like ChatGPT, weight credibility signals when determining which sources to cite. A technically broken website is not a credible source, regardless of how well the content is written.
Site health: 71% to 91%
The institution underwent a significant technical audit and remediation programme over the course of 2025, which ran alongside the content work. Key areas addressed included:
- Domain migration and URL structure migration, with comprehensive redirect mapping and post-migration validation
- Resolution of broken links, soft 404 errors, redirect chains, and canonicalisation inconsistencies
- XML sitemap corrections including paid page indexation issues
- Metadata optimisation across 85+ pages, standardising title and description patterns
- Image optimisation, render-blocking resource identification, and Core Web Vitals improvement work for LCP and CLS
The result was a site health score improvement from 71% to 91%, a 32% improvement, measured consistently across the year. Domain authority improved by 42%, following the disavowal of 81 spam backlinks and the structured content quality improvements.
llms.txt implementation
We implemented an llms.txt file, a structured, plain-text file placed at the root of the website that explicitly signals to large language models how to interpret and index the institution's content. This is an emerging standard, analogous to robots.txt for traditional crawlers, designed to improve how LLMs like ChatGPT and Perplexity understand site structure, content scope, and entity relationships.
Early data suggested this implementation contributed to improved brand recognition within LLM-driven platforms, though isolating its specific impact from broader content improvements is methodologically difficult at this stage.
Schema markup at scale
FAQPage schema was implemented across all new and restructured blog posts. Article schema was added to editorial content. Program pages were structured with relevant schema types to improve entity understanding. The goal was to help AI systems, which use structured data to reduce ambiguity in content interpretation, extract accurate, reliable information rather than synthesising from unstructured prose alone.
Phase Three: Understanding How LLMs Actually Choose Sources
The most significant insight from this engagement came from a prompt analysis exercise conducted across the portfolio of institutions. We identified 125 prompts where the institution (or sister institutions) was missing from ChatGPT's cited sources, and then analysed what types of sources were cited instead.
The findings were counterintuitive for teams accustomed to traditional SEO thinking:
| Source type | Share of citations | Why LLMs prefer them |
|---|---|---|
| Third-party aggregators & rankings | 33% | Structured lists, neutral framing, multiple institutions compared, easy to summarise |
| Third-party blogs | 20% | Advisory, non-branded, written in how-users-ask language, LLMs borrow this reasoning structure |
| Owned institution pages | 20% | Only cited when content behaves like reference documentation, factual, location-specific, low promotional tone |
| PR / UGC (Wikipedia, Reddit) | 13% | Entity validation and existence confirmation, trust anchors, not primary recommendation sources |
| Mixed / unclear | 13% | LLM fallback when authority signals across source types are fragmented |
The core insight: ChatGPT is not ranking institutions. It is ranking source credibility patterns. An institution can rank #1 on Google for a query and still be entirely absent from the ChatGPT answer for that same prompt, because the LLM's source ecosystem for that topic is dominated by third-party aggregators and editorial blogs that don't reference the institution at all.
This finding fundamentally reshaped the LLM optimisation strategy. Improving owned content was necessary but insufficient. The institution also needed to build presence in the third-party ecosystems, aggregator sites, ranking directories, advisory blogs, that LLMs already treat as trusted sources for higher education queries.
ChatGPT brand visibility tracking
We implemented prompt-level brand visibility tracking using a dedicated AI optimisation tool, monitoring 50 tracked prompts relevant to the institution's programs and target student profile. Brand visibility, the percentage of tracked prompts where the institution was mentioned, reached 35% by November 2025, with negative sentiment stable at 3%.
LLM-driven referral traffic across the institutional portfolio grew 255% year-over-year. While LLM referral traffic is inherently limited in volume (AI chatbots drive 95–96% less referral traffic than equivalent Google search rankings), the directional growth confirmed improving brand presence in AI-generated answer ecosystems.
Phase Four: Blog Strategy as the Primary AIO and Top-Funnel Asset
Blogs were identified early as the highest-leverage content type for AIO visibility. Unlike program pages, which are inherently promotional and location-specific, well-structured blog posts can match the advisory, explanatory tone that AI systems favour when answering research-stage queries.
18 new blog posts were published across 2025, targeting career-focused, comparison-style, and AI-adjacent topics that prospective students at the awareness stage actually search for. The top-performing topics included AI tools for international students, cross-disciplinary career outcome questions ("what can you do with a biomedical science degree"), and data literacy comparisons ("data science vs data analytics").
Total blog traffic for the year reached 2,245 clicks from zero in 2024, a channel built entirely from scratch. In the final quarter (September to November), the blog generated 1,531 clicks with a 213% increase versus the preceding three months, as AIO visibility for blog-targeted keywords compounded.
The blog restructuring approach, adding tables of contents, short structured paragraphs, keyword-rich introductions, and FAQ blocks, produced measurable AIO ranking improvements within one week for targeted keywords. The "leadership vs management" blog post, for example, moved to position 1 for its target keyword within days of restructuring.
Traffic Stability in a Zero-Click Search Environment
Overall organic traffic for the year was effectively flat, 279,021 sessions in 2025 versus 279,178 in 2024, essentially unchanged. In the context of the search landscape shift described above, this was a deliberate and meaningful outcome.
By late 2025, approximately 60% of all Google searches ended without a click. When AI Overviews were present, 92% of users did not click through to a website. Maintaining the same traffic volume against this backdrop, while growing AI Overview coverage by 465% and non-branded keyword rankings by 409%, represents significant structural improvement in search presence, even if the session count appears flat.
The analogy I used with the institution's leadership team: traffic staying flat while the search landscape fundamentally changes is not stagnation. It is the equivalent of holding revenue during a recession. The underlying position is stronger; the headwinds are real.
What This Case Study Demonstrates About AI Search Optimisation for Higher Education
Several principles from this engagement apply broadly to any post-secondary institution marketing to international students in a search environment increasingly dominated by AI-generated answers:
1. AIO and LLM optimisation require different strategies, and different KPIs
AIO is a Google-native channel. LLM visibility is an ecosystem challenge. Both require structured, authoritative content, but LLM visibility additionally requires presence in third-party source networks (aggregators, editorial blogs, Wikipedia) that AI systems treat as credibility anchors. Measuring both by traffic alone will consistently understate their value.
2. Non-branded keyword growth is the most durable AIO investment
The most AIO-visible content answered questions prospective students were asking before they knew which institution they wanted to attend. Category-level, program-adjacent, and career-outcome keywords outperformed branded terms in AIO citation rates and top-of-funnel reach.
3. Owned content must behave like reference documentation, not marketing material
LLMs cited owned institution pages in only 20% of the prompts where third-party sources dominated. The pages that were cited had a consistent profile: clear program intent, explicit location signals in headings, summary data tables, low promotional tone, and factual specificity. Pages that led with brand language, aspirational copy, or vague value propositions were not cited.
4. Technical credibility is a prerequisite for AI visibility
Site health, canonical consistency, schema implementation, and Core Web Vitals are not optional extras in an AI-first search environment. AI systems weight source reliability, and a technically inconsistent website signals lower trustworthiness regardless of content quality.
5. Brand visibility in AI is a lagging metric, start measuring it now
Brand visibility in tracked LLM prompts is the leading indicator of where AI-influenced enrolment attribution will eventually show up. Institutions that begin tracking and optimising this metric now will have a compounding advantage as AI-assisted student decision-making becomes the norm rather than the exception.