School AI Visibility in ChatGPT and Perplexity: KPIs and Monthly Ritual

The message arrived on a Tuesday morning. A communications director at a mid-sized UK university forwarded a screenshot to her senior team: ChatGPT had just recommended a direct competitor by name in response to "best marketing degree in the North West." Her institution — which had run a well-resourced admissions campaign that cycle — was entirely absent. No mention, no link, no context. The competitor she saw had not changed its website significantly in two years.

That moment is becoming a familiar inflection point for UK university communications leaders. AI visibility is no longer a digital team footnote. It is a recruitment channel with measurable impact, and it needs to be governed as one.

This article is about the management layer: how to define AI visibility as a formal KPI, how to run a structured monthly review, and how to present the data to senior leadership in a way that drives resource decisions. For the technical setup behind the numbers, see our GEO monitoring guide. For the 90-day implementation programme, see the 90-day action plan to get cited by ChatGPT and Perplexity.

Why AI visibility is now a communications KPI — not just a digital team concern

AI visibility is a communications problem, not a technical one. The evidence sits in the data: in the UK, only 29% of ChatGPT responses mention a higher education institution when a prospect asks about UK universities and courses — Perplexity reaches 38%, against a European average of just 19% (Source: Skolbot GEO Monitoring study, 500 queries × 6 countries × 3 AI engines, Feb. 2026). That gap between 19% and 38% is not explained by technical infrastructure. It is explained by how institutions present their credentials, outcomes, and authority to AI engines — which is exactly what a communications function owns.

The Office for Students and QAA both require institutions to make key information accessible and accurate. Every piece of structured, verifiable content that satisfies those obligations also strengthens AI visibility. The communications director who frames AI visibility as an extension of existing quality-information duties — rather than a separate technical project — will have a far easier conversation with their Vice-Chancellor.

UCAS data consistently shows that sixth-formers use multiple research channels before choosing where to apply. When an AI engine becomes one of those channels — and it already has — the institution that appears in the answer shapes the consideration set. The institution that does not is excluded before the student has even visited a website.

The practical consequence: AI visibility belongs in the same governance conversation as organic search rankings, media coverage, and league table positioning. It needs an owner, defined metrics, and a review cadence. Most UK communications teams have not yet formalised this. Those that do gain first-mover advantage in a channel that rewards early structure.

The 3 AI visibility KPIs that actually matter

Three metrics give a complete picture of how your institution performs in AI-generated answers. Tracking all three prevents the common mistake of optimising for presence at the expense of quality or traffic impact.

These three KPIs map to three distinct business questions. Citation Rate answers: "Can AI engines find and recognise us?" Attribution Rate answers: "Does AI visibility translate into visits and applications?" Mention Context answers: "How does the AI position us relative to competitors?"

A school with high Citation Rate but low Mention Context is being named, but not recommended — a common situation for post-92 institutions on generic queries dominated by Russell Group universities. The fix is different from a school with low Citation Rate across the board, which has a foundational recognition problem. Separating the three metrics makes the diagnosis precise.

For the underlying framework and how to run the full diagnostic, see is your school visible on ChatGPT? and the Perplexity audit guide.

The monthly 90-minute ritual

Ninety minutes, once a month, is sufficient to maintain meaningful governance of your AI visibility programme. The ritual has three equal blocks. Schedule it for the first week of each month, immediately after your monthly data pull.

Block 1 — 30 minutes: the audit

Submit your fixed query battery to ChatGPT and Perplexity. Your query set should include 20 to 30 queries covering: branded queries (your institution name, your flagship programmes), generic programme queries ("MSc data science London", "nursing degree accredited NMC"), geographic queries ("universities in [your city]"), and differentiation queries (accreditations, TEF rating, employability outcomes).

For each response, record three data points in your tracking spreadsheet: cited (yes/no), attributed (link present yes/no), context (primary / alternative / list / caveat). This takes roughly 90 seconds per query if your spreadsheet is pre-formatted. At 25 queries across two engines, that is just over an hour — but 30 minutes is achievable if you run ChatGPT and Perplexity in parallel across two screens or use the Skolbot AI Check tool to accelerate the data collection phase.

Do not vary the queries month to month. Consistency is what makes trends legible. Add new queries only at the quarterly review, not mid-cycle.

Block 2 — 30 minutes: the analysis

Calculate your three KPIs for the month and compare them with the previous three months. You are looking for four patterns:

Improving trend: Citation Rate is rising month-on-month. Identify which query category is driving the improvement — usually reflects Schema.org deployment or new citable content published in the previous cycle.

Plateau: Citation Rate has stabilised at a level below your target. This signals that the current content approach has reached its ceiling. The next move is usually external mentions — UCAS profile completeness, QAA review outcomes, sector media coverage.

Drop: A month-on-month decline of more than 3 percentage points warrants investigation. Common causes: a competitor has published a data-rich page that is now outranking your content; an AI model update has shifted citation patterns; or a technical issue (robots.txt change, page 404) has made previously cited content inaccessible.

Gap between engines: Strong Perplexity performance with weak ChatGPT performance indicates that your web content is solid but your presence in AI training corpora — rankings, QAA reviews, accreditation directories — needs investment.

Note the single most important finding from this block. It becomes the action item for Block 3.

Block 3 — 30 minutes: action planning

Translate the Block 2 finding into one concrete action for the next 30 days. Keep the scope narrow. Communications directors who try to act on five findings simultaneously make progress on none.

Typical actions by scenario:

• Citation Rate <15% on generic queries: Commission two data-rich programme pages with Schema.org EducationalOccupationalProgram markup — target: live within 3 weeks.
• Attribution Rate <20% on Perplexity: Audit robots.txt for PerplexityBot blocks; verify that crawlable HTML versions of key pages exist and that Schema.org canonical URLs are correctly set.
• Mention Context dominated by list mentions: Identify the queries where a competitor holds primary-recommendation position. Analyse what data they surface (employment outcomes, accreditations, TEF rating) and publish a more precise, sourced version on your own site.
• Strong plateau after Schema.org deployment: Shift focus to external corroboration — a sector media pitch to Times Higher Education or Wonkhe, or a UCAS profile completeness review.

Assign each action to a named owner with a delivery date. Record it in your tracking spreadsheet alongside the month's KPI figures.

Reading the data — when to act, when to monitor

Not every change in the data requires an immediate response. The monthly ritual generates signal and noise in roughly equal measure, and distinguishing between them is a core management skill.

Act immediately when: Citation Rate drops by more than 5 percentage points in a single month; a previously cited page has disappeared from AI responses (often indicates a technical issue); a competitor has moved from absent to primary-recommendation on a high-priority query.

Monitor for a second month when: a single query category shows an unexpected dip; Attribution Rate fluctuates by 2 to 3 percentage points without a clear cause; Perplexity and ChatGPT diverge sharply on a query where they previously agreed. AI engine responses have inherent variability — a one-month anomaly is not a trend.

Accept and deprioritise when: generic queries dominated by Russell Group institutions show Citation Rate below 15%. This is structural, not a content failure. On queries like "best university in the UK," four to six institutions will be cited in 90% of responses regardless of your GEO programme. Invest your 30-minute action block in niche queries where your institution can achieve primary-recommendation position — specialist programmes, geographic specificity, distinctive accreditations.

Escalate to a quarterly strategic review when: three consecutive months of monitoring reveal a structural gap that cannot be resolved within the communications team's remit. This typically means a sustained underperformance on ChatGPT driven by insufficient presence in AI training sources — a matter that requires coordination with the digital, academic, and external relations teams, and potentially board-level resource allocation.

Institutions with structured Schema.org data gain an average of +12 points of AI visibility (Source: Skolbot GEO Monitoring study, 500 queries × 6 countries × 3 AI engines, Feb. 2026). If your institution has not yet deployed structured data, that single action should appear in every action plan until it is complete — it is the highest-leverage lever available at any stage of the programme.

Presenting AI visibility to your Vice-Chancellor or board

Senior leaders respond to data presented in the context of institutional strategy, not channel-specific metrics. The communications director who presents "our Perplexity Citation Rate rose from 22% to 31%" is presenting a channel metric. The one who presents "AI-referred visits to admissions pages rose 18% quarter-on-quarter, and we now appear in 31% of Perplexity responses for our target programme queries — up from 22% — placing us ahead of two direct competitors" is making a strategic argument.

The following dashboard template translates monthly KPI data into a board-ready summary. Update it quarterly, not monthly — senior leaders do not need monthly granularity, but they do need trend data.

Accompany the table with three sentences: one on the headline trend, one on the single most important action taken in the quarter, one on the strategic risk or opportunity for the next quarter. This is the format that boards can act on. The ICO guidelines on data accuracy also apply here — any figures you present about your institution's AI visibility that reference external data sources should be clearly attributed and stored with provenance documentation.

For the full strategic framework underpinning this governance approach, see the GEO pillar guide for schools.

FAQ

How do we establish a baseline if we have never tracked AI visibility before?

Run the 30-minute audit block described above before your first action-planning session. Submit 20 queries — 8 branded, 12 generic — to both ChatGPT and Perplexity, and record results in a simple spreadsheet. Do not wait for a perfect query set or a formal project sign-off. A rough baseline from this week is more valuable than a precise baseline from next quarter. The first month's data is purely diagnostic; trends become meaningful from month three onwards.

Should we track Google AI Overviews alongside ChatGPT and Perplexity?

Yes, but treat it as a separate metric. Google AI Overviews (formerly Search Generative Experience) draws primarily from pages already ranking in organic search, making it more responsive to traditional SEO signals than to GEO-specific interventions. Include 5 to 10 Google AI Overviews queries in your monthly audit, particularly for high-intent programme queries, but do not conflate AI Overviews performance with ChatGPT or Perplexity performance — the optimisation levers are different.

Our TEF rating is Gold. Why are we not appearing more prominently in AI answers?

TEF Gold is a powerful authority signal, but only if it is surfaced in a format AI engines can extract. If your TEF rating appears in a PDF, in a press release buried on a news archive page, or in a sentence without explicit Schema.org markup, the AI cannot reliably associate the rating with your institution. Publish a dedicated "TEF Rating" or "Quality and Rankings" page with your rating stated in plain HTML, linked directly from your homepage, and marked up with structured data referencing the Office for Students register entry. A well-structured TEF page can improve your Mention Context classification within four to six weeks on Perplexity.

How do we handle a situation where AI engines are citing inaccurate information about our institution?

Address it on two fronts simultaneously. First, publish accurate data on a clearly structured page — a "Fast Facts" or "Key Statistics" page is ideal — and add Schema.org EducationalOrganization markup that anchors the correct figures to your institution's canonical entity. Second, report persistent inaccuracies to OpenAI via their feedback mechanism and to Perplexity via their correction request process. Do not wait for the AI to self-correct; the structured data intervention is the lever you control directly. Cross-check your UCAS profile and QAA published data for consistency — mismatches between your website and authoritative third-party sources reduce the AI's confidence in your figures.

What is a realistic timeframe for seeing Citation Rate improvements after starting this programme?

Perplexity is the most reactive engine: new citable content and Schema.org changes can produce measurable Citation Rate improvements within two to four weeks. ChatGPT responds more slowly, typically four to eight weeks, because its corpus is updated in waves rather than continuously. Board-level reporting should therefore set a 90-day window as the minimum meaningful evaluation period, with monthly data points used for operational adjustments rather than strategic conclusions. Institutions that expect results within a fortnight will undervalue genuine progress; those that wait a full year to evaluate will lose the window of competitive advantage that early movers currently hold.