A citation audit of ChatGPT, Perplexity, Gemini, Copilot, and AI Overviews.

Every quarter we run a structured audit across the major AI answer surfaces. The most recent run covered 200 higher-ed queries spread across five engines, 40 institutions, and a controlled set of intent types: branded, comparative, transactional, informational.

Here is what we saw.

The aggregate picture

The .edu share is up from 17% in our previous quarter's audit. The forum share is down from 41%. The shift is most pronounced in branded queries; comparison queries still lean heavily on forums.

ChatGPT

What it rewards

• Long, structured pages with clear headings and machine-readable JSON-LD.
• Pages that present comparisons in tables with consistent row labels.
• Wikipedia and high-authority third-party sources for fact-checking.
• Recent content; the freshness signal is real.

Source preferences

In our audit, ChatGPT's citation set leaned heaviest on the institution's own .edu (28%), Wikipedia (24%), third-party rankings (14%), and Reddit (12%). The remaining 22% was a long tail of news, accreditor sites, government data, alumni profiles on LinkedIn, and the like. Reddit's share is meaningfully lower than two quarters ago, because the engine has tuned the source weighting.

What we'd do to win citations on ChatGPT

Ship EducationalOccupationalProgram schema across every program page, update your Wikipedia article inside COI guidelines, and run a quarterly content sprint that produces 4–6 fresh long-form pages per faculty. That is the recipe that moves ChatGPT share most reliably.

Perplexity

What it rewards

• Pages with explicit numbered citations or footnotes already on them.
• Recent content — Perplexity's recency weighting is the most aggressive of the five engines.
• Forum content; Perplexity is the most forum-friendly of the major engines.

Source preferences

Perplexity's citation set in our audit was 35% forum, 22% .edu, 18% Wikipedia, 11% rankings, 14% other. The forum share is the highest of any major engine.

What we'd do

Comparison content with explicit numbered citations is the single highest-leverage Perplexity-specific move. Perplexity's UI rewards traceable source-by-source claim mapping; your content should be written to make that easy.

Google AI Overviews

What it rewards

• Pages that already rank well in organic Google search.
• Pages with structured FAQPage schema (very heavily rewarded).
• Pages with rich EducationalOccupationalProgram markup.
• Pages with author or organization E-E-A-T signals.

Source preferences

Google's AI Overview citation set is the most .edu-heavy of any engine — 38% in our audit. It is the strongest reward for traditional SEO investment.

What we'd do

Don't deprioritize SEO. AI Overviews are still the engine where best-in-class traditional SEO has the most direct AEO carry-over. The investment that wins on Google AI Overview is the same investment that won on Google organic — schema, content depth, internal linking — done at a slightly higher bar.

Gemini

What it rewards

• YouTube content. Gemini's video integration is the strongest of any major engine.
• Wikipedia and Wikidata.
• Google Business Profile data — local queries lean heavily on this.

Source preferences

Gemini's citation set leaned heavily on YouTube (18%) and Google Business Profile (13%), which makes its source mix the most distinctive of the five engines we tested.

What we'd do

If you don't have a YouTube channel with at least monthly upload cadence, start one. Campus tours, program explainers, faculty interviews. Gemini cites video in a way the other engines do not yet.

Microsoft Copilot

What it rewards

• Bing's organic index — Copilot leans almost exclusively on Bing's web index, which has materially different coverage than Google.
• LinkedIn content. Copilot's LinkedIn integration is the strongest of any major engine.
• Edge browser activity for personalized answers (less relevant to institutional marketing).

Source preferences

Copilot's citation set leaned 31% on LinkedIn, 24% on .edu, 16% on Wikipedia, 11% on rankings, 18% on other.

What we'd do

Make sure your institution has a complete LinkedIn School page. Most institutions have a complete LinkedIn Company page (for the central administration) but an incomplete or unclaimed LinkedIn School page (which is what student-facing search surfaces).

Cross-engine patterns

A few things were true across all five:

1. Schema-tagged program pages outperformed unmarked pages by 3–5× in citation rate, across all engines.
2. Wikipedia article freshness was a cross-engine signal. Schools with an article last meaningfully edited in the last twelve months were cited in 2–3× more comparison queries than schools with stale articles.
3. Comparison content with named rivals outperformed evasive comparison content by 4×, across all engines.
4. llms.txt presence correlated with a 7–9% citation lift, controlling for other factors, on the schools we tracked through the deployment.

The recipe that moves citation share is the same across engines: schema, freshness, comparisons, authority. The mix of channels is the only thing that differs.

A practical allocation

If you have a fixed AEO budget and you're picking where to invest:

• 60% across-the-board GEO: schema, llms.txt, comparison content, Wikipedia. These investments pay off in all five engines.
• 15% Google AI Overview-specific work: deep FAQPage, internal linking, E-E-A-T signals.
• 10% YouTube cadence: for Gemini.
• 10% LinkedIn School page completion + cadence: for Copilot.
• 5% Perplexity-specific work: comparison pages with explicit numbered citations.

The mix shifts over time as the engines themselves evolve. Re-audit every quarter.

Methodology note

We use a fixed query bank of 200 queries spread across the four intent types. We run each query against each engine with a clean session (cleared cookies, neutral location, fresh account). We record citation count, source URLs, and source category. We do this every quarter. The data isn't open — we use it for our clients — but the patterns above are robust across the last three quarters of audits.

If you'd like a citation audit on your own institution, the live tool is free and runs against ChatGPT, Perplexity, Gemini, and AI Overview citation patterns. Paste any program page; the audit takes 30–60 seconds.