Government Websites Are Strong on SEO but Empty on AEO — Structure and Improvements for 8 Korean Metropolitan Governments

What We Measured and How

Following the previous post, we diagnosed the homepages of the 8 special and metropolitan city governments three times each, independently, with the same logic as AEGIS Insight, and used per-axis medians. This article does not publish per-government scores or rankings; the reason is below. Measured October 20–21, 2026.

The Three Axes and the AEGIS Index — Best, Average, Worst

We do not say which institution is which; we only show where the 8 values fall (three-run medians).

The gap between best and worst is about 40 points on the Index. The average looks high, but it hides sites with a weak structural base.

Seen Through the 8 Scenarios

AEGIS classifies the current state into 8 scenarios depending on whether SEO, AEO, and GEO each exceed 60. Here is the result for the 8 sites (three-run medians) and for all 24 individual runs.

At a glance, 7 of 8 at the top stage seems reassuring, but three things must be read together.

• The bar is low and they sit on the edge. The scenario threshold is 60, and four sites have AEO in the 60s. In fact three sites changed scenario between their three runs (S8 ↔ S7 ↔ S3), and 6 of 24 runs fell short of the top stage.
• GEO favors public bodies. High recognition and external listings push GEO to around 90, which lifts the scenario classification. One site had high GEO but SEO and AEO did not support it, landing in S3 MIRAGE — "known to AI but structurally fragile."
• Readiness scores and actual use as evidence are far apart. Structure scores were mostly top stage, but as the next post shows, an official domain of the government itself was used as evidence in fewer than half of AI answers, and fewer than one in ten by source count.

A high score is a starting point for checking, not grounds for relief. A top-stage classification does not guarantee AI citation, and an institution on the edge can change stage with a small configuration change.

GEO came out highest because public bodies have rich external listings such as wiki and news coverage. Reading it as "AI cites them well" would be a mistake. Which sources AI actually used is covered in the next post.

How Much Scores Wobbled — Why We Publish No Rankings

Across three independent runs, one site's SEO score spanned 29 points and its AEO score 36. In one case, the score the model reported and the score recalculated from weighted items differed by up to 23 points. A score is only an approximation of structural readiness, and ranking inside that variation inflates meaningless differences. So we publish group averages and structural patterns only.

What "Empty on AEO" Looks Like

Schema stopping at one organization-level type is the biggest gap. What citizens ask AI are questions like "how to file a move-in report" or "youth rent subsidy conditions," yet no page structured as FAQPage or HowTo was found. It is not that the information is missing; it is not in a form AI can easily lift into an answer. Details on writing these are in our FAQPage and HowTo guide.

What Was Not Measured Is a Result Too

Half the sites had no PageSpeed measurement. That does not mean they are slow; it is a limit of the measurement environment. Filling such items with plausible defaults would make a score look measured, so we record "not measurable" and disclose that fact. This principle is discussed in our post on labeling estimates.

Six Improvements

Limits of This Diagnosis

GEO and AEO scores are search-grounded estimates from a single Gemini model, not actual citation rates. They cover one homepage address and do not reflect the structure of lower pages. The 8 are the whole of their category but we do not generalize statistically.

Next

Finally, we check which sources AI used when answering questions about these governments, across 128 answers — including the share that cited official domains and a case of regional confusion around Gwangju.