Medical AI Alliance
Reports

Can AI read Korean healthcare and biotech companies?

A census of 418 listed companies · measured 2026-08-14

Investors and overseas partners increasingly ask an AI assistant before they ask anyone else. That makes one question worth measuring: can AI actually read Korean healthcare and biotech companies?

Published 2026-08-14 · Measured 2026-08-14 · Medical AI Alliance

Summary

We measured the websites of all 418 healthcare, biotech, medical-device and medical-AI companies listed on KOSPI, KOSDAQ and KONEX. No company was contacted and no private data was used.

Across the 371 sites that responded, the mean score was 53.0 out of 100.

GapCompaniesShare
No machine-readable statement of what the company does32888.4%
No sitemap25468.5%
Almost no machine-readable text (under 300 characters)12333.2%
No English pages9826.4%
HTTPS not working9124.5%
AI crawlers blocked from the site8322.4%

Most of these are configuration accidents, not company policy.

What we measured, and how

We drew 418 companies from the Korea Exchange listing and kept those in healthcare, biotech, medical devices and medical AI. Standard industry codes have no category for medical AI, so filtering by code alone drops the very companies that matter here. We read the industry code together with the stated products, then cross-checked against regulatory clearance records.

Each company's main page was requested twice: once as an ordinary browser, once as an AI crawler (GPTBot, ClaudeBot). We compared the two responses, then checked whether robots.txt, sitemap.xml and llms.txt genuinely exist, whether structured data declares the organisation, whether English pages are reachable, and how much text is readable without executing JavaScript.

47 sites did not respond and are excluded. All shares below are against the 371 that did.

88.4% do not tell machines what they do

A person reading a homepage understands the business. A machine does not. Separately from the visible text, a site has to declare the company's identity in a form machines read. That is structured data.

328 of 371 companies (88.4%) carry no such declaration. When an AI has to describe one of these companies, there is no primary source to draw on. News articles and forum posts fill the gap instead.

Some declarations exist but are wrong — a company declared as a person rather than an organisation, or a name with no description attached.

22.4% block AI crawlers outright

83 sites refuse AI crawlers. The same URL opens normally in a browser and is rejected when requested as GPTBot or ClaudeBot.

Many of them block and permit at the same time. Their stated AI policy file says commercial use is allowed; the server says no. The sign on the door says welcome and the door is locked.

A harder variant returns a normal 200 status while the body is a block notice. Crawler-side logs record success, so neither the company nor the AI vendor sees a problem.

This is rarely a decision anyone made. Security appliances and CDNs ship with AI crawlers on the default block list. Within the same corporate group, affiliates on different infrastructure were not blocked at all.

33.2% have almost nothing for a machine to read

Getting in is not enough; there has to be something to read. We measured the text available without executing JavaScript.

The median was 1,637 characters. 123 sites (33.2%) fell below 300, and 143 (38.5%) below 1,000 — a few paragraphs to explain an entire company.

Many of these look full to a human. The content is drawn in by JavaScript: visible to the eye, an empty page to the machine.

By sector and by score

Medical AI companies scored highest. Their technical staff tend to own the website directly, and overseas visibility is tied to the business.

SectorCompaniesMean
Medical AI1669
Pharma & CDMO11255
Healthcare infrastructure3154
Medical devices10651
Biotech & drug discovery10649
Score bandCompanies
80+60
60-79102
40-5972
20-39129
0-198

What this means

None of this is a technical problem. It means nobody owns the question. Websites are designed for human readers, and the people who maintain them work on what appears on screen, not on what a machine reads.

Most of it is cheap to fix. Unblocking crawlers is a configuration line; sitemaps and structured data are a day of work. What costs something over time is having material worth reading.

And one thing a company cannot do from its own site. AI leans on third-party listings where companies appear side by side more than on any single company's homepage — and no company can put its competitors on its own site. A comparable listing can only come from an alliance or a third party.

Method and limits

External observation only. No company was asked for data and no internal configuration was inspected. Everything here is what a visitor to the site can determine.

One page per company. Sub-pages may hold more. We chose the main page because that is where an AI reaches first.

Scores are for comparison, not certification. The weighting is ours. The per-item findings matter more than the total.

A point-in-time measurement. Sites change. These figures are as of the measurement date and some companies may have improved since.

Every check and threshold is stated above. We will share the per-item findings for any company on request. If anything here is wrong, tell us and we will re-measure and correct it.

This report is based entirely on public information and may be quoted freely. Please cite Medical AI Alliance as the source.