Summary

  • ICANN's 2026 IDN and Universal Acceptance report says 20.6% of 923 websites reachable by its automated survey accepted internationalized email addresses.
  • The report gives 11.1% for a 2025 survey of 1,000 global and 1,000 country-specific sites, but says the 2026 survey was automated and some sites requiring human interaction, such as CAPTCHA, may not have been reachable.
  • The difference between those rates is not, by itself, a like-for-like measurement of progress. Neither the direction nor size of any selection effect is established by the published comparison.

An acceptance percentage always has a gate in front of it: which websites could enter the test? ICANN's report, announced on 22 September, gives a striking number for 2026. Of 923 global and country-specific websites that its automated survey could access, 20.6% accepted internationalized email addresses. On the same page, it recalls an 11.1% rate from a 2025 survey covering 1,000 global and 1,000 country-specific sites.

The numerators invite subtraction. The denominators resist it. ICANN says it automated the survey this year and warns that websites needing human interaction to access them, including CAPTCHA, may not be accessible. The 923 are therefore not simply a smaller edition of the 2,000 sites described for 2025. The public comparison does not establish that the same kinds of sites, still less the same individual sites, were observed in both years.

That is a limit on inference, not an accusation that the 2026 test is wrong. Automation can make testing repeatable and scalable; it can also define which pages are visible to the tester. The report does not tell a reader whether excluded sites would have performed better or worse on internationalized email. Calling the displayed percentage-point difference an improvement in the whole website population would supply an answer the comparison has not measured.

A website form is one control point

The measured behavior is website acceptance of a test email-address category. It is not an end-to-end finding that an address can be validated, stored, displayed, processed and delivered across every system. ICANN separately evaluates mail-server readiness; those servers have another population and another denominator. The report's counts of internationalized domain names likewise describe deployment, not the outcome of this website-input test. Each is useful evidence if its object is named.

For a site operator, the practical failure remains concrete: a valid address may be rejected at a registration or contact field, blocking a user before any mail is sent. For an institution assessing global adoption, the separate problem is whether the headline rate describes a stable test frame. An operator can fix a field; a measurement owner has to explain which fields were reachable.

Lu Heng's emphasis on running behavior offers a discipline, not a substitute authority: observe what a system actually accepts, and keep the conditions of that observation attached. In this case, the published figures show two observations. They do not yet provide a matched time series.

Sources