ICANN Guidelines on Visual Similarity Assessments

The Internet Corporation for Assigned Names and Numbers (ICANN), as the global body responsible for coordinating the domain name system, plays a central role in ensuring the stability, security, and predictability of internet identifiers. As the number of generic top-level domains (gTLDs) and Internationalized Domain Names (IDNs) has increased, so too has the risk of confusion between visually similar domain strings. In response, ICANN has developed and refined a set of guidelines for conducting visual similarity assessments to mitigate the danger of user deception, brand confusion, and namespace collisions. These assessments form a critical component of gTLD delegation processes and are particularly vital in the evaluation of IDNs, where the range of scripts and characters introduces significant complexity.

Visual similarity assessments under ICANN’s guidelines are designed to prevent the delegation of top-level domains that are likely to be mistaken for existing TLDs or for each other. This principle was formally enshrined during the implementation of the New gTLD Program, launched in 2012. One of the foundational documents in this process is the Applicant Guidebook (AGB), which outlines the criteria and procedures for gTLD evaluation. The AGB specifies that applied-for strings must not be visually confusingly similar to existing TLDs or other applied-for strings. This is intended to protect the usability and integrity of the DNS, ensuring that end users do not inadvertently navigate to incorrect domains due to character resemblance.

To implement these evaluations, ICANN established the String Similarity Panel, an expert group tasked with comparing new gTLD applications against existing strings. The panel uses a combination of algorithmic tools and human judgment to assess whether strings are visually confusing. One of the tools initially used by ICANN was the Sword algorithm, which calculates a similarity score based on typographic resemblance. However, algorithmic analysis alone is insufficient, especially in cases where strings use different scripts or languages but still produce visually similar output due to homoglyphs or transliteration.

Human evaluators play a crucial role in identifying context-dependent similarities that automated systems may miss. They consider not only individual characters but also the entire visual impression of the string as rendered in common fonts and formats. For example, two domain strings may differ in code points but appear nearly identical in standard sans-serif typefaces. A string like “rn” may be easily confused with “m” depending on spacing and font rendering, and Cyrillic characters like “с” and “е” can mimic their Latin counterparts “c” and “e” to an imperceptible degree. Evaluators must also account for how domains are perceived at typical screen resolutions and font sizes, where subtle differences become even harder to detect.

ICANN’s guidelines stress that the threshold for confusion is not legal or linguistic equivalence but visual resemblance in a way that could reasonably mislead an average internet user. This standard, while practical, introduces subjectivity and has sparked controversy in several instances. Notable disputes have emerged where applicants argued that their proposed gTLDs were unfairly rejected based on subjective assessments of similarity. For instance, there were objections when applications such as “.unicom” and “.unicorn” were considered too visually similar, despite clear semantic and branding distinctions. These cases highlight the challenge of balancing linguistic nuance, trademark rights, and technical constraints with user safety and navigational clarity.

Visual similarity assessments become even more complicated in the realm of IDNs. With hundreds of characters and multiple writing systems in play, the range of potential confusable strings grows exponentially. ICANN has worked closely with the Unicode Consortium and language experts to develop script-specific tables that define allowable code points for each TLD, limiting the risk of creating homograph pairs. Furthermore, ICANN’s IDN Implementation Guidelines encourage registry operators to apply strict script policies and avoid mixing characters from multiple scripts in a single label, a known vector for visual spoofing. For example, an IDN in the Cyrillic script should not incorporate Latin characters, even if they visually align, to prevent misleading combinations.

To support global consistency, ICANN also relies on the Root Zone Label Generation Rules (LGRs), which are script-specific rule sets developed by expert panels. These LGRs define which characters and variant forms are permitted in the root zone for a given script, helping enforce uniformity and reduce the chances of delegating visually confusable strings. Each LGR is developed with community input and linguistic validation, ensuring that the rules reflect real-world language use while maintaining technical clarity.

The stakes of visual similarity assessments extend beyond user confusion—they touch on brand protection, cybersecurity, and competition. If a TLD resembling a well-known brand or service is delegated, it may be exploited for phishing attacks or brand dilution. Conversely, overly stringent similarity rules could stifle innovation and limit the diversity of new gTLDs, especially for communities seeking to express identity in their native scripts. As a result, ICANN has faced ongoing pressure to balance the need for safety with the demand for linguistic and market expansion.

In recent years, ICANN has initiated reviews of its visual similarity processes to improve transparency and consistency. These reviews include proposals for updated algorithmic tools, clearer criteria for panel assessments, and enhanced appeals mechanisms for applicants. The organization has also engaged with global stakeholders through policy development processes to refine the treatment of IDNs and visual similarity rules in preparation for future rounds of gTLD applications.

Visual similarity remains a critical dimension of domain name evaluation, rooted in both technical standards and human perception. ICANN’s guidelines attempt to operationalize this complex concept through a blend of algorithmic scoring, expert review, and community input. While challenges remain, especially as the domain namespace continues to expand linguistically and structurally, these guidelines serve as a crucial safeguard against confusion and abuse in a multilingual internet. As digital identities grow ever more vital to communication, commerce, and culture, the precision and fairness of visual similarity assessments will remain a cornerstone of trust in the DNS.

You said:

The Internet Corporation for Assigned Names and Numbers (ICANN), as the global body responsible for coordinating the domain name system, plays a central role in ensuring the stability, security, and predictability of internet identifiers. As the number of generic top-level domains (gTLDs) and Internationalized Domain Names (IDNs) has increased, so too has the risk of…