How to Disambiguate Entities

Why it matters

Vague references force an AI system to guess which entity you mean — and guesses lower confidence and invite misattribution. "Studies suggest the framework is effective" gives a model nothing to anchor to; "Aggarwal et al.'s GEO study (KDD 2024) found that adding quotations, statistics or citations raised a source's visibility in generated answers by up to ~40%" names the study, the year, the finding, and the effect size, so the claim can be resolved and re-cited accurately.

Disambiguation is the prose-level counterpart to entity mapping (Phase 4.3) and schema (Phase 5.1): you name entities specifically in the text, then reinforce those names with sameAs links so they resolve to known references.

What good looks like

Every entity is named specifically, with its relationships made explicit:

• People, organisations, concepts, and claims are named specifically, not referred to vaguely.
• Each named entity carries disambiguating attributes — affiliation, year, role, or version.
• Relationships between named entities are stated explicitly.
• Names are reinforced with sameAs and schema so they resolve to known records.
• Specific systems and tools are named (GPTBot, ChatGPT, Claude) rather than "the crawler" or "the AI".

Step by step

1. Replace vague references with named entities Hunt for "the researcher", "a study", "the framework", "the AI", "the crawler". Replace each with the specific name: who, which study, which framework, which system.
2. Attach attributes that pin the entity down Add the details that disambiguate: affiliation and year for a person ("Pranjal Aggarwal, IIT Delhi, KDD 2024"), a version for a tool, a publisher for a framework. Attributes separate your entity from everything sharing its name.
3. State relationships between named entities Make the connections explicit — who published what, which tool obeys which standard, which framework improves which outcome. Relationships turn isolated names into a resolvable graph (Phase 4.3).
4. Reinforce names with sameAs and schema Back the on-page names with sameAs links and Person/Organization schema (Phase 5.1) so a model can tie your text to authoritative external records.
5. Name the specific systems and tools Don't write "the crawler ignores robots.txt". Write "OpenAI's GPTBot crawler obeys robots.txt directives published at /robots.txt". Specific system names are themselves disambiguating signals.

Worked examples

Vague:        "Signal" (could be the app, the protocol,
              the newsletter, or a statistics term).

Disambiguated: "Signal - AI Knowledge: ASO & GEO
              Insights, the monthly newsletter published
              by AI Knowledge Signal at
              aiknowledgesignal.io."

Common mistakes