AI & Infrastructure
Entity Recognition

Entity Recognition

Entity recognition, or named entity recognition (NER), is the NLP task that finds names in text. It labels each one with a type, such as person, company, product or place.
July 3, 2026
Cihan Geyik
Table of Content

Before a system can say anything about your company, it has to notice your company was named. Entity recognition does the noticing. It scans text, marks the spans that are names and gives each a type. It’s the first step in turning a sentence into facts. Brand names fail at it more often than marketers expect.

Recognition comes before everything else. A name the model never tags as an entity can’t be linked, counted or tied to facts later in the pipeline.

How does entity recognition work?

Split text into tokens → Predict a label for each token → Join labeled tokens into spans → Assign a type to each span

Most systems label every token with a tag. A common format is BIO, also called IOB: B marks the first token of an entity, I marks the tokens inside it, and O marks everything outside.

Ansvisor → B-ORG
tracks → O
citations → O
in → O
Google → B-PRODUCT
AI → I-PRODUCT
Mode → I-PRODUCT

The hard part is visible in the last three lines. “Google AI Mode” is one product, and a weaker model might tag “Google” as the company and drop the rest. Getting span edges right matters as much as getting the type right.

From rules to language models

ApproachHow it finds namesWeak spot
Rules and gazetteersMatches lists of known names and patternsMisses any name not on the list
Statistical modelsLearns from capitals, word shape and contextNeeds labeled data for each domain
Transformer modelsReads every word in light of its neighborsStill unsure about brand-new names
Prompted LLMsExtracts entities from plain instructionsOutput can vary between runs

The field’s best-known benchmark came from the CoNLL-2003 shared task. Sixteen systems tagged four types in English Reuters news and in German text: persons, organizations, locations and miscellaneous names.

How accurate is entity recognition?

Accuracy is reported as F1, the balance of precision (tags that are correct) and recall (entities that were found). At CoNLL-2003, the top system scored 88.76 on English and 72.41 on German.

That 16-point gap between languages still matters for brands selling in many markets. A model that tags your brand reliably in English may miss it in another language.

Today’s open models do better on news-style text. The spaCy English transformer pipeline, en_core_web_trf 3.8, reports an entity F-score of 89.93 across 18 types. Its small CPU model scores about 84. Text from your niche, full of product names, will usually score lower than the benchmark.

Why brand names trip up entity recognition

  • Common-word names. Brands named Notion, Square or Slack read like ordinary words.
  • Lowercase and stylized names. Many models lean on capital letters, so a lowercase brand loses a strong clue.
  • Product versus company. A product that contains the company name is often split or typed as an organization.
  • New names. A brand launched after the model was trained has no learned pattern behind it.
  • Other scripts. Names written in, or transliterated into, Japanese or Arabic script can be tagged inconsistently.

Mention tracking runs into the same problem. To count your mentions in AI answers, a tool first has to find your name in each answer. Our guide on tracking brand mentions in ChatGPT walks through that side.

Entity recognition and AI search

Entity recognition is one task inside natural language processing, and its output feeds the steps after it. A tagged name goes to entity linking, which picks the matching knowledge base entry. The types it uses are described on our named entities page.

In search, recognized entities tell an engine what a query or passage is about. That supports semantic search, where results are matched on meaning. For your own company, the record those names point to is your brand entity.

Your own pages are the easiest place to fix it. A full first mention, with the category beside the name, gives every model the context it needs to tag you correctly.

How to test entity recognition on your pages

  1. Paste a product page into an NER tool such as spaCy or Google’s Cloud Natural Language API.
  2. Check that your brand and each product come back as entities, with the right type and full span.
  3. Rewrite missed first mentions: full name, category next to it, no pronoun.
  4. Repeat for each language version of the page.
  5. Retest after edits, since small wording changes can flip a tag.

How Ansvisor helps

Ansvisor’s Site Audit scores any public page on 47 weighted AEO and GEO signals, with entity checks among them. Run it on your main product pages in every language you publish. Then look at the answers themselves. Answer Engine Insights reports, market by market, whether each AI platform names your products correctly.

Named Entity Recognition, NER, Entity Extraction, Entity Detection, Entity Identification

Let AI Search Work for Your Growth

Each prompt you win brings more mentions, more citations and more buyers. Ansvisor keeps finding the next one.

AI Answers Mentions Citations Visibility Score AI Share of Voice AI Traffic Revenue
Win Back Search VisibilityTry Ansvisor 14-day Free →✓ Zero risk. Cancel anytime.

FAQ

Frequently asked questions.

What is Entity Recognition?

Entity recognition is the task of finding the names in a text and labeling each with a type, such as person, organization, location or product. It’s usually the first step in turning text into structured facts. Search engines, AI assistants and brand trackers all depend on it.

Why is Entity Recognition important?

A name that is never tagged as an entity can’t be linked to a knowledge base, counted as a mention or connected to facts. If a model misses your product name or labels it as an ordinary word, everything built on top of that step misses it too.

What types of entities can AI systems recognize?

Most models tag people, organizations, locations and products, and many add events, works, dates, money and percentages. The set depends on the training data. A model trained only on news, for example, may have no product type at all.

How accurate is Entity Recognition?

On standard news benchmarks, good models score around 0.9 on the F1 measure, which balances precision and recall. Accuracy drops on new domains, rare names and some languages. Brand names that look like ordinary words, or that use lowercase styling, are frequent misses.

Which tools help analyze Entity Recognition?

Paste a page into an NER tool, such as the spaCy library or Google’s Cloud Natural Language API, and it lists the entities it found with their types. Check that your brand and each product come back complete and correctly typed.

Help us grow the AI Visibility Glossary

New terms are added regularly.

Help us improve the page or suggest a new term →
About the Author
Cihan Geyik

Cihan Geyik

Co-founder at Ansvisor

Cihan Geyik is the co-founder of Ansvisor, an open-source AI Visibility platform for AI Search. With more than 15 years of experience in digital marketing and growth, he writes about AI visibility, AI search, AEO, GEO, citations, and answer engines. He focuses on helping brands understand and improve their presence across ChatGPT, Gemini, Google AI Overviews, Bing AI Mode and other AI answer engines.

Google Preferred Sources
Get more Ansvisor insights in Google

Add Ansvisor as a preferred source and Google will show you more of our AI search, AEO and GEO guides.