Entity Types

Every project tracks one entity type. The type decides what the analysis detects in each AI answer: brands, places, people, and so on. Flags, aliases, groups and metrics work the same way for every type.

​

Choosing a type

You pick the type in the first step of the project wizard, under Entity type. The website analysis then drafts competitors and prompts for that type. If you change the type in the wizard, the competitors and prompts it drafted are cleared. Pick the type from your queries, not from your company. A hotel group that asks “What are the best hotels in Lisbon?” tracks Facilities. The same group asking “Which hotel chain has the best loyalty programme?” tracks Brands. Exports name their entity columns after the type. For example, a Countries & Cities project exports locations and location_citations where a Brands project exports brands and brand_citations.

---

​

What each type covers

Each type has rules for what it detects and what it leaves out. Every type detects only names that are written in the answer. When a name is unclear, the analysis leaves it out.

​

Brands

The default type. One entry per brand: the parent company or primary brand name (Apple, not iPhone 15). Detects

• Brands that compete in the project’s industry, or that the answer offers as an option.
• A product, product line, loyalty programme or single store counts as its brand: “Brand X Air 90” is Brand X.
• A hotel of a chain counts as the chain. An independent hotel is a brand of its own.

Does not detect

• A brand named only as a tool, integration, device, payment method or data source used alongside the options: a CRM, a phone, the car the user drives.

​

Brands & Products

Like Brands, but each named product is an entity of its own. “Apple iPhone 15” gives the brand Apple and the product iPhone 15. Detects

• Brands, at level brand. A brand is detected whenever its name is written, also inside a product name.
• Named products, product lines and models, at level product: iPhone 15, Air Max 90, RAV4.

Does not detect

• A loyalty programme or a single store or hotel location as a product. It belongs to its brand.
• A brand or product named only as a tool, integration, device, payment method or data source.
• Product categories (“running shoes”, “SUVs”).
• Features or technologies not sold on their own (“OLED”, “4G”).
• People, places and events.

​

Countries & Cities

Detects

• Regions, at level region: Europe, Southeast Asia, the Nordics. Only when the answer talks about the region itself.
• Sovereign countries, at level country.
• Cities and towns, at level city.

In “When comparing the Nordics, Sweden and Norway top most livability rankings”, the Nordics is detected. In “Rotterdam hosts Europe’s largest port”, Europe only measures Rotterdam, so it is not detected. Does not detect

• Brands, products, companies or organisations, and people.
• Landmarks and facilities. An airport, a hotel, a stadium or a port is a facility.
• Nationalities (“French”, “Germans”), unless the country itself is named.

​

Organisations

Detects companies, institutions, universities, government bodies, agencies and NGOs, on three levels:

• parent: a parent, holding or umbrella body (Alphabet, United Nations).
• organisation: a standalone organisation (Google, WHO).
• unit: a division, department, subsidiary brand or team (Google DeepMind).

Does not detect

• Consumer products. The organisation that makes the product is detected instead.
• People. A CEO or a founder is a person.
• Cities, countries and places.
• Policies and laws.
• Industry terms (“Big Tech”, “the automotive sector”).

​

People

Detects named individuals: executives, founders, researchers, politicians, athletes, artists. Does not detect

• The organisation or team a person belongs to. In “Hinton spent years at Google, while LeCun leads AI research at Meta”, only Hinton and LeCun are detected.
• Products, brands, places, countries and cities.
• A role with no name (“the CEO”, “a senator”).

​

Places

Detects

• Landmarks, monuments, natural sites (parks, mountains, beaches, lakes) and points of interest, at level place.
• The city, country or region a place sits in, at level area.

In “In Paris, the Eiffel Tower and the Louvre draw millions”, Paris is detected as an area, and the Eiffel Tower and the Louvre as place. Does not detect

• Companies, brands and people.
• Facilities run by an organisation. An airport, a hotel, a stadium or a port is a facility.

​

Facilities

Detects

• Named venues and built structures, at level facility: airports, train stations, stadiums and arenas, hotels, hospitals, ports and convention centres.
• The organisation, chain or authority the answer says runs the facility, at level operator. In “Wembley Stadium, operated by the Football Association”, the Football Association is detected.

Does not detect

• An organisation that is only associated with a venue. In “Old Trafford is Manchester United’s home ground”, Manchester United is not detected.
• Products, brands and people.
• Natural places. A mountain, a beach or a park is a place.
• Countries and cities.

​

Events

Detects

• Recurring event series, at level series: festivals, conferences, championships, awards (the Olympic Games, the Cannes Film Festival).
• Specific editions, at level edition (Paris 2024, Super Bowl LVIII).

Does not detect

• The organiser. It is an organisation.
• The venue. A stadium or an arena is a facility.
• Places, cities, people (performers, athletes, speakers) and products.

---

​

Levels

A level is the tier of an entity inside its type, for example country or city. The analysis fills it when the entity first appears. Brands and People have no level. To correct a level, open the entity sheet in Project settings → Entities and change Level. The analysis only fills an empty level, so your choice stays.

---

​

Changing the type

Change the type in Project settings → General, under Analysis → Entity type.

---

​

Analysis instructions

Analysis instructions are extra rules for the detection, written by you. Set them in Project settings → General, under Analysis → Analysis instructions. The limit is 500 characters.

• They apply to every new answer. Past answers keep their results.
• They take priority over the default rules of the type.
• They cannot change the output. The type’s levels stay the same, and the sentiment, stance and ranking still come from each answer.
• They cannot add a name that the answer does not write.

​

Writing good instructions

• Name what to detect or ignore. Give examples of the names you mean.
• Write one rule per sentence.
• Stay inside the type. Instructions cannot add a level, so they cannot turn a Facilities project into a city tracker. Pick the type that covers what you want.
• Use aliases for spellings. To group the spellings of one brand, add aliases to the entity.
• Check the next run. Read the entities of the next run before you add more rules.

Examples of rules that work:

• Do not count retailers or marketplaces as brands.
• Ignore car models; only detect car makers.
• Exclude dealerships; detect the car brands they sell.
• Only detect cities in Europe.
• Do not detect news websites cited as sources.
• Ne détecte pas les distributeurs, seulement les fabricants.

​

When instructions are refused

MentionLab checks new instructions when you save them. Instructions that could skew the results are refused, and the message tells you why: If the message is “The instructions could not be checked”, nothing was saved. Save again.