what3words Explained

what3words is the most prominent commercial competitor to /learn/plus-codes-explained and offers a fundamentally different design trade-off: human-memorable word triples vs computable alphanumeric strings. This article covers what the system is, who uses it, the technical approach, and the documented criticisms that have shaped its evolving role.

This is a balanced support article — significant adoption warrants coverage; documented criticisms warrant note.

What a what3words address is

A what3words address is three English (or other-language) words separated by dots, conventionally prefixed with three forward slashes:

///daring.lion.race

This particular address identifies a 3 m × 3 m square near Buckingham Palace in London.

Key properties:

• Each ~57-trillion-square cell has a unique three-word identifier in each supported language.
• Languages are independent grids: the same physical 3 m × 3 m square has different three-word identifiers in English vs French vs Spanish.
• Words are curated: ~40,000 words per language, chosen to avoid profanity, easily-confused pairs, and rare/technical vocabulary.
• Pronunciation-aware: words are selected for clarity when spoken aloud (no “through” vs “threw” in the same dictionary).
• Domain-specific encoding: word triples that sound similar are placed in widely separated locations to reduce ambiguity if mistyped.

How the algorithm works (sketch)

The full algorithm is proprietary — what3words has not published the encoding/decoding details, and the reference implementation is not open source.

What is publicly known:

1. Earth's surface is divided into a global grid of 3 m × 3 m squares (~57 trillion squares total).
2. Each square is assigned a unique three-word identifier from a per-language dictionary.
3. The assignment is not purely random — similar- sounding triples are placed in distant locations.
4. Each language uses its own dictionary and grid.

The implementation is closed; the only way to encode or decode is via what3words' commercial API or one of their licensed SDKs (iOS, Android, web).

The dictionary

what3words' English dictionary contains ~40,000 words chosen against several criteria:

• No profanity — even when combined.
• No easily-confused pairs — “there” and “their” not both in dictionary.
• No racially or culturally insensitive terms.
• No singular-plural pairs unless distant grids.
• Pronunciation-friendly — no obscure technical vocabulary or jargon.

Similar dictionaries exist for other languages — typically 25,000 to 40,000 words per language. Languages supported include English, Spanish, French, German, Italian, Portuguese, Russian, Chinese, Japanese, Hindi, Arabic, Bengali, and many others (60+ total).

The dictionary is curated by the company; words are periodically added or removed. There's no public specification of the current dictionary content.

Real-world adoption

what3words has substantial enterprise and government adoption, particularly in emergency services and remote addressing:

UK emergency services (2018)

The UK National Police Chiefs' Council (NPCC) formally endorsed what3words in 2018. UK 999 (emergency) call operators accept what3words addresses from callers reporting incidents at remote or unaddressed locations — in moorland, beaches, parks, and so on. Major incidents in the UK in the late 2010s and 2020s have been resolved partly through what3words addresses.

RNLI (Royal National Lifeboat Institution)

The UK's lifeboat service uses what3words for water-based rescue. Beach signage in many UK coastal areas includes what3words addresses to help callers report locations precisely.

Mongolia (2016)

Mongolia's national postal service (Mongol Post) adopted what3words as the country's official addressing system in 2016. Approximately 70% of Mongolians live in areas without formal street addresses; the what3words system allows postal delivery, business registration, and emergency services to reference specific locations.

Mercedes-Benz (2017)

Mercedes-Benz integrated what3words into its in-car navigation system in 2017, allowing drivers to enter destinations as three-word addresses. The integration was scaled back in later years (Mercedes' current models support what3words via voice control rather than as a primary navigation interface).

Other adopters

• AirBnB: shows what3words addresses for some listings to help guests find rural properties.
• Various delivery and logistics companies: particularly for last-mile rural delivery.
• Olympic events: 2018 Pyeongchang and others have used what3words for spectator-area routing.
• Many emergency-response NGOs in disaster zones.

The pattern: what3words excels in verbal-communication contexts (telephone, radio) where word triples are easier to dictate than numeric coordinates or alphanumeric codes.

Critical analysis

Several independent analyses have raised concerns. Two documented prominently:

Andrew Tierney / Pen Test Partners (2019)

In 2019, security researcher Andrew Tierney (of the UK-based Pen Test Partners) published an analysis highlighting cases where similar-sounding what3words addresses point to widely separated locations. Examples included pairs like:

• ///filled.count.soap vs ///filled.count.soaps (one word plural) — possibly miles apart.
• ///fence.gross.bats vs ///fence.gross.batch — near-homophones.

The implication for emergency services: a caller who mishears or mispronounces one word may direct rescuers to the wrong location, potentially with safety consequences.

what3words responded that:

• The dictionary explicitly excludes singular-plural pairs where ambiguity matters.
• Similar-sounding triples are placed in distant locations to avoid local confusion.
• Mispronunciations are within the company's documented design considerations.

Aaron Toponce (2021)

Aaron Toponce published a more detailed 2021 analysis that documented numerous specific ambiguous pairs and argued the cumulative risk was meaningful for safety-critical applications.

The WhatFreeWords incident (2019)

In 2019, an open-source clone called WhatFreeWords emerged, with a reverse-engineered algorithm and dictionary that could encode/decode without the proprietary what3words API. what3words pursued legal action, the WhatFreeWords site was taken down, and mirrors were removed from GitHub under DMCA-style notices.

The episode raised concerns in the open-source community about whether critical addressing infrastructure should depend on a private company with active enforcement of proprietary status. Plus Codes' open-source model is often cited as a contrast.

Proprietary lock-in

Critics argue that emergency services and government addressing systems should not depend on a single private company. Mongolia's adoption is sometimes cited as both a success (addressing the unaddressed) and a concern (national infrastructure dependent on a foreign company).

what3words' position is that the proprietary system enables active quality control of the dictionary and grid assignments — work that would be hard to coordinate in a community open-source model.

Comparison with Plus Codes

The two systems target similar problems with different approaches:

| Property | what3words | Plus Codes | | -------- | ---------- | ---------- | | Type | Three words | 10-11 alphanumeric chars | | Memorability | High | Low–moderate | | Voice-friendly | High | Moderate | | Computable | Via API only | Locally | | Open source | No | Apache 2.0 | | Languages | 60+ (separate grids) | Alphabet-based | | Cell size | 3 m × 3 m | ~14 m × 14 m (default) | | Globally unique | Yes | Yes | | Adopted by emergency services | Yes (UK, others) | Limited | | Adopted as postal | Yes (Mongolia) | Yes (Kolkata, others) | | Proprietary risk | Yes | No | | Documented ambiguity risk | Yes (similar-words) | No (curated alphabet) |

The trade-off is verbal usability (what3words advantage) vs technical openness and computability (Plus Codes advantage).

Worked example

The Empire State Building (40.7484° N, 73.9857° W):

• Lat/lon: 40.7484, -73.9857
• Plus Code: 87G7PXQ4+4W
• what3words: ///pizza.gushy.passed (per the what3words public API as of 2026; the exact triple varies slightly depending on which 3 m square is selected near the building entrance).

Word triples for the same physical location in different languages:

| Language | Three-word address | | -------- | ------------------ | | English | ///pizza.gushy.passed | | French | ///pivot.cabane.signaler | | Spanish | ///lengua.simio.estilo | | German | ///rückwärts.scheune.kontakt | | Japanese | ///さかみち.ひとりごと.せきにん |

Each language has its own grid; the triples are unrelated across languages even though they all identify the same physical location.

When what3words is appropriate

what3words is well-suited for:

• Emergency-services reporting where callers may be panicked or distracted and need to dictate a location verbally — UK 999 and similar services have shown measurable success.
• Postal addressing in regions without formal addresses — Mongolia's adoption is a viable national case.
• Disaster response where formal infrastructure may be damaged.
• Voice-controlled navigation (e.g., car infotainment systems).

what3words is less suited for:

• Critical infrastructure where vendor lock-in is unacceptable.
• Open-data applications where the algorithm and data must be free.
• Verbal precision in low-quality audio where the documented homophone/plural risk has been concerning.
• Long-term archival addressing where the dictionary could change.

Pricing and licensing

Limited free-tier API access is available for non-commercial use. Commercial licensing varies by use case (mapping, logistics, emergency services). The exact pricing is not public; enterprise contracts are negotiated.

This contrasts with Plus Codes, which can be encoded/decoded locally with no licensing or API call required.

Common misconceptions

“what3words is open source.” It's not. The algorithm and dictionary are proprietary. A 2019 clone (WhatFreeWords) was taken down via legal action. For open-source location encoding, see /learn/plus-codes-explained.

“what3words is more accurate than Plus Codes.” Cells are smaller (3 m × 3 m vs 14 m × 14 m for the standard Plus Code). But Plus Codes can use the extended 11-character form for ~3.5 m precision. “Accuracy” in the sense of cell granularity is similar at the extended precision; the question is whether the cell size matters for the application.

“Three-word addresses are unambiguous.” Not entirely. Independent researchers have documented cases where similar-sounding triples point to widely separated locations. The company has responded with curation choices but the risk persists. For critical applications, verify the triple by some other means (cross-check with the lat/lon, confirm with reverse-decoded address).

“what3words is used officially in every country.” Major adopters are UK (emergency services, partial), Mongolia (postal addressing, near-full), and several specific organizations in many countries. It is not a global postal standard the way the Universal Postal Union conventions are.

“Three-word addresses are easy to type.” For users in the language of the dictionary, yes. For non-native users, spelling errors are a real concern (///pizza.gushy.passed works fine; ///piza.gushy.past might not be a valid triple at all, or might point to a different location).

“what3words is just a UI for lat/lon.” Conceptually yes — it's an alternative encoding of the same underlying coordinates. But the encoding is proprietary, so applications cannot independently encode/decode without an API call.

“what3words solves the addressing problem.” It addresses the identification problem (giving any location a memorable identifier). It does not address the delivery problem (getting mail/services to that location), which requires postal infrastructure investment regardless of identifier format.

“Mongolia's adoption is universal.” Mongolia's postal service uses what3words as the official addressing system, but adoption is partial in practice. Urban Ulaanbaatar uses traditional addresses where they exist; rural and ger-district areas use what3words; international mail uses both.

“The dictionary will eventually be open sourced.” No public indication this is planned. The proprietary status is a deliberate business choice; the company has actively defended it against clones.

“what3words and Plus Codes are competitors that will one win.” They serve different needs and likely coexist. Plus Codes is the open computational standard; what3words is the proprietary verbal-friendly standard. Both have meaningful adoption; neither is displacing traditional postal addressing universally.