Frequently Asked Questions

NameGender estimates the gender statistically associated with a name in available datasets. It does not determine a person’s identity, legal gender, biological sex or self-identification.

You get the more common gender with its share, for example Riley as female at 57% in US records, and the number of people that share rests on. When the evidence is too thin to call, the answer is unknown rather than a coin flip. Send best_guess only if you would rather have the likelier gender than an unknown.

Some names change gender across borders: Andrea is mostly male in Italy and mostly female in Germany. A country makes the engine prefer that country’s records before the worldwide result. Send a two-letter code (DE), a three-letter code (DEU) or the country name as your CRM stores it (Germany, Deutschland). Without a country, a browser language tag such as it-IT or an IP address can stand in, and country_source in the response says which one was used.

For counted sources, probability is the observed share of the dominant label. Confidence describes the evidence level using sample size. A record without an auditable count is marked unverified and cannot show 100%.

Latin, Arabic, Cyrillic, Chinese characters, Korean Hangul and Japanese are read directly. In our published benchmark the share of names answered was 98% for Latin script, 99% for Arabic, 98% for Cyrillic and 100% for Chinese. Coverage is lower for Greek (81%), Hebrew (66%) and Thai (51%), and names we cannot place come back unknown instead of guessed.

No. Müller and Muller, Ayşe and Ayse, José and Jose find the same records, and Turkish dotted and dotless i are handled. Upper case, extra spaces and a surname after the first name are fine as well.

The active snapshot contains 8,235,264 unique name-country records across 201 country codes. Size is not treated as proof of accuracy; benchmark coverage and errors are reported separately.

Mostly from official birth and population registers that publish counts per name: the US Social Security Administration, the statistics offices of England and Wales, Scotland, Northern Ireland, France, Canada, Spain, Ireland, Norway, Austria, Switzerland, Poland and Israel. Openly licensed records such as Wikidata fill gaps, and records without counts are kept apart from counted ones.

Most official sources publish once a year, and we rebuild the data when a new year appears. Every response carries data_version, so a changed answer can be traced to a data release, and changes that affect results are listed on the changelog.

On our published benchmark of 4,610 labelled names from 25 countries, 94.3% got an answer and 96.6% of those were right. We also ran a sample of 991 of them through genderize.io and Namsor and published the result: Namsor came first, and we are roughly level with genderize.io. The research pages show the method, the names and the misses.

Yes. Names can be shared across genders, countries and writing systems. Use country, sample size, confidence and source fields, and keep unknown results as unknown instead of forcing a decision.

First names, full names (the first name is picked out), email addresses, usernames and lists of up to 100 values per request. The dashboard and the file API also take CSV and XLSX files of up to 1,000,000 rows.

Four more things from the same key and credits: a ready salutation line in ten languages (“Sehr geehrte Frau Müller,”), a check that flags junk names such as “asdf qwerty” in sign-up forms, the age spread of the people with a first name in the US, France and Norway, and the countries whose birth records contain a name.

By default 1,200 requests per minute per API key, counted per key and not per IP address. Every response carries X-RateLimit-Remaining, and a 429 response tells you how long to wait in Retry-After. A bulk request for 100 names counts as one request.

Names are looked up in a read-only index held in memory, not by a database query, so the lookup itself takes a few milliseconds. Every response reports took_ms, and a bulk request answers up to 100 names in one round trip.

Official client libraries for Python, JavaScript, PHP, Ruby, Go, Java, .NET, R, PowerShell and Rust, a command line tool, a Google Sheets add-on, an Excel add-in, Zapier, Make and n8n, and an MCP server for Claude, Cursor and other AI assistants. The integrations page lists each one with its status.

Yes. Send the endpoint, response status and a redacted request example to support. Never email a full API key.

By default the request log keeps the value you sent so you can review your usage. You can switch this off for any API key in the dashboard: from then on the name, email address or username is not written at all, not even hashed, and only the result, country, key and time are kept for billing.

The full key is displayed once. Only a cryptographic hash, prefix and final characters are stored. Keys can be restricted by IP or origin and revoked from the dashboard.

The source file is deleted when processing finishes, and the result file after 7 days. Job totals and credit records can remain after the files are removed.

Yes. The data processing agreement is published on the site and takes effect when you create an account, with no signature needed; a countersigned copy is available on request. Treat inferred gender as personal data and mention the enrichment in your privacy notice.

No. A name says something about a group of people, not about the person in front of you. Use the results for salutations, segments, research and data quality, never for hiring, credit, housing, insurance, eligibility or age checks.

No. Purchased credits remain in the account until used. Daily free credits reset and do not accumulate as a cash balance.

The current free plan provides 100 credits per day. Limits can change for future use and are always shown on the pricing page.

Yes. One credit per name, whether the answer is male, female or unknown, because the lookup is done either way. The exception is an age request for a country we do not cover, which costs nothing.

Yes. We email you when your credits will last less than a week at your current pace and again if they run out, and we also email you if your requests keep failing. With webhooks, the low and depleted credit events also reach your own system.

By card or with USDT. Card payments go through Paddle, which charges any tax that applies and emails an invoice you can download again from the dashboard. Crypto orders are listed in the dashboard with their date, amount and transaction reference.

For card payments, as soon as Paddle confirms the payment. For crypto, when the payment network reaches the required confirmations; the order page updates as the payment is detected and confirmed.

Yes, for the credits you have not used, within 14 calendar days of activation. The credits you did use are priced at the published rate for that volume and deducted from the refund, so a half-used package is not refunded at half price. Network fees are not refundable, subject to mandatory legal rights.