2.4 million Americans share the surname Smith

The 2010 Census counted 2,442,977 people with the surname Smith. That is not a trivia fact. It is the reason a people-search product that treats a shared name as identity will eventually describe the wrong person.

The number comes from the Census Bureau's Frequently Occurring Surnames from the 2010 Census table, last revised on 8 October 2021. The technical documentation behind it, Joshua Comenetz, October 2016, is more useful than the top-ten list: about 6.3 million distinct surnames were reported; 11 of them appeared more than a million times; 3.9 million surnames, 61.9 percent of the distinct set, were reported only once; and surnames were recorded for 295 million people, 95.5 percent of the enumerated population.

We did not collect a new sample. We read the official table, added the published counts, and asked what those figures do to a name search. The product note explains the matching rule we chose. This post is the dataset underneath that rule.

The finding

People counted · 2010 Census

1. Smith2,442,977
2. Johnson1,932,812
3. Williams1,625,252
4. Brown1,437,026
5. Jones1,425,470
6. Garcia1,166,120
7. Miller1,161,437
8. Davis1,116,357
9. Rodriguez1,094,924
10. Martinez1,060,159

The ten most common surnames in the 2010 Census, as the Bureau printed them:

| Rank | Surname | People counted | | --- | --- | ---: | | 1 | Smith | 2,442,977 | | 2 | Johnson | 1,932,812 | | 3 | Williams | 1,625,252 | | 4 | Brown | 1,437,026 | | 5 | Jones | 1,425,470 | | 6 | Garcia | 1,166,120 | | 7 | Miller | 1,161,437 | | 8 | Davis | 1,116,357 | | 9 | Rodriguez | 1,094,924 | | 10 | Martinez | 1,060,159 | | | Top ten | 14,462,534 |

Those ten names account for 14,462,534 people. Against the 295 million people for whom a surname was recorded, that is 4.9 percent. One surname, Smith, is 0.83 percent of that same base.

The Bureau also published File B: 162,253 surnames that occurred at least 100 times. Most of the 6.3 million distinct surnames sit below that threshold. A rare name is common in the file. A common name is common in the population. People search fails in the second case, and it fails quietly in the first when two rare-looking strings are actually the same person spelled two ways.

What the figures do not say

They do not say how many people are named John Smith.

The 2010 product tabulated surnames. Comenetz is explicit that first names were used only as an edit check carried over from the 2000 programming, not as a published frequency list. Anyone who quotes an official “John Smith” count from this table invented it.

They also do not describe 2026. The underlying census is 2010. The documentation is 2016. The HTML table was last revised in 2021. Garcia's rise to sixth place, noted in the same documentation, already showed the list moving with the population. A 2026 search will not see the same counts. It will see the same shape: a few names shared by millions, a long tail of names shared by almost no one, and a large middle where “uncommon in my circle” is still common in a country.

The edits matter for search. The Bureau deleted punctuation and spaces, so O'Hara and O Hara both become OHARA in the file. Suffixes were stripped. Obvious non-names were removed. A people-search index that keeps the punctuation the person actually uses, or that does not, will miss or merge records for reasons that have nothing to do with identity.

The presentation “does not in any way identify any specific individuals.” That is the Bureau's sentence, and it is the right one. These are aggregates. They tell you about collision risk. They do not tell you who anyone is.

The regulator already did the next step

Surname frequency is only half of a name search. The other half is the first name. The Consumer Financial Protection Bureau stated the joint problem in 2021: thousands, and in some cases tens of thousands, of people may share one first-and-last combination. Name-only matching, in that statement, means using first and last name without address, date of birth, or Social Security number.

The Bureau of the Census and the CFPB are answering different questions. One counted last names. The other told consumer reporting agencies that matching on a name alone is not a reasonable accuracy procedure under the Fair Credit Reporting Act. The advisory opinion is about regulated reports, not public-web search. The arithmetic is still the same. If a first-and-last pair can describe tens of thousands of people, a product that fuses every record sharing that pair is not being thorough. It is being lazy in a way the data already forbids.

The CFPB also said the risk is not evenly distributed. Name-only matching is more likely to mix people in Hispanic, Black, and Asian communities because those populations have less surname diversity than the non-Hispanic white population. Comenetz's documentation is consistent with that warning: the 2010 list is not a neutral ranking of English names. Garcia, Rodriguez, and Martinez are in the top ten. A matcher that treats “common name” as a Smith problem has already misunderstood the file.

We are not a consumer reporting agency. The FCRA line is a separate post. The reason the advisory opinion belongs here is narrower. If name-only matching is too weak for a report that can cost someone housing, it is too weak for a public profile that can be screenshotted, forwarded, or believed.

What a collision looks like in a result

The failure is not an empty page. It is a finished-looking one.

Two public pages share a surname. One has a city. The other has a job title. A third has a photograph. The product writes a single card. The card has a face, a timeline, and three sources. Completeness is the error. A thin “possible match” can be checked. A fused biography asks the reader to believe a person who is actually two people, or three.

The Census table tells you how often that setup is available. There are more than two million Smiths. There are more than a million people for each of eleven surnames. File B has 162,253 names that already occur at least a hundred times. A searcher who types a name from that list and accepts the first rich card is not using a rare identifier. They are using a bucket.

A rare surname does not rescue the method. 3.9 million names appear once in the 2010 file, including unique spellings and transcription errors. Two records that share an unusual string can still be different people, or the same person recorded twice, or a relative. Rarity changes the prior. It does not finish the proof.

How to use the table without misusing it

Treat the Census file as a collision prior, not as a biography.

1. If the surname is in the top ten, or anywhere in File B, do not search the name alone. Add a public anchor you already know: a city, an employer, a school, a handle, a domain, a date range. The common-names guide is the worksheet.
2. If the surname looks rare, still keep candidates separate. A unique string in 2010 can be a spelling variant of a common name, or a person who has since been joined by namesakes.
3. Do not compute a first-and-last count from this table. The table does not support it. Use the CFPB's qualitative range — thousands, sometimes tens of thousands — and then look for independent evidence.
4. Do not treat a 2010 rank as a 2026 rank. Use the shape: a short head, a long tail, and a large collision zone in between.
5. Watch punctuation. A search that requires O'Hara will miss OHARA, and the reverse. The Bureau collapsed those forms on purpose. A people-search index may not.

Independent evidence means a separate path, not a second copy of the same directory row. Two aggregator pages that reprint one listing are one source. A company page, a personal domain, and a handle that all name the same employer is the beginning of a case.

A conflict stops the merge. Two cities in the same year, two ages that cannot both be true, a photo history that belongs to someone else: those are not details to average. They are the reason to keep the records apart.

Methodology

This is a reading of a published official dataset, not an original survey.

• Primary table: Census Bureau, Frequently Occurring Surnames from the 2010 Census, top-ten counts as printed, page last revised 8 October 2021.
• Technical documentation: Comenetz, “Frequently Occurring Surnames in the 2010 Census,” October 2016, including the 6.3 million distinct surnames, the 11 names over one million, the 3.9 million singleton surnames (61.9 percent), the 295 million people with a recorded surname (95.5 percent coverage), and File B's 162,253 names occurring 100 or more times.
• Arithmetic we performed: the top-ten sum (14,462,534) and the two percentages against 295 million (4.9 percent and 0.83 percent). We rounded those percentages to one decimal place and two decimal places respectively.
• What we did not do: download File A or File B and re-rank them; estimate first-name frequencies; inflate the 2010 counts to 2026; identify any person; or treat DeepSearch results as a sample.
• Second source: the CFPB's 2021 advisory opinion and the accompanying statement on name-only matching. Those documents are about consumer reporting agencies. We use them for the joint first-and-last collision problem and for the distributional warning, not as a claim that DeepSearch is a consumer reporting agency.

If the Bureau publishes a 2020 or 2030 surname product, this post should be revised against that file. Until then, the honest vintage is 2010.

What we refuse to do with this file

We will not turn it into a “how unique is your last name” toy. The Bureau already warned that the presentation does not identify individuals. A uniqueness score built on a 2010 aggregate would be a novelty metric with a government logo behind it.

We will not use it to imply that a rare surname makes a DeepSearch match safe. The product rule is unchanged: a name alone never establishes a match, a single overlapping detail is a hypothesis, and results are ranked so you choose a person rather than being handed one. That rule is in the transparency post and in our terms.

We will not pretend the table is current. It is the best official national surname-frequency file the Bureau has published in this series. That is a reason to cite it. It is not a reason to date a 2026 headline with 2010 counts as if they were this year's.

If you are searching, or being searched

If you are looking for someone, start with the strongest public clue you already have and keep every same-name result in its own row until independent sources agree. The worksheet is How to search common names without mixing people. How to check a specific claim is How to verify a DeepSearch result using its sources.

If a result mixes you with someone else, that is a correction. Send the result URL, the wrong field, and the source it cites. How to correct or remove information from DeepSearch is the process. Remove my info is the public form.

The Census file will not tell us which of those cases you are in. It only tells us why the cases keep happening. Millions of people share a few dozen names. A product that ignores that is not searching. It is collapsing.