When a census hides as much as it shows: the 1891 census of Victoria, BC

In this post, I explore a genealogy question I’ve had for decades with a new tool: AI. I begin by explaining the genealogical problem, the AI usage, the results, and then the verdict on how well AI did.

What is going on with the 1891 census for Victoria, BC?

Imagine you’re looking for your family in the census. You have information they were living in Victoria, BC, Canada, in about 1890 and you think, Great! It should be simple to locate a family in the 1891 Canada census. But you can’t find them. All search results are nil. But you know they were there so what is going on? You start reading the hundreds of census pages manually and then you see the problem. They might be in the census but their names have been erased.

Trigger warning: I use historic language in this blog. It may not be suitable for all readers. Please be forewarned.

The first time I saw this, about a decade ago, I genuinely thought it was a mistake. The enumerator recorded all the characteristics – male, “dom” [domestic], from China, both parents born in China, religion recorded as heathen, employed as Cook – but neglected to get the name and wrote a racial description in the style of the times: Chinaman.¹ Just reading that word makes me suck in my breath.

Weird, I thought, and moved on. And it seemed that every page there was another mistake, until I found entire households of “mistakes.” And it seemed the enumerators only mostly recorded Chinese residents as Chinaman, because there were times when they seemed able to record the name. In this record, there are nine Chinese people recorded: two with names – Ah Lou and Ah Yow, and seven named Chinaman.²

I was shocked. The instructions to enumerators were clear. They were to record the names of every person in Canada where they normally resided. That’s why the census is called the Nominal Return: nominal, meaning names. At the time, I was researching families in the Johnson Street ward, and I put it down to one rogue enumerator. Here are three or more families. Same treatment: Chinaman, Chinawoman, China girl, and China Boy. I had thought perhaps the enumerators might record the names for Canadian born citizens, but apparently not. The last two entries in this images are for boys born in B.C.³

It’s now 2026. I was thinking about the importance of names. Genealogical practise practically centres around names, by entering a first name and surname in a database. I thought, wouldn’t it be an interesting research project to see how many Chinese were affected? To glean what we could from what information was recorded? To take back identity from that enumerator gone rogue? In years past, I’d have needed a research grant but today I have ChatGPT Plus. I already know it is capable of analyzing and processing vast reams of documentary data. I’m a curious person and I wanted to see what ChatGPT would do with a fairly meaty project.

Discovery no. 1 – Two thousand people were affected

Midway through the analysis, I realized I needed to read every census page and capture every Chinese person, named or not named, because the practise of erasing Chinese names runs across all the Victoria districts. I eventually read all 391 pages and downloaded a further 129 pages. Working with Chat was educational. As I revised the questions from Who was the enumerator for the Johnson Street Ward? to How many Chinese in the Victoria area were affected, what were their demographic characteristics, and what did they do for work, I had to do the work in batches.

The 1891 census manuscripts reviewed for Victoria and its surrounding census units contain at least 2,135 identifiable Chinese people. Only 98 were recorded under proper personal names. For 2,037 people, or 95.4%, the census substituted a generic racial designation for all or part of their personal name. Within Victoria’s three city wards, the rate was 98.8%. That is erasure for almost 96% of the population.

Here is a group of Chinese laundrymen in Victoria’s Esquimalt Ward.⁴

Here are two Chinese employed as cook and houseboy in the household of George A. Keefer, Yates Ward.⁵

Discovery no. 2 – Occupations for Chinese in Victoria

The hard work was compiling the data and finding the gaps. (Given there were some issues with transcribing 350 census pages, consider these results illustrative, not thorough.)

With the absence of 96% of the Chinese names in the area, I wondered what else could be gleaned from the data results. The analysis was solid for some wards, such as James Bay Ward, where relatively few Chinese lived. For example, here are how many Chinese who worked as domestic labourers in James Bay Ward.

Occupation	Number
Domestic	56
General servant/ servant	13
Servant total	69
Cook	14
Gardener	5
Steward	1
Total Chinese employed as domestic labour in James Bay Ward	89

Here is a one entry for a Chinese working as a domestic in James Bay Ward.⁶

And here is a breakdown of occupations for the Victoria area.⁷ What struck me was not the most commonplace occupations – labourers, laundry, domestics – but the sometimes startling and original entries. Under the category of “one-offs” were Actor, Cigar dealer, Coachman, Drying-house work, Hotel keeper, Milkman, Oil burner, Opium labourer, Physician, Piano maker, Saddler, and Tinsmith.

Occupation, normalized	Minimum documented count	Minimum share of 2,135 Chinese
Labourers	1343	62.9%
Laundry / washing	150	7.0%
Cook / cook helper	145	6.8%
Domestic / household servant	89	4.2%
Gardener / Garden labourer	40	1.9%
Saw-mill labourer	32	1.5%
Farmer	22	1.0%
Shoemaker / boot & shoe	22	1.0%
Tailor	20	0.9%
Tea / grocery merchant	20	0.9%
Peddler	17	0.8%
Boot / store keeper	17	0.8%
Brick / masonry work	16	0.7%
Farm labourer / farm hand	15	0.7%
Porter	14	0.7%
Woodcutter	10	0.5%
Lodging-house work	10	0.5%
Store / grocery keeper	10	0.5%
Butcher	9	0.4%
Carpenter	8	0.4%
Lumberman	8	0.4%
Market gardener	7	0.3%
General / produce dealer	7	0.3%
Agent	7	0.3%
Merchant / dry-goods	7	0.3%
Railway labourer	6	0.3%
Barber	6	0.3%
Restaurant work / waiter	7	0.3%
Dockyard labourer	4	0.2%
Sectionman	4	0.2%
Cart driver	3	0.1%
Compositor	3	0.1%
Salesman	3	0.1%
Baker	3	0.1%
House keeper	2	0.1%
Jeweler	2	0.1%
Manager	2	0.1%
Other specified one-off occupations	23	1.1%
No occupation recorded in census	22	1.0%
TOTAL	2,135	100%

Discovery no. 3: No rogue enumerators

You’ll recall I started this research thinking I was dealing with one rogue enumerator. As the research spread out, from one ward to two, and finally all the Victoria area wards, I saw the same pattern repeated. I was not looking at the work of one rogue enumerator, but rather the work of 11-13 enumerators.

Enumerator as recorded	Area	Division / source unit
T. T. Wilson	Johnson Street Ward	Division 1
B. W. Ward	Johnson Street Ward	Division 2
Noel F. Townsend	Johnson Street Ward	Division 3
Robert Edgson	Johnson Street Ward	Division 4
G. G. O’Driscoll	Yates Street Ward	Division 5
E. A. Bickmore	Yates Street Ward	Division 6
F. Hammond	Yates Street Ward and part of James Bay	Division 5
John Kirsch	James Bay Ward	Division 7
Jno. G. O’Driscoll (could be G.G. O’Driscoll)	James Bay Ward	Division 8
Alex W. Hardie	James Bay Ward	Division 9
N. F. Townsend (could be Noel F. Townsend)	Esquimalt	sub-district d / Polling District 11
Edward Day Twiss	Victoria District	sub-district f / Polling District 10
W. Fisher	Metchosen District	District 12

However this happened, whoever decided, whoever approved the final results, this was the result of many people across the entire Victoria census area.

Was this legal?

In a word, no. The 1891 instructions to enumerators was clear.⁸

11. Every enumerator, by visiting every house and by careful personal inquiry, shall ascertain, in detail, with the utmost possible accuracy, all the statistical information with which he is required to deal, and no other, and shall make an exact record thereof, and attest the same under oath…

The “Directions Concerning the Separate Schedules, Schedule No. 1, Nominal Return of the Living,” was even more clear.⁹

The names of living persons, all belonging to each Census family… are to be entered in full, in the following manner, and as shown in the specimen schedule: —

Jones, William.

” Mary.

The family name being first given, and then the Christian name.

How did this happen?

This is the question. To understand how it happened, I first ask who was involved. Who was up the food chain?

The chain of command was fairly clear. Wilson, Ward, Townsend, Bickmore, Hardie and the other Victoria enumerators worked under John B. Lovell, the census commissioner for Victoria, whose job was to coordinate their work and review the nominal schedules line by line for accuracy before forwarding them onward. Lovell in turn reported to George Sargison, Chief Census Officer for British Columbia, who had been trained in Ottawa by George Johnson, the federal chief statistician and head of the temporary Census Branch. The census itself was administered by the Department of Agriculture, under Deputy Minister John Lowe and Minister John Carling.

Breaking that down, it was the enumerators responsibility to enumerate correctly, Lovell’s job to check every line was right, and the census administration to detect and remedy a systemic breach. Historian Patrick Dunae independently found Lovell problematic: Sargison complained that Lovell resisted training, ignored established procedures and preferred methods of his own. Dunae describes him as an incompetent census commissioner.¹⁰ Research to date has not uncovered evidence supporting who gave the orders, or even if it was strongly questioned. As genealogists say, further research is needed.

Did the same practice of erasure occur elsewhere in Canada?

I can’t say it did not but what I can say is that the same erasure did not seem to happen in other B.C. districts. Here are three examples from the census districts of New Westminster, the Cariboo, and Yale.

Here is the 1891 census of WON Ling Sing of New Westminster.¹¹

Here are a group of gold miners in the Cariboo District.¹²

And here is a group of men working at a store in Yale.¹³

The second Victoria area census of 1891

Surprise: there are actually two Victoria censuses for 1891.¹⁴ After the federal census reported a population of 16,841, civic leaders protested that Victoria had been undercounted. The city responded by organizing its own check census in September 1891, only a few months after the federal enumeration. That municipal count produced a population of 22,981.

At first glance, this sounds like an ideal second source for checking the federal census and perhaps recovering some of the Chinese names that federal enumerators had replaced with Chinaman and similar labels, but the second census was not done the same way. It generally names only a household representative, with the other people in the household recorded as numbers. Chinese householders and business operators do sometimes appear under personal names, which is evidence that their names could be obtained and recorded. Sadly, it can’t be considered a replacement census where the erased names can be recovered.

Its methodology was also in doubt. Although it counted more than 6,000 additional residents overall, it actually counted fewer Chinese residents than the federal census. The two enumerations had different purposes and methods, but together they provide a revealing comparison of who was counted, how they were described, and whose identity made it onto the page.

Working iteratively with AI

The findings in this post came after days of working with ChatGPT. My original research question – How many Chinese were affected in the Johnson Street Ward? – was the wrong question to ask. Because I’d assumed the substitutions of words like Chinaman, Chinawoman, and Chinaboy were the work of one enumerator, my first group of forty-nine records only looked at the clusters of Chinese living together in one ward, with a few samples from the James Bay Ward and the Yates Street Ward.

I hit Chat’s limits with this project:

• Too many words in one thread
• Too many uploads at once
• Too many records to analyze at one time

I learned to do work in batches. I uploaded documents in groups of 5, 10, and 20. I asked for results in stages, i.e., process and analyze half these records and update the workbook. I checked up on Chat as it was working – still there? – and twice I had overloaded its processing capabilities. This project had me carefully composing my questions, because the responses took from 30-55 minutes to come back. I think it’s important to recognize what AI can and cannot do. It is an amazing tool and improving rapidly, but it can’t currently process a couple hundred documents at once. It struggles with illegible records and tables like censuses. This analysis is therefore a work in progress. It is more than I could do solo but it is less than could be done by, say, a research team transcribing records by hand.

How well did ChatGPT do with this project?

I looked online for a checklist to help me evaluate AI’s results and weirdly did not see anything. It seemed wrong to ask ChatGPT to come up with ideas, so at the risk of approaching a subject like a complete noob, here are my questions and answers.

• How well did ChatGPT understand the subject?
• Evaluate the accuracy of the analysis.
• How well did ChatGPT process the raw files?

How well did ChatGPT understand the project?

9/10. I have trained ChatGPT to be skeptical of my claims and to ask me to prove my assertions. At the same time, nine months of working with me continually on Chinese Canadian genealogy is a body of work that it can see. It knows already what I’m interested in seeing, what I will accept as sources, and when to qualify its answers.

Evaluate the accuracy of the analysis

6/10. During the course of this mini-project, I had AI produce a series of workbooks containing statistical analysis. With each iteration, I instructed AI to update the workbook, but not everything in the iterative process was captured. It was instructive to see what Chat did and did not do. The longer the processing time to complete, the more important it became to check the results. It’s easy and tempting to skip the check but the few times I didn’t check, I regretted it.

Here is an example from the analysis of the Esquimalt ward census created by ChatGPT.

How well did ChatGPT process the raw files?

5/10. As is known, AI does best when it is recognizing text in a clear, text recognized document. Asking ChatGPT to process hundreds of handwritten census records, which are a table format, with uneven levels of legibility, is a risk. AI did well recognizing the key terms (Chinaman, Chinagirl, etc.) but did poorly on reading across and down. It was confused by ” marks. ChatGPT Plus has a document upload limit of 20/text and a further limit of total uploads. For a large project, one is going to run into both of these limits. I found that when I hit the limit of words in a thread, ChatGPT had some trouble recalling previous work.

Final score: 67%

Afterword

It was fascinating to apply new technology to an old problem. I learned more about the problematic 1891 census of Victoria, found an alternative census I didn’t know existed, and was able to tabulate some information about the men, women, and children of Chinese ancestry in the Victoria area. I have yet to positively connect any of these erased names to their owners but this exercise showed me what was possible.

As with every genealogical question, I’ve answered one question and created many more. Now this is done, I’d like to know:

• How was the erasure of over two thousand Chinese names and identities overlooked by the census administrators whose responsibilities included an accurate census?
• If this was not oversight and was instead methodical, where is the directive?
• Why was the erasure performed in Victoria and nowhere else?
• And also: What else can I do to leverage AI to restore the identities of this important population?

The joke in genealogy is: “My work is done,” said no genealogist. Ever.

Thank yous

Thanks this week go to my ancestors, who survived a sometimes hostile environment to give me the great gift of Canadian citizenship.

This post was written by the author and draws on her research project of the 1891 Victoria census using ChatGPT.

Resources and Notes

1. 1891 census of Canada, British Columbia, Victoria City, Johnson Street Ward subdistrict, digital image, image 8 of 179, p. 16, family 85, line 4, census of Finlayson, R household, “Chinaman,” Ancestry (https://www.ancestry.com : accessed 12 August 2026), database and images, “1891 Census of Canada.” ↩︎
2. __ Johnson Street Ward subdistrict, digital image, image 46 of 179, p. 91, family no. 54, line 2, census of Ah Lou household, and family no. 55, line 9, census of Ah Yow household, Ancestry (https://www.ancestry.com : accessed 12 August 2026), database and images, “1891 Census of Canada.” ↩︎
3. __ Johnson Street Ward subdistrict, digital image, image 56 of 179, family no. 89, line 2, household of “Chinaman,” family no. 90, line 7, household of “Chinaman,” and family no. 91, line 12, household of “Chinaman,” Ancestry (https://www.ancestry.com : accessed 12 August 2026), database and images, “1891 Census of Canada.” ↩︎
4. __ Esquimalt subdistrict, digital image, image 3 of 32, family no. 23, line 9, household of “Chinaman,” family no. 90, line 7, household of “Chinaman,” and family no. 91, line 12, household of “Chinaman,” Ancestry (https://www.ancestry.com : accessed 12 August 2026), database and images, “1891 Census of Canada.” ↩︎
5. __Yates Ward subdistrict, digital image, image 88 of 88, lines 12-13, household of George Keefer with “Chinaman” cook and also houseboy, Ancestry (https://www.ancestry.com : accessed 12 August 2026), database and images, “1891 Census of Canada.” ↩︎
6. __James Bay Ward subdistrict, digital image, image 20 of 82, line 16, household of “Chinaman,” Ancestry (https://www.ancestry.com : accessed 12 August 2026), database and images, “1891 Census of Canada.” ↩︎
7. Chat had difficulty analyzing the Johnson Street Ward entries for occupations, for reasons ranging from illegible records to a great number of copied entries. For this reason, the category of labourers is larger than it should be, as I made a decision to group the uncategorized occupations with labourers. ↩︎
8. Canada, Revised Statutes of Canada, 1886, chap. 58, § 11, “An Act respecting the Census,” reproduced in Canada, Department of Agriculture, Census Branch, Manual Containing “The Census Act,” and the Instructions to Officers Employed in the Taking of the Third Census of Canada (1891) (Ottawa: Brown Chamberlin, Printer to the Queen’s Most Excellent Majesty, 1891), 2; digital image, Government of Canada Publications, https://publications.gc.ca/site/eng/9.828167/publication.html : accessed 10 August 2026. ↩︎
9. Canada, Department of Agriculture, Census Branch, Manual Containing the Census Act and the Instructions to Officers Employed in the Taking of the Third Census of Canada (1891) (Ottawa: Brown Chamberlin, Printer to the Queen’s Most Excellent Majesty, 1891), 12, “Schedule No. 1—Nominal Return of the Living,” instructions for column 6; digital image, Government of Canada Publications, https://publications.gc.ca/collections/collection_2016/statcan/CS98-1891I.pdf : accessed 10 August 2026. ↩︎
10. Patrick A. Dunae, “Making the 1891 Census in British Columbia,” Histoire sociale / Social History 31, no. 62 (November 1998): 227–28, 235–36; digital images, Histoire sociale / Social History, https://hssh.journals.yorku.ca/index.php/hssh/article/view/4666 : accessed 10 August 2026. ↩︎
11. 1891 census of Canada, British Columbia, New Westminster District, New Westminster City subdistrict, family 202, 1891 census of WON Ling Sing household, image 121 of 142; Library and Archives Canada, RG31-C-1, Census of Canada, 1891, microfilm T-6290; digital image, Ancestry (https://www.ancestry.com : accessed 10 August 2026), database and images, “1891 Census of Canada.” ↩︎
12. 1891 census of Canada, Library and Archives Canada, RG31-C-1, Census of Canada, 1891, Statistics Canada fonds, microfilm series T-6290–T-6427, image 7 of 23, digital image, Ancestry (https://www.ancestry.com : accessed 10 August 2026), database and images, “1891 Census of Canada.” ↩︎
13. 1891 census of Canada, Yale District, Lytton subdistrict, family no. 5, household of Ow Kwong Chin, image 1 of 28; Library and Archives Canada, RG31-C-1, Census of Canada, 1891, Statistics Canada fonds, microfilm series T-6290–T-6427; digital image, Ancestry (https://www.ancestry.com : accessed 10 August 2026), database and images, “1891 Census of Canada.” ↩︎
14. City of Victoria (British Columbia), 1891 Census Return for the City of Victoria, Library and Archives Canada, MG 9, F 3, microfilm H-1408; digital images, Héritage Canadiana (https://heritage.canadiana.ca/view/oocihm.lac_reel_h1408/1 : accessed 10 August 2026.) ↩︎