From Newspaper Search to Historical Investigation: What SNEWPapers AI Tools Can Do

All source materials for this post were acquired using SNEWPapers. We are in the process of the most accurate extraction, in terms of correct segmentation and textual accuracy, of the Library of Congress' Chronicling America collection ever conducted. You can track our progress here. This post is a demonstration of what you can find using our tools. Connect your Grok, ChatGPT, Claude and more today to our vast database and research whatever you're interested in (and hopefully share it with the rest of us!). What would've taken you eons yesterday, is possible in a few hours with SNEWPapers today. Estimates put the total count of historical newspaper pages at over a billion for U.S. Newspapers alone, with 25+ million sitting in the Library of Congress. By and large, nobody knows what's in them... until now

What happens when you ask SNEWPapers a question?

When you connect your AI tools to SNEWPapers, they acquire all the skills you need to start deep diving into the components we've extracted from the corpus. You can ask it what it's capable of and it will tell you, but sometimes an example is useful as well. Everything below comes directly from a ChatGPT chat hooked up to SNEWPapers about the following question:

What did the newspapers report about the Johnstown Flood of 1889, how did their understanding of the disaster evolve, and who did they hold responsible?

A brief note

We chose this subject intentionally. The LoC actually uses this example here. Their results are mostly focused on the first two days, and they describe how you need to use keywords and a tight date range. I'd encourage you to try it their way before the SNEWPapers way!

Another brief note

Some of the things mentioned below might happen in the background while it's reasoning about your questions, but they can all be explicitly requested as well, whenever you want

The corpus

Your AI will be able to investigate the overall corpus, and it might first do some preliminary investigations. You could also ask it directly yourself. It'll gather the most recent stats for you, and perhaps display something like:

The corpus is actively growing, so these numbers are not permanent.
Follow the live progress here

November 4, 2026 snapshot:

Historical span: 1736–1963
Newspaper titles: 2,732
Indexed components: 4.77M

Decade of publication, November 4, 2026 snapshot. The corpus is actively growing, so these numbers are not permanent.

Show data

	Components
1730s	4,637
1740s	1,417
1750s	8,994
1760s	38,250
1770s	73,904
1780s	45,936
1790s	113,959
1800s	143,420
1810s	197,107
1820s	202,027
1830s	237,838
1840s	232,980
1850s	255,215
1860s	298,227
1870s	374,021
1880s	476,419
1890s	546,656
1900s	482,150
1910s	414,641
1920s	342,323
1930s	272,851
1940s	23,860
1950s	11,706
1960s	119

1. Local news 1,025,305 21.3%
2. Domestic news 944,979 19.6%
3. Notices 710,893 14.8%
4. Filler 402,415 8.4%
5. Foreign news 307,950 6.4%
6. Editorials 278,420 5.8%
7. Personal ads 224,338 4.7%
8. Obituaries 122,714 2.6%
9. Stories 110,212 2.3%
10. Other labels* 684,887 14.2%

*Other labels (23 types, largest first): Auction, Commercial, Letter to editor, Sports report, Wedding, Literary, Society, Poem, Shipping, Cost of goods, Reward, Runaway, Train schedules, Theater listings, Review, Weather report, Birth announcement, Recipe, Epigraph, Errata, Radio listings, Crossword puzzle, TV listings.

Show data

	Components	Share
Local news	1,025,305	21.3%
Domestic news	944,979	19.6%
Notices	710,893	14.8%
Filler	402,415	8.4%
Foreign news	307,950	6.4%
Editorials	278,420	5.8%
Personal ads	224,338	4.7%
Obituaries	122,714	2.6%
Stories	110,212	2.3%
Other labels*	684,887	14.2%

From this it will know that it might be able to answer your question, since it probably is already aware internally that the Johnstown Flood happened in 1889.

Slicing and dicing

SNEWPapers might suggest your AI tool zoom into total indexed event documents from newspapers published in each state during May 1–July 31, 1889 (top 15).

• May
• June
• July

1. Michigan 1,752
2. California 1,560
3. West Virginia 987
4. Indiana 810
5. Kansas 730
6. Texas 653
7. Kentucky 565
8. Rhode Island 543
9. District of Columbia 533
10. Tennessee 502
11. Montana 474
12. Colorado 451
13. Delaware 416
14. Wisconsin 393
15. Maryland 336

Everything indexed from each state's papers in those three months, not only coverage of the flood. A missing month means none of that state's issues for the month have been processed yet.

Show data

	May	June	July	Total
Michigan	339	342	1,071	1,752
California	268	755	537	1,560
West Virginia	246	181	560	987
Indiana	0	284	526	810
Kansas	730	0	0	730
Texas	0	136	517	653
Kentucky	118	0	447	565
Rhode Island	213	129	201	543
District of Columbia	95	438	0	533
Tennessee	311	191	0	502
Montana	0	428	46	474
Colorado	0	0	451	451
Delaware	191	0	225	416
Wisconsin	0	0	393	393
Maryland	212	124	0	336

It might also suggest using semantic and keyword filters, for example:

• Required keyword: Johnstown
• Additional keywords: flood deluge dam disaster conemaugh reservoir relief
• Semantic description: “Reports and reactions to the catastrophic May 1889 Johnstown Pennsylvania flood caused by the South Fork dam failure, including deaths, rescues, relief and responsibility.”

Now we're narrowing in on the interesting content...

1. Indiana 32
2. California 15
3. District of Columbia 12
4. Michigan 12
5. New York 12
6. Texas 11
7. Montana 10
8. West Virginia 9
9. Colorado 8
10. New Jersey 7
11. Maryland 6
12. North Dakota 6
13. Missouri 5
14. Nevada 5
15. Kentucky 3

Pennsylvania, where the flood happened, is missing from both lists. That is a coverage issue, not a lack of reporting at the time: we have processed only 7 Pennsylvania issues from all of 1889 so far, and Chronicling America may hold few Pennsylvania papers for the period as well (the Library of Congress' own Johnstown guide includes only one, the Pittsburg Dispatch of June 2, 1889). Expect this to change as more of the collection is processed.

Show data

	Matching components
Indiana	32
California	15
District of Columbia	12
Michigan	12
New York	12
Texas	11
Montana	10
West Virginia	9
Colorado	8
New Jersey	7
Maryland	6
North Dakota	6
Missouri	5
Nevada	5
Kentucky	3

Aggregations

SNEWPapers will also suggest various aggregations, depending on the context. Labels in SNEWPapers are a top-level categorization of content type. For example, the top labels for the narrow query above are:

1. Domestic news 96 56.8%
2. Local news 27 16.0%
3. Editorial 21 12.4%
4. Filler 5 3.0%
5. Notice 5 3.0%
6. Story 5 3.0%
7. Foreign news 4 2.4%
8. Commercial 2 1.2%

Show data

	Matching documents	Share
Domestic news	96	56.8%
Local news	27	16.0%
Editorial	21	12.4%
Filler	5	3.0%
Notice	5	3.0%
Story	5	3.0%
Foreign news	4	2.4%
Commercial	2	1.2%

So now we know more or less where to zoom in.

Each label type, for example Editorial above, also has a vast system of sub-types, keywords and such beneath it. Each type is also extracted in its own unique way.

A document can have one primary type, and multiple sub-types based on context. If we asked what the sub-types of the Editorial documents found were:

1. Social reform 8
2. Infrastructure 6
3. Religion / morality 3
4. Science / medicine 3
5. Legal reform 2
6. Crime / punishment 2
7. Satire 2

A document can carry several sub-types, so these 26 tags come from the 21 editorials above.

Show data

	Tags
Social reform	8
Infrastructure	6
Religion / morality	3
Science / medicine	3
Legal reform	2
Crime / punishment	2
Satire	2

Summaries

Now you could say something like:

"let's focus on domestic news, editorial and story, and using just the summaries, grab what you think could be the top 20 most relevant and display their summaries."

To which you'll see something along the lines of:

The hybrid search returned 122 matching components across domestic_news, editorial, and story, covering May 1–July 31, 1889.

Below are the top 20 by relevance, using only SNEWPapers' generated summaries, without retrieving full article text. I've preserved the summaries verbatim so we can assess their usefulness.

One result (#5) concerns a different flood in Johnstown, New York. I've retained it to illustrate an important false positive.

And so on... (all 20 are listed under References at the end of this post).

What stands out

From ChatGPT:

Even from these summaries alone, several research threads emerge: dramatically changing death estimates, early identification of the South Fork club as the reservoir's owner, eyewitness accounts and warnings, competing arguments over responsibility, the national relief effort, and eventually legal action against the dam's owners.

The most promising contrast is between the June 8 editorial that hesitates to assign direct blame and the July 19 editorial that explicitly accuses the owners. That could give the historical investigation a compelling narrative arc.

The horseman story also appears independently in both domestic_news and story labels. That's worth investigating: it may be the same syndicated account republished elsewhere, rather than two independent eyewitness sources.

Diving into the full text

Pulling too much full-text will eat up your AI's context, so you might not want to throw a thousand articles at it, but at this point, we have a decent cohort so we can deep-dive to an interesting result. SNEWPapers will also suggest a nice research folder layout if you want to store more articles, your queries and such. If whatever you're using has access to your local folders, it can use that, or perhaps if you're connected to Google Drive or something like that, it can use that. It will know how to save everything on your behalf.

For this demonstration, I'll skip those features, and just ask it:

OK, so let's do a quick writeup. Grab the full-text of these articles into your memory first. Can we generate summary of what happened and when (general synopsis of the event). Then, in a timeline fashion, blending in information from the full-texts (with links), could you tell me about early reports, changing death estimates, eyewitness accounts and warnings, culprits, arguments over responsibility, relief efforts and legal actions. Feel free to dive into that horseman story too..

The story, as told by ChatGPT, according to the Newspapers at the time...

The Johnstown Flood of 1889: Disaster, Warnings, Blame and the Search for Answers

An investigation through American newspapers, using SNEWPapers

On May 31, 1889, a wall of water swept through the Conemaugh Valley in Pennsylvania, destroying communities and devastating the industrial town of Johnstown. The catastrophe became one of the deadliest floods in American history. But in the days and weeks that followed, Americans reading their newspapers encountered a much less settled story.

How many people had died? Had anyone warned the townspeople? Was the disaster an unavoidable consequence of extraordinary rainfall—or the result of a dangerous dam maintained for the recreation of wealthy men? And would anybody be held accountable?

Drawing from twenty historical newspaper articles, we'll reveal how those questions developed between June and July 1889.

What happened at Johnstown?

Johnstown was a busy industrial town situated where the Little Conemaugh River and Stony Creek meet. Its ironworks, railways, shops and workers' homes occupied a valley already vulnerable to flooding.

Upstream stood the South Fork Dam, holding back Lake Conemaugh, a reservoir owned by the South Fork Fishing and Hunting Club. Originally built to support Pennsylvania's canal system, the reservoir had become a private recreational lake.

After exceptionally heavy rainfall on May 30 and 31, the reservoir rose beyond the dam's capacity. Workers attempted to release water and prevent a breach, but the structure failed at approximately 3:10 p.m. on May 31.

The released water raced down the valley, sweeping through South Fork, Mineral Point, East Conemaugh and Woodvale before reaching Johnstown around 4:07 p.m. Houses, trees, railway equipment and other wreckage were carried into the town. Debris accumulated against the Pennsylvania Railroad's stone bridge, where a fire added another horror to the flood.

The official death toll was eventually recorded as 2,209, although the actual number may have been higher.

Those are the broad facts established by later historical research. The contemporary newspaper accounts show how uncertain and contested the picture initially was.

June 1: A catastrophe reported through broken communications

The first reports were fragmented. Telegraph lines were down, railway traffic was interrupted and correspondents struggled to establish what had happened.

The Evening Capital of Maryland, June 1, reported that violent rains had caused the Conemaugh Lake reservoir to burst. Its dispatch described a flood rushing through the valley and the apparent devastation of South Fork, Mineral Point, Conemaugh and Woodvale.

The article conveyed the scale of the disaster, but its language also revealed uncertainty. It feared that entire communities might have been swept away without knowing how many people had escaped.

Another Evening Capital dispatch carried a report from a courier arriving at Blairsville that 1,500 lives had been lost.

That was not a verified death toll. It was an early estimate transmitted under extraordinary circumstances.

A separate report from Sang Hollow described bodies floating downstream and quoted a railway superintendent's assessment that Johnstown was virtually destroyed. The same dispatch described a mass of buildings and driftwood piled near the bridge and burning.

The earliest reporting thus combined specific observations, secondhand testimony and sweeping estimates.

This is an important feature of historical newspaper research: an article published close to an event may preserve valuable firsthand information while simultaneously repeating claims that later prove exaggerated or mistaken.

June 1: The dam's wealthy owners were already identified

Questions about the reservoir's ownership appeared almost immediately.

In another June 1 Evening Capital report, readers learned that Lake Conemaugh belonged to the South Fork Fishing and Hunting Club, an organization associated with wealthy Pittsburgh men.

The newspaper described the reservoir's dimensions and the dam's construction, then offered a preliminary explanation: a cloudburst might have increased the volume of water beyond what the structure could withstand.

That explanation matters because it presents the breach primarily as the result of exceptional natural forces.

Yet the same article also supplied information that would become central to arguments over responsibility: the dam was no longer merely a piece of public infrastructure. It was associated with a private leisure club.

Ownership was established early. Responsibility was not.

June 3–5: Eyewitness accounts, extraordinary escapes and the mysterious horseman

As journalists and survivors supplied more detail, newspapers began publishing stories of individual experiences.

The June 22 edition of Fair Play collected dramatic accounts of people escaping on floating buildings, rooftops and wreckage.

One account described a man named Dougherty surviving by riding a telegraph pole downstream. Another told of the McMillan children escaping across the roofs of wrecked houses. The article also described a family reportedly saved with the assistance of their St. Bernard dogs.

Such stories gave the disaster a human dimension that casualty statistics could not convey. They also demand caution: highly dramatic accounts, especially when collected and republished at a distance, should be checked against independent evidence before being treated as established fact.

The nameless Paul Revere

Among the most memorable stories was that of an unidentified rider on a large bay horse.

The Evening World, June 4, described a man galloping through the valley, warning residents that the dam was failing and urging them to flee to higher ground.

According to the report, some residents ignored him because they had heard similar warnings before. The rider continued ahead of the approaching flood until both horse and rider were overwhelmed.

The Morning Appeal, June 5, published a strikingly similar narrative under the heading A Nameless Paul Revere.

Both versions share the same essential details: an unidentified man, a large bay horse, warnings shouted through the valley, residents who initially failed to respond and a final encounter with the flood near a railway bridge.

Do these two reports corroborate the horseman's existence?

Not necessarily.

Their similarity could indicate that the story circulated through a shared news report or was republished by different newspapers. Two publications carrying essentially the same narrative do not automatically represent two independent witnesses.

The horseman's identity also remains absent from both accounts.

What we can say is that, within days of the disaster, newspapers were circulating a vivid story of a heroic but unidentified messenger. Whether every detail was accurate is a separate question.

A documented warning effort: John G. Parke

The horseman accounts become especially interesting when compared with reporting about engineer John G. Parke.

The Evening World article separately describes Parke recognizing the danger at the dam, helping organize attempts to release water, riding to warn people at South Fork and arranging for messages to be sent toward Johnstown.

The Indianapolis Journal, June 8, also recounts Parke's efforts and reproduces his reported explanation of events.

These accounts establish that warnings were part of the contemporary reporting. They do not establish that the unnamed horseman and Parke were the same person; the Evening World actually presents them as separate figures.

The distinction is important. There is a documented history of attempted warnings, alongside a dramatic anonymous rider story whose origins deserve further investigation.

June 3–July 2: The death toll becomes a moving target

One of the clearest patterns across the newspapers is the instability of casualty estimates.

Newspaper date	Publication	Reported estimate
June 1	Evening Capital	1,500 deaths reported by a courier
June 3	Los Angeles Daily Herald	Approximately 5,000 in one passage, with higher figures elsewhere
June 5	Great Falls Tribune editorial	Possibly 10,000
June 14	Telegram-Herald	Between 3,000 and 4,000
July 2	Daily Kennebec Journal	Conflicting estimates of roughly 3,000, 5,000 and nearly 10,000

These are contemporary claims, not confirmed casualty counts.

The Great Falls Tribune editorial of June 5 suggested that the loss might approach 10,000 people.

By contrast, the Telegram-Herald, June 14, reported a prevailing estimate of 3,000–4,000.

Crucially, it offered an explanation for the discrepancy: early estimates had assumed a substantially larger population in Johnstown and its neighboring boroughs than later investigators believed was accurate. Some people initially counted as missing may never have been residents of the affected communities.

The newspaper also reported that an existing local directory could help establish a more reliable list of residents and victims.

Yet uncertainty persisted.

The Daily Kennebec Journal, July 2, acknowledged that earlier estimates were too high while still reproducing conflicting figures. Its own discussion did not settle on a consistent total.

The story is therefore not a simple sequence of steadily improving numbers. It is a record of different authorities, incomplete population information and inconsistent reporting.

The official figure of 2,209 emerged later. Reading the original newspapers makes clear why a reliable count was so difficult to establish.

June 3–July 19: Natural catastrophe or human negligence?

Perhaps the most revealing development is the argument over responsibility.

The newspapers did not speak with one voice.

June 3: The force of nature

The Los Angeles Daily Herald, June 3, published an extended editorial describing the flood as part of an extraordinary meteorological catastrophe.

The writer emphasized the intensity of the storm, the mountainous terrain and the immense forces unleashed by the water.

In this interpretation, the dam's destruction was chiefly the consequence of an exceptional natural event. The editorial suggested that it was difficult to blame the dam's constructors for failing to withstand such forces.

The piece also devoted considerable attention to Johnstown's industrial population. It portrayed the victims as hardworking people subjected to an overwhelming disaster.

Sympathy for the victims did not necessarily imply blame toward the dam's owners.

June 8: Doubts about assigning responsibility

Five days later, the Indianapolis Journal examined the responsibility question directly.

The editorial cited reports that engineers had previously examined the dam and judged it safe. It also reproduced Parke's reported view that the catastrophe could not have been prevented.

The paper was reluctant to conclude that the owners bore direct responsibility.

But its argument contained a significant qualification.

It questioned the justification for maintaining a potentially dangerous reservoir for private pleasure once its original public purpose had disappeared.

The editorial's conclusion was forward-looking: other dangerous recreational dams should be eliminated rather than allowed to threaten downstream communities.

This was an intermediate position. The writer did not confidently assign legal blame for Johnstown, but nevertheless saw the disaster as a warning about the risks of private reservoirs.

July 19: A much harsher accusation

By July, a very different argument appeared in Colorado's State Herald.

The editorial, attributed at its end to the Chicago Herald, accused the dam's owners and local officials of culpable conduct. It argued that those responsible for maintaining the reservoir should face criminal consequences.

The writer criticized the dam's construction, its maintenance and the alleged failure to respond to warnings.

The language was severe, presenting the disaster as an example of wealthy people's disregard for working-class lives. It also connected Johnstown to wider criticism of Pennsylvania's coal-mine owners and industrial inequalities.

That connection is historically revealing. For this writer, the flood was not simply an engineering failure. It represented a broader imbalance of power between wealthy property owners and industrial workers.

There is an important limitation, however: this is an editorial opinion, and one apparently republished from another newspaper. Its allegations are evidence of contemporary accusations, not proof that every allegation was true.

Taken together, these three editorials show a spectrum of interpretations:

• A catastrophe overwhelmingly attributed to extraordinary natural forces.
• A cautious refusal to assign blame, combined with concern about dangerous private dams.
• An explicit accusation that the owners' negligence had made the disaster possible.

The contrast is compelling, but we should not claim that every newspaper's opinion changed in the same direction. These are different publications, writing from different perspectives.

June: A national relief effort takes shape

Even as arguments over the disaster's cause continued, relief became an immediate national concern.

The Great Falls Tribune, June 5, praised Pittsburgh's initial fundraising and urged additional contributions.

The needs on the ground were enormous.

On June 8, the Indianapolis Journal described residents of smaller valley settlements lacking food and clothing.

Its report placed roughly a thousand people around a railway depot commissary seeking rations. Soldiers helped control the crowd while relief workers distributed supplies.

The article also recorded the organization of a Masonic relief committee.

Relief was therefore not a single centrally directed operation. Local committees, religious organizations, fraternal groups, private donors and state authorities all played roles.

By June 14, the Telegram-Herald reported substantial reconstruction activity, including workers clearing debris, railway repairs and a state commission established to distribute donated funds.

The article described 1,500 men beginning work on the ruins under state supervision, as well as tensions between workers, contractors and militia.

It also reported that Johnstown's surviving business community was considering how to rebuild. Pittsburgh merchants were offering credit to help local businesses reopen.

The immediate emergency was gradually becoming a long-term recovery effort.

July: Relief becomes an administrative challenge

The Wood County Reporter, July 25, published a detailed account of the Flood Relief Commission's work.

According to its report, approximately $1.7 million had been expended on relief in the Conemaugh Valley and surrounding area.

The number of people receiving food had fallen from a maximum of 31,950 to a commissary roll of about 7,000.

The commission also faced nearly 4,000 distinct claims.

Its guiding principle was that donated money should assist people according to need, rather than reimburse every person for their financial losses.

That distinction created difficult decisions. A wealthy property owner might have lost a large amount, but a poorer family could have lost everything necessary for survival.

The report reveals a transition from emergency rescue to organized social assistance, with administrators deciding how charitable funds should be distributed.

July 30: The search for legal accountability

By the end of July, questions about responsibility were reaching the courts.

The Hopkinsville Kentuckian, July 30, reported that the widow of John A. Little, who had drowned in Johnstown's Hurlburt Hotel, had brought a lawsuit seeking $50,000 in damages against the South Fork company.

The newspaper characterized it as a test case that might lead to further litigation.

It also reported that a coroner's jury had found the dam's owners responsible.

This is a significant final point for the initial investigation. Within two months of the flood, the question had progressed from newspaper speculation to formal legal action.

But a lawsuit is not a judgment.

Later historical accounts establish that the South Fork Fishing and Hunting Club was never successfully held legally liable for the flood, despite the strong public criticism it faced.

The difference between public blame and legal accountability is one of the defining themes of the disaster.

---

Research note: This account draws primarily on the full texts of 20 newspaper components retrieved through SNEWPapers, published between June 1 and July 30, 1889. Newspaper claims are attributed to their contemporary sources, and the distinction between reported estimates and subsequently established facts is maintained throughout. The selected articles are a research sample, not an exhaustive survey of all coverage. Some accounts may be syndicated or republished rather than independently reported.

What the newspapers reveal

The Johnstown Flood is often summarized as a terrible storm, a failed dam and thousands of deaths. Those facts matter, but the newspapers reveal a richer history.

The first dispatches show journalists trying to report an enormous catastrophe through damaged communications. Later articles reveal the difficulty of counting the dead, the spread of heroic stories, and the tension between the need for immediate relief and the demand to establish responsibility.

The editorials expose competing interpretations of the same disaster: overwhelming natural forces, preventable engineering danger and wealthy indifference toward the lives of industrial workers.

Meanwhile, the relief reports document the less dramatic but essential work of feeding survivors, restoring transport, administering donations and rebuilding a devastated community.

Finally, the July lawsuit demonstrates that arguments over responsibility did not end with newspaper commentary.

Perhaps the most important lesson is that historical events do not arrive in newspapers as finished narratives. They emerge through fragments, competing claims, eyewitness accounts, corrections, arguments and the gradual accumulation of evidence.

Following those fragments across publications and dates is what turns newspaper search into historical investigation.

References

All 20 summaries ChatGPT returned. The SNEWPapers links below provide the full transcription and metadata, as well as links to the original source material at the Library of Congress and to PDFs of the original pages: