Creatos Logo
Compare
Buy License

Comparison · tested on

NotebookLM alternative: Gemini Notebook, ChatGPT and Claude on the same PDF

NotebookLM is now called Gemini Notebook. Google renamed it on 16 July 2026 and says it is the same standalone research tool, now also inside the Gemini app and Google Search. The usual alternatives are the projects in ChatGPT and Claude, which also keep your files and answer questions about them. We gave all three the same 39-page jobs report and the same request, then checked their answers against the PDF. Two of the three got the same number wrong. We make Creatos, a desktop canvas that puts several models' answers to one source side by side, so it is here too, with what it doesn't do.

Short answer

  • Stay with Gemini Notebook if you want it free, want citations that jump to the exact passage, or want Audio Overviews, mind maps, flashcards and quizzes from your sources. ChatGPT and Claude showed no citations in our run.
  • Use ChatGPT Projects if you already pay for ChatGPT. It was the only one of the three to report the unemployment change correctly, and it named the pages it used.
  • Use Claude Projects if you already pay for Claude and your sources are large: on paid plans, Claude switches to searching project knowledge once it nears the context limit, for up to 10 times the capacity.
  • Use Creatos if you want to see where models disagree before you trust a number: one PDF on a canvas, Claude, GPT and Gemini on your own keys or local models, answers side by side. It costs $69 once.

Watch the side-by-side test

The same 39-page jobs report and request in NotebookLM (now Gemini Notebook), ChatGPT and Claude, every number checked against the PDF: which two got unemployment wrong, the citations, and what each costs. It plays from YouTube when you press play. Watch on YouTube

One report, one request, three apps

The document was the US Bureau of Labor Statistics' The Employment Situation, September 2026, a 39-page PDF of text and tables. On 14 October 2026 we added it to each app and sent this request once, with no follow-ups:

Summarize this report in 10 bullet points for a reader who won’t open it. Include the key numbers.
AppWhere the PDF wentPlan and model shown
Gemini NotebookA new notebook, through Upload filesNo plan or model name anywhere in the app
ChatGPTA new project, under SourcesPlus; the model picker said “ChatGPT”
ClaudeAttached to a new chat (see below)Max; Claude Opus 5.5

Each app answered in under 50 seconds. Claude ran as an ordinary chat because, in our account, New project on claude.ai created a Claude Code project instead: Anthropic says it is rolling out a new version of projects in stages, starting with Claude Code. For one document and one request the input is the same, but strictly this was not a Projects run.

What each one got right and wrong

Gemini NotebookChatGPTClaude
Numbers checked424360
Wrong301
Unemployment rate“held steady at 4.2%”: wrong“rose slightly to 4.2% … from 4.1%”: right“held steady at 4.2%”: wrong
Labor force participation“remained at 61.8%”, and the employment ratio “held at 59.2%”: wrong61.8%, “little changed”: right“rose 0.2 points to 61.8%”: right
Payrolls and revisions+29,000; July and August down 60,000: right+29,000; down 60,000: right+29,000; down 60,000: right
CitationsNumbered chips on each bulletOne line naming pages 1–4 and Summary Table ANone
ExtrasOffered a “visualization or table” of the industry changesA bottom line and a “numbers at a glance” blockThe date of the next report, 6 November

All four mistakes come from the same place. The report's own text says the unemployment rate, the participation rate and the employment-population ratio all "changed little in September", which is statistical language for a change too small to be sure of. Its tables show unemployment at 4.1% in August and 4.2% in September, participation at 61.6% and 61.8%, and the employment ratio at 59.1% and 59.2%. Gemini Notebook and Claude turned "changed little" into "held steady", which tells a reader nothing moved. All three repeated the report's line that the rate has stayed between 4.1% and 4.3% since March; only ChatGPT said which way it moved in September.

We checked every number in the three answers against the PDF. Everything else was right in all three: payrolls, revisions, wages, hours, the industry figures and the rates by group. The one figure not in the report, Claude’s three-month average of 51,000, is correct arithmetic from the revised months. Claude's answer was the most detailed, with sub-bullets on slack and weak industries and the report's own ±122,000 margin of error. Gemini Notebook's citations are the one thing the others can't match: Google says you can select a citation to jump to the quote in your source.

This was one run of one document, so it shows how these apps slip, not which one always wins. If a number matters, check it: our free source-check skill gives every claim in an AI answer a verdict and a quote from the PDF.

The three answers, word for word

Each answer is exactly what the app's Copy button gave us on 14 October 2026, formatting included. Copy leaves out Gemini Notebook's citation chips, so they are missing here. The wrong lines are the unemployment rate in Gemini Notebook's and Claude's answers, and participation and the employment-population ratio in Gemini Notebook's.

Gemini NotebookA new notebook, PDF added through Upload files

Here is a 10-bullet summary of the September 2026 Employment Situation report:

  • Nonfarm Payroll Growth: Total nonfarm payroll employment was little changed in September 2026, adding 29,000 jobs, below the average monthly gain of 45,000 over the prior 12 months.
  • Unemployment Rate: The official unemployment rate held steady at 4.2%, remaining within a narrow range of 4.1% to 4.3% since March 2026, with 7.1 million people unemployed.
  • Downward Job Revisions: Payroll growth for prior months was revised down by a combined 60,000 jobs — July 2026 was revised down by 31,000 (from +21,000 to -10,000) and August 2026 was revised down by 29,000 (from +162,000 to +133,000).
  • Hourly Wages: Average hourly earnings for private nonfarm employees rose 5 cents (0.1%) to $37.81, marking a 3.0% increase over the last 12 months.
  • Workweek Length: The average workweek across all private nonfarm employees remained unchanged at 34.4 hours, while the manufacturing workweek held steady at 40.6 hours with 3.0 hours of overtime.
  • Industry Performance: Health care led job growth with +17,000 jobs (slower than its 12-month average of +33,000), construction gained +11,000, and manufacturing added +9,000, while financial activities fell by -7,000 and government employment dropped by -17,000.
  • Demographic Trends: The unemployment rate for Black workers increased to 7.0%, while rates for White (3.6%), Asian (2.9%), Hispanic (4.7%), adult men (3.9%), and adult women (3.6%) showed little change over the month.
  • Labor Force Participation: The labor force participation rate remained at 61.8% and the employment-population ratio held at 59.2%, showing minimal net change since January 2026.
  • Involuntary Part-Time & Underemployment: The number of individuals working part-time for economic reasons was 4.5 million, while the broader U-6 labor underutilization rate stood at 7.6%.
  • Unemployment Duration & Multiple Jobs: The long-term unemployed (jobless 27+ weeks) held at 1.9 million (27.1% of all unemployed people), discouraged workers numbered 414,000, and multiple jobholders totaled 9.0 million (5.5% of total employed).

📊 Would you like a visualization or table comparing industry-by-industry job changes across July, August, and September?

ChatGPTA new project, PDF under Sources; Plus, picker “ChatGPT”
U.S. Jobs Report — September 2026: 10 Key Takeaways

Bureau of Labor Statistics (BLS) · Released October 2, 2026

Bottom line: The U.S. labor market remained weak in September. Hiring was sluggish, unemployment edged higher, wage growth slowed, and previous job gains were revised downward. However, the report does not show a sharp, economy-wide collapse in employment.

1. Only 29,000 jobs were added in September. Nonfarm payrolls increased by just 29,000, below the already modest 12-month average of 45,000 jobs per month. This indicates persistently weak hiring.

2. Unemployment rose slightly to 4.2%. The rate increased from 4.1% in August, with approximately 7.1 million Americans unemployed. Since March, unemployment has stayed between 4.1% and 4.3%.

3. Previous job numbers were revised down by 60,000. July payroll growth was revised from +21,000 to −10,000, while August was reduced from +162,000 to +133,000. July therefore experienced a net loss of jobs rather than a gain.

4. Healthcare remained a key source of job growth. The sector added 17,000 jobs, including 13,000 in outpatient healthcare and 12,000 in hospitals. Nursing and residential care facilities lost 9,000 jobs. Healthcare hiring was slower than its 12-month average of 33,000.

5. Construction and manufacturing showed modest gains. Construction added 11,000 jobs, while manufacturing added 9,000. Manufacturing employment has recovered by 72,000 jobs since its December 2025 low, but neither industry showed strong September growth.

6. Financial-sector employment continued to struggle. Financial activities lost 7,000 jobs in September and have shed 129,000 jobs since May 2025. Insurance carriers and related businesses accounted for 90,000 of those losses.

7. Wage growth was unusually slow in September. Average hourly earnings increased just 0.1% ($0.05) to $37.81. Annual wage growth was 3.0%. Production and nonsupervisory workers earned an average of $32.60 per hour, up 0.2% for the month.

8. Labor force participation remained low. The participation rate was 61.8%, and the employment-to-population ratio was 59.2%. Both were little changed in September and have shown limited net improvement since January.

9. Long-term unemployment and underemployment remained significant. Approximately 1.9 million people had been unemployed for at least 27 weeks, representing 27.1% of all unemployed workers. Another 4.5 million were working part-time for economic reasons, while 5.8 million people outside the labor force wanted a job.

10. Unemployment disparities widened, particularly for Black workers. Black unemployment increased from 6.0% to 7.0%. Other September rates included White workers at 3.6%, Hispanic workers at 4.7%, Asian workers at 2.9%, and teenagers at 14.5%. Most other major demographic groups saw little monthly change.

The numbers at a glance

September jobs

+29K

Weak hiring

Unemployment

4.2%

+0.1 pp monthly

Annual wage growth

3.0%

Nominal increase

Overall assessment: The strongest signal is not mass layoffs but a labor market with very little hiring momentum. The combination of 29,000 new jobs, 60,000 in downward revisions, and just 0.1% monthly wage growth suggests a cooling economy. Healthcare remains relatively resilient, but employment growth is too limited to indicate a broad-based recovery.

Source: BLS, The Employment Situation — September 2026, pages 1–4, including Summary Table A.

ClaudeA new chat with the PDF attached; Max, Claude Opus 5.5
  • Hiring nearly stalled. Nonfarm payrolls rose 29,000 in September, below the 45,000 monthly average of the previous 12 months. No major industry changed much. BLS considers a monthly change significant only above about ±122,000, so this is effectively flat.
  • Earlier months were revised down. July went from +21,000 to −10,000, and August from +162,000 to +133,000. Together that's 60,000 fewer jobs than first reported. The three-month average is now 51,000 per month.
  • Unemployment held steady at 4.2%. About 7.1 million people were unemployed. The rate has stayed between 4.1% and 4.3% since March and is down from 4.4% a year ago.
  • Black unemployment jumped. The rate for Black workers rose a full point to 7.0% (from 6.0%), the only major group with a notable increase. Other rates were adult men 3.9%, adult women 3.6%, teenagers 14.5%, White 3.6%, Asian 2.9% and Hispanic 4.7%.
  • Participation edged up but is still below last year. The labor force participation rate rose 0.2 points to 61.8%, and the employment-population ratio was 59.2%. A year earlier they were 62.5% and 59.7%.
  • Broader slack was mostly unchanged.
    • About 1.9 million people had been jobless for 27 weeks or more, 27.1% of all unemployed.
    • 4.5 million people worked part time because they couldn't get full-time work.
    • The broadest underemployment measure (U-6) was 7.6%.
    • People who want a job but recently stopped looking fell 236,000 to 1.5 million.
  • Health care still led, but more slowly. It added 17,000 jobs, about half its 12-month average of 33,000. Outpatient care added 13,000 and hospitals 12,000, while nursing and residential care lost 9,000.
  • Goods producers posted small gains. Construction added 11,000, mostly nonresidential specialty trade contractors (+12,000). Manufacturing added 9,000 and is up 72,000 since its December 2025 low, led by plastics and rubber and by machinery (+5,000 each).
  • There were several weak spots.
    • Government lost 17,000 jobs, temporary help 10,900, information 10,000 and financial activities 7,000.
    • Financial activities is down 129,000 since its May 2025 peak, mostly in insurance (−90,000).
    • Slightly more private industries cut jobs than added them.
  • Wage growth slowed slightly. Average hourly earnings rose 5 cents (0.1%) to $37.81 and are up 3.0% over the year. Production and nonsupervisory workers gained 7 cents to $32.60. The average workweek held at 34.4 hours, and manufacturing stayed at 40.6 hours with 3.0 hours of overtime.

The October report is scheduled for Friday, November 6.

Plans, limits and prices

Gemini NotebookChatGPT ProjectsClaude Projects
Free plan100 notebooks, 50 sources each5 files per projectUp to 5 projects
Paid plans (US)Google AI Pro $19.99 a month: 500 notebooks, 300 sources eachGo $8 and Plus $20 a month: 25 files; Pro: 40 filesPro $20 a month; Max $100 or $200
Size of one source500,000 words or 200 MB512 MB, and 2 million tokens30 MB per file
Usage limitsCompute-based, refreshed every 5 hours up to a weekly limit; Pro gets 4 times the free amountBy planBy plan; large projects switch to search on paid plans

Gemini Notebook's figures come from Google's plans table, which it labels "subject to change", its sources page and its usage limits page; Google publishes no fixed number of chats. Audio Overviews work in more than 80 languages. The Pro price is from Google's blog. ChatGPT's file counts are on its projects page, the size limit on its file uploads page, and the prices on its Go and Plus pages. Claude's come from its projects, project search, file limits and Pro plan pages.

In our notebook, Gemini Notebook's Studio panel offered Audio and Video Overviews, a slide deck, a mind map, reports, flashcards, a quiz, an infographic and a data table, all made from your sources. ChatGPT projects can use your default memory or keep a memory of their own. On paid plans, Claude switches a project to searching its files once they near the context limit.

Where Creatos fits, and where it doesn’t

Creatos is not a notebook app. It is a desktop canvas: you put a source on it, such as a PDF, a YouTube video or a web page, and connect it to several AI nodes, each running a different model. You ask once and read the answers next to each other, so a disagreement like 4.1% against "held steady" shows up before you repeat it.

  • Models. You add keys for the providers you use: Anthropic, OpenAI, Google and ten others, any OpenAI-compatible endpoint, or Ollama for models on your own computer. You pay the provider for what you use, or nothing for local models. In our API test on this same report, three questions cost $0.22 on Gemini, $0.44 on Claude and $0.51 on GPT.
  • Price. $69 once, with no subscription and a full refund within 14 days if you ask. A new download gets a 7-day free trial with $0.50 of free usage and no API key.
  • Where files live. On your computer. When you use a cloud model, the text it needs goes to that provider under your key; with Ollama, nothing leaves the machine.

What you give up compared with Gemini Notebook:

  • Audio and Video Overviews, mind maps, flashcards, quizzes and the other Studio outputs. Creatos has none of them.
  • A free plan. Creatos has no free tier beyond its 7-day trial.
  • The web, phones, sync and sharing. Creatos runs on macOS 12 or later on Apple Silicon and 64-bit Windows 10 or later, with no sync and no sharing.

FAQ

What happened to NotebookLM?

Google renamed it Gemini Notebook on 16 July 2026. Google says it is the same standalone research tool, now also working inside the Gemini app and Google Search. It also added a secure cloud computer to notebooks that can write and run code, first for Google AI Ultra and business customers. The app is at notebook.google.com.

Is Gemini Notebook still free?

Yes. The free plan allows 100 notebooks with up to 50 sources each, and each source can hold up to 500,000 words or 200 MB. Usage is compute-based: it refreshes every 5 hours up to a weekly limit. Google AI Pro, $19.99 a month in the US, raises that to 500 notebooks, 300 sources each and four times the usage.

Can ChatGPT or Claude replace NotebookLM?

For asking questions about your files, yes: both have projects that keep files attached to every chat. ChatGPT allows 5 files per project on Free, 25 on Go and Plus, and 40 on Pro. Claude has projects on every plan, with up to 5 projects on Free and 30 MB per file. Neither showed citations to the passage in our run.

Which one got the numbers right?

In our run on the September 2026 jobs report, only ChatGPT reported the unemployment rate’s change correctly, 4.1% in August to 4.2% in September. Gemini Notebook and Claude both said it held steady at 4.2%, and Gemini Notebook also said labor force participation and the employment-population ratio stayed put when both rose slightly. That is one run of one document, so it shows how they slip, not which one always wins.

Does Creatos make audio overviews, flashcards or quizzes?

No. Creatos is a desktop canvas where one source, such as a PDF, connects to several AI models so you can read their answers side by side. If you want Audio Overviews, mind maps, flashcards or quizzes from your sources, Gemini Notebook is the better tool.

Does Creatos have a free trial?

Yes. A new download gets a 7-day free trial and can run a real comparison of Claude, GPT and Gemini right away, with $0.50 of free usage and no API key. After that, Creatos costs $69 once, and activating the license works as before. Any purchase can be refunded in full within 14 days: email hello@creatos.io, no questions asked. With the license, AI use is billed by the provider whose key you add, or costs nothing with local models through Ollama.

Who wrote this

We make Creatos, so read this with that in mind. We ran the three apps ourselves on 14 October 2026 with their default settings and checked each answer against the BLS report. Every plan, limit and price comes from Google's, OpenAI's and Anthropic's own pages, read the same day and linked above. Google's rename announcement is from 16 July 2026.