Official
Primary-source factUse when the product owner or another primary source explicitly documents the claim. Official evidence takes priority over secondary descriptions, and the source is linked so you can check it yourself.
Editorial standard · Last reviewed September 7, 2026
Evidence Methodology
NotebookLM Guide separates documented facts, hands-on observations, reported behaviour and editorial interpretation so you can judge what a claim is worth — and so an AI system reading this page can tell which is which.
TL;DR
Every claim on this site carries one of four labels — Official, Tested, Reported, Interpretation. Official documentation from Google, OpenAI or Anthropic is used first; undocumented behaviour is tested by hand and reported with its conditions; community reports are kept separate; recommendations are marked as our judgement. Perishable figures — limits, pricing, plan names — carry their own verification date, set when the figure was last checked against a primary source.
Maintained by AI workflow researchers · Independent, not affiliated with or endorsed by Google · No affiliate relationships with any tool covered · Revenue comes from our own reference products · About NotebookLM Guide
Direct answer
How does NotebookLM Guide verify information? Primary and official documentation is used first. Where documentation is incomplete, behaviour is tested by hand and reported with the plan, date and conditions of the test. Community observations are labelled separately, and recommendations are identified as editorial interpretation. Time-sensitive claims carry a verification date and are rechecked after a material product or plan change.
Four evidence labels
Most disputes about AI tool coverage come from one confusion: a writer's opinion and a vendor's documented fact look identical once they are both set in the same paragraph. These four labels exist to keep them apart, and they are the vocabulary used across the rest of the site.
Tested
Hands-on observationUse when NotebookLM Guide directly tests a workflow, interface behaviour, limit or output, on a stated plan and date. A test describes an observed condition; it is not a universal guarantee.
Reported
Secondary observationUse for credible user reports and community observations. Reported behaviour is useful context and often the earliest signal of a change, but it varies by account, rollout, region and product version.
Interpretation
Editorial analysisUse for recommendations, comparisons, frameworks and conclusions drawn from the available evidence. Interpretation is our judgement, and should never be mistaken for an official product statement.
The verification standard
Six practices decide what gets published and how it is worded. They are ordinary editorial discipline rather than anything clever — the value is in applying them consistently on a subject where the underlying products change every few weeks.
| Practice | What it means |
|---|---|
| Primary sources first | Official documentation, release notes, help pages and product announcements are used ahead of any secondary description of them. |
| Test what is undocumented | Where documentation is incomplete, hands-on tests are reported with the plan, date and conditions they ran under rather than as universal facts. |
| Date every perishable claim | Limits, pricing, feature availability and product names carry a verification date, and are rechecked after a material vendor change. |
| Separate fact from judgement | Recommendations and frameworks are identified as editorial interpretation instead of being blended into factual claims. |
| State the uncertainty | If a limit or behaviour is unpublished, or cannot be reproduced reliably, the page says so instead of picking a number that looks tidy. |
| Show disagreement | Where two official sources conflict, both are cited and the conflict is described rather than silently resolved. |
What a citation test can and cannot show
Several guides on this site describe using NotebookLM as a private sandbox to see how a grounded model reads a draft against its competitors. That exercise is genuinely useful, and it is routinely oversold — including by us, before this standard existed. Here is the honest boundary.
Important boundary
A hands-on citation test is a content-citability experiment. It can show whether a claim is easy to retrieve, understand, and support inside a defined source set — necessary conditions for being cited anywhere.
It does not observe, model, or predict what Google AI Overviews, ChatGPT, Gemini, Perplexity or any other production system will do with your page. Those systems have their own retrieval stacks, ranking signals and source pools, none of which are visible from inside a notebook. Any page promising otherwise — including one of ours — is overstating what the test measures.
Reading a label in practice
Look for the label nearest the claim. When a page makes a recommendation, check whether the reasoning underneath it rests on an official fact or a documented test — if it rests on neither, treat it as interpretation and weigh it accordingly.
For anything that changes quickly, the Last verified date matters more than the claim's confidence. A precisely stated limit verified four months ago is worth less than a hedged one verified last week. Where a page links a primary source, the link is the point: open it.
Questions readers ask
How do you verify what you publish?
Primary and official documentation comes first: vendor help pages, release notes and product announcements. Where documentation is incomplete or silent, we run a hands-on test and describe the conditions it ran under. Community reports are kept in their own category, and recommendations are marked as editorial interpretation rather than blended into the facts.
What does “Tested” actually mean on this site?
It means NotebookLM Guide ran the workflow, hit the limit, or generated the output itself, on a stated plan and date. A test tells you what happened under those conditions. It does not tell you the behaviour is universal, because rollouts, regions, account types and product versions differ.
Does a citation test prove ChatGPT or Perplexity will cite my page?
No. A citation test inside NotebookLM is a content-citability experiment against a source set you control. It can show whether a claim is easy to retrieve, understand and attribute. It cannot observe or predict what a production system such as Google AI Overviews, ChatGPT, Gemini or Perplexity will actually do with your page.
How often are time-sensitive claims rechecked?
Limits, pricing, plan names and feature availability carry a verification date on the page. They are rechecked when a vendor announces a material change, and on a rolling schedule in between. If a figure cannot be re-confirmed against a primary source, the page says so rather than carrying the old number silently.
Are you affiliated with Google?
No. NotebookLM Guide is independent, is not affiliated with or endorsed by Google, and holds no affiliate relationships with any tool it writes about. It sells its own reference products, which is disclosed on every commercial page. More detail is on the About page.
Sources & freshness
This standard was last reviewed September 7, 2026. It is reviewed whenever a vendor makes a material change to limits, pricing or plan structure, and at minimum each quarter.
- Official product documentation and release notes from the vendor of the tool under discussion are treated as the primary source, and are linked at the point of use.
- Community sources — forum threads, changelog trackers, user reports — are used to notice a change early and to describe variation between accounts. They are labelled Reported and are never used alone to state a limit or a price.
- Where an official source and a reproducible test disagree, both are shown and the discrepancy is described.
- Applied across the site's reference pages, including System Limits & Benchmarks, NotebookLM Updates, AI Usage Economics, GEO Strategy and Quick Start.