Statista is a reliable pointer, not a primary source

Statista is reliable in the narrow sense that matters most: when it names an outside source, the number it displays usually matches that source. We traced 13 publicly visible Statista statistics to the organisations Statista credits, and 8 matched the original exactly or to rounding, though one of those matched its named source rather than the official national series. The other 5 are the useful part. One changed the population the survey measured. Three could not be fully checked against a reachable original. One is Statista’s own model with no external source at all. Use Statista to find a number and its origin, then cite the origin.

Statista is an aggregator with its own forecasts on top

Statista describes itself as a data company with statistics “on over 80,000 topics from 50,000+ sources in 170 industries” (About Statista). Most statistic pages re-chart a figure published by someone else: a statistics office, an international body, a company filing, a survey firm. Its research commitment page says content passes “a tested multi-stage peer-review process prior to publication”. That is an internal editorial review. It is not academic peer review, and Statista is not a peer-reviewed publisher.

Statista also produces figures of its own. Market Insights pages carry market sizes and forecasts with “Source: Statista Market Insights”, built by a “combined top-down and bottom-up approach” from company reports, third-party studies, Statista’s own consumer survey and macro indicators (eCommerce worldwide, methodology section). These are modelled estimates. They can be well built, but no outside document states them, so nobody can trace them.

Statista uses AI in its product. Its Research AI answers questions from Statista content, and the page itself warns: “Our AI can make mistakes. Please check important information.” The site navigation also lists an API and MCP integration for enterprise AI tools.

Thirteen Statista figures traced to their originals

Method: we opened each Statista page logged out and unpaid on 2026-09-23 and recorded the figure as displayed and the source Statista names (from the page’s source field or its citation formats, which list the originating organisation as author). Then we fetched the original and compared. The sample covers government, international-body, company-reported, survey and Statista-modelled figures. It is 13 figures chosen for variety, not a random sample and not a study of Statista’s catalogue.

#	Statista figure as displayed	Kind	Source Statista names	Original reached	Result
1	Internet users worldwide: 6,000 million in 2025; 5,500 million in 2024	International body	ITU	ITU press release, 17 Nov 2025: “6 billion” in 2025, “up from a revised estimate of 5.8 billion in 2024”	2025 matches. The 2024 row is ITU’s earlier 5.5 billion, since revised
2	US unemployment rate: 4.2% in 2025	Government, via international body	World Bank	World Bank API: 4.198, a “modeled ILO estimate”	Matches the named source. The official BLS annual average is 4.3%, an 11-month average because October 2025 was not collected (BLS API, LNU04000000)
3	US resident population: about 342.6 million, July 2026, “projected”	Government	U.S. Census Bureau	Census NA-EST2025-POP: 342,620,143 for July 1, 2026	Matches. It is a Census short-term projection, not a count
4	US GDP, current prices: 30.77 trillion USD in 2025	International body	IMF	IMF DataMapper: 30,767.1 billion	Matches. BEA’s own figure is 30,762.1 billion (FRED GDPA). Rows for 2026 to 2031 are IMF projections
5	Average inflation rate worldwide: 4.13% in 2025	International body	IMF	IMF DataMapper: 4.1	Matches, rounded
6	Facebook monthly active users: 3,065 million, Q4 2023	Company-reported	Meta Platforms	Meta Q4 2023 release, SEC Exhibit 99.1: “3.07 billion”	Matches, rounded. The series ends at Q4 2023, so it is not a current figure
7	Tesla deliveries: “around 480 thousand” in Q2 2026	Company-reported	Tesla	Tesla Q2 2026 release, SEC Exhibit 99.1: 480,126	Matches, rounded. Tesla’s investor site refused our fetcher; the SEC copy is the same release
8	Netflix paid subscribers: “surpassing *** million” in Q4 2025	Company-reported	Netflix	Netflix Q4 2025 letter: “crossed the 325M paid memberships milestone”	Statista’s value is paywalled. Netflix stopped regular quarterly membership reporting from Q1 2025 (Q4 2024 letter), so check what fills 2025 in any quarterly series
9	Social media use: YouTube 81%, Facebook 69% of “internet users”, 2021	Survey	Pew Research Center	Pew, Social Media Use in 2021: 81% and 69% of U.S. adults, 1,502 surveyed by phone	Numbers match; the population differs. Pew measured all adults, not internet users. Pew has newer data with a survey-mode change
10	Smartphone subscriptions: “almost 7.3 billion” in 2025, over 7.9 billion by 2028	Company forecast	Ericsson	Ericsson’s Mobility Visualizer is interactive and exposed no values to our fetch	Original unreachable. Ericsson’s current outlook now runs to 2031; the Statista page was released June 2025
11	Global CO2 emissions: 38.11 billion metric tons in 2025 (projection)	Research consortium	Statista and Global Carbon Budget	Global Carbon Budget, 13 Nov 2025: “38.1 billion tonnes of fossil carbon dioxide”	Matches. Statista converted from carbon to CO2 (x3.664) and the figure is fossil only, which the notes say and the title does not
12	Retail e-commerce sales worldwide: 2025 value paywalled	Third-party forecast	Ecommerce Germany News	The named source is a news outlet; the public page does not name who produced the forecast	Original unreachable from the public page
13	eCommerce market worldwide: US$3.86tn revenue in 2026	Statista model	Statista Market Insights	None exists outside Statista	Statista’s own estimate, no external source. Method is published; the inputs are not

What the traces say about Statista’s reliability

Statista’s transcription is accurate. Across the figures we could check, every number Statista attributed to an outside body appeared in that body’s publication, sometimes rounded. The problems sit around the number, not in it.

• The source is not always the source you would pick. Row 2 is correct, but it is the ILO’s modelled rate for the US, not the BLS rate a US reader expects. Row 4 is the IMF’s number, which happens to agree with BEA’s.
• Scope drifts in the title. Row 9 turns “U.S. adults” into “internet users”. Row 11 drops “fossil” from the title. Both are visible only when you read the original.
• Vintages mix. Row 1 pairs ITU’s newest 2025 estimate with a 2024 value ITU has since revised upward by 300 million.
• Some figures cannot be traced. Rows 10 and 12 lead to a closed interactive tool and a news outlet. Row 13 leads nowhere outside Statista, by design.

Cite the original, not the aggregator

The working rule for reports, papers and decks is to cite the organisation that produced the number, at the document where it published it, and use Statista to get there. Statista’s own citation formats put the originator first (“ITU. (November 14, 2025)…”), which points the same way.

To find the original from a Statista page, read the source name from the “Source” panel or the citation formats. On the older page layout the detailed source panel asks you to register for a free account; the originator’s name still appears in the citation formats. Then search that organisation’s site for the release date and the exact title, and check the unit, the population, the year and whether the value is an estimate or a projection. Rows 2, 9 and 11 show that each of those can differ.

Citing Statista itself is reasonable when the figure is a Statista Market Insights estimate or a Statista survey, because Statista is then the producer. Say so in the sentence (“Statista estimates…”), not only in the footnote. In academic work, most style guides and instructors expect the primary source; a Statista page is a finding aid.

Pitch decks and client decks need the primary source

A deck slide is often read without the speaker, forwarded, and checked by the one person in the room who knows the market. A primary-source footnote survives that; “Source: Statista” invites the question of which Statista page, which vintage and whose number. There is also a licensing question. Statista’s plan table lists publication rights only on its Professional plan (plans), so a chart exported from a lower plan is not automatically cleared for external publication. Citing and re-charting the original avoids both problems.

Statista’s price reflects licensing, not only data

On 2026-09-23 Statista’s plans for individuals were: Basic, free, with free statistics and 5 Research AI queries; Starter, $199 per month billed annually, adding premium statistics and downloads; Personal, $649 per month billed annually, adding reports and automated citations; and Professional, priced by sales, the only tier listing publication rights and Market Insights (Statista plans). The page states the account “requires an annual contract and will renew after one year to the regular list price.” What the higher tiers sell is breadth, exports and rights. When the figure you need comes from ITU, the IMF, the Census Bureau or an SEC filing, the original is free.

How Hyperresearch checks a citation against its source

Hyperresearch, the AI deep research API, applies the same test mechanically: its cite-checker binds each citation to the sentence that uses it, reads the cited source, and returns supported, partially supported, unsupported or wrong source with the evidence span. Sentences carrying numbers, percentages or “billion” are always checked; the receipt states how many of the remaining pairs were read. The findings are written to a verification receipt. A citation the checker did not read is counted on the receipt as not checked, never as passed. The same check runs as a standalone endpoint on a document Hyperresearch did not write.

Hyperresearch reads public pages logged out, so a paywalled Statista value is out of its reach; for statistics like the ones above it goes to the originals, which are public. See how it works and pricing.

Private market research firms raise a related question: why their market-size figures for the same industry differ. Why market size reports disagree compares eight of them on one market.