Measurement guide · International platform examples
Bots and fake accounts: what the statistics count
An account estimate, an enforcement count and a bot rate answer different questions.
These disclosures do not provide a universal bot percentage—or a U.S. bot rate. A platform can estimate account categories within an active audience or count accounts it removed over a year. Those figures have different denominators and periods.
For example, Meta estimated that approximately 4% of worldwide Facebook MAUs were false accounts in the fourth quarter of 2023. Its category includes misclassified profiles as well as violating accounts; it is not a bots-only measure. Meta 2023 Form 10-K, limitations of key metrics.
Historical example · Worldwide Facebook accounts · Fourth quarter 2023 · Company estimate, not a survey of U.S. people or a current count
Duplicate and false accounts are different categories
Worldwide Facebook MAUs · Fourth quarter 2023
Source: Meta’s 2023 Form 10-K, Limitations of Key Metrics and Other Data. Original Social Usage graphic. Approximate company estimates; the categories are not presented as parts of a complete distribution and should not be added.
- Duplicate account
- An additional account a person maintains alongside a principal account. That definition alone does not establish automation, impersonation or malicious intent.
- User-misclassified account
- A personal profile used for a business, organization or nonhuman entity, such as a pet, instead of the source’s intended Page format.
- Violating account
- A profile Meta believes is intended for uses that violate its terms, with bots and spam given as examples. This is one part of the broader false-account category.
Definitions paraphrased from Meta’s account-metric limitations. They are the definitions in this historical disclosure, not universal categories for every platform.
Do not add 10% and 4% and call the result a “fake-user rate,” or subtract both from 100% to estimate real people. The source does not establish the mutually exclusive, exhaustive classification that calculation would require. Accounts are also not the same thing as distinct people.
A removal count measures recorded enforcement
In a separate disclosure, Bluesky reported removing 3,619 accounts for involvement in suspected influence operations during 2025. This is a service reporting count; no U.S.-only filter is specified. Bluesky 2025 Transparency Report, Influence Operations.
The count covers accounts identified through manual investigations and automated systems for suspected influence-operation activity. Bluesky cautions that additional affected accounts may have been removed through other enforcement processes without being specifically attributed to influence operations. The source’s “suspected” qualification matters.
View the separately scoped enforcement record
| Measure | Count | Denominator |
|---|---|---|
| Accounts removed for suspected influence operations | 3,619 | No rate denominator supplied for this measure |
Source: Influence Operations section. Accounts, not reports, posts, identified people or proven bots. No sampling interval supplied; this is an operational count with detection and attribution limits.
This figure does not say how many such accounts remained on the service, how many distinct people operated them, or what share of all accounts were bots. Dividing it by an unrelated year-end registered-account or active-user count would not repair those missing definitions. More removals can reflect different detection or enforcement activity; the count alone does not establish a change in underlying prevalence.
Facebook’s approximate MAU shares and Bluesky’s annual removals belong in separate panels. They cannot rank the platforms by bot prevalence or safety.
Start with the question the number can answer
| Your question | Evidence needed | What does not substitute |
|---|---|---|
| What share of an audience falls in a category? | A defined account population, category, observation window and estimation method. | Annual removals with no matching population at risk. |
| How much enforcement happened? | A defined action, counted entity and reporting period, including attribution limits. | An estimate of accounts remaining active. |
| What share are automated bots? | An explicit automation definition and classification method for the stated sample or population. | Every duplicate, misclassified profile or suspected influence-operation account. |
| How many followers of one creator are fake? | Evidence about that creator’s followers and the assessment method. | A service-wide account estimate or removal count. |
This is Social Usage’s reading framework, derived from the different entities and methods in the cited disclosures. It does not supply a new prevalence estimate or endorse a follower-audit service.
Methods, dates and reuse
Facebook’s active-user base and estimate limits
A Facebook MAU in the cited filing is a registered and logged-in user who visited Facebook or used Messenger while also registered with Facebook during the preceding 30 days. The account-quality percentages are described as fourth-quarter estimates; we do not relabel them as an exact December 31 census.
Meta uses internal signals and reviewer judgment, and warns that signal access, methodologies and product changes can affect estimates. Actual duplicate and false-account numbers may differ substantially. No subgroup, U.S.-only rate, trend, confidence interval or numerical “correction” to its MAU total is calculated here.
Meta 2023 Form 10-K: MAU definition and limitations of key metrics. For the separate historical active-audience figures, see the Facebook guide.
Source and review dates
- Meta Platforms, Inc.
- Data: Q4 2023. Publication: 2024; exact filing day not established here. Evidence checked 2026-10-08.
- Bluesky Social, PBC
- Data: 2025. Publication: 2026-01-29. Evidence checked 2026-10-08.
- Assisted review
- 2026-10-08 · Source definitions, extraction and presentation checked.
These are dated examples of different measurements. No claim is made that they are the newest available account-quality figures. New selected evidence was retrieved for this guide; earlier Facebook and Bluesky release histories and policy-retrieval dates remain unchanged.
This is assisted checking, not independent human editorial approval.
Attribution and reuse
Sources: Meta Platforms, Inc., 2023 Form 10-K; Bluesky Social, PBC, 2025 Transparency Report, published January 29, 2026. Social Usage supplies the original explanation and chart and is independent of both companies.
Selected factual editorial use is scoped to this presentation. The SEC’s EDGAR reuse guidance supports access to the Meta filing. Bluesky retains its report rights; its terms are not an open dataset licence. This page offers no CSV, image export or embed and reproduces no source graphics, account records or full source table.