Bot-traffic percentages measure network activity

One automated client can send thousands of requests while a person reads a page once. Request counts measure the resulting traffic, leaving the number of people to be measured separately.

Claims about fabricated online conversation sometimes begin with a bot-traffic percentage. Moving from requests to people changes the denominator. That step needs evidence about the audience before the traffic measurement can support it.

Account impersonation creates a trust problem. Automated abuse can damage a service, while misleading media can deceive its readers. Each problem needs a measurement tied to the behavior under investigation.

A request counts network activity

Thales' 2026 Bad Bot Report analyzes activity from 2025. It reports bots at 53% of traffic, with bad bots accounting for 40%. Those categories describe the vendor's classified observations. Estimating the population of internet users would require a different sampling method. The 2026 report overview

Cloudflare Radar labels its bot-traffic panel as a share of HTTP requests. Readers can also choose the observation period. Assessing the people in a comment thread requires evidence about that thread's participants. Cloudflare Radar's traffic view

Imagine a site with 100 human visits while a crawler sends 10,000 requests. The request count is dominated by automation. The example leaves the visitors' identities unspecified, as well as the crawler's purpose. It also supplies no evidence about the truth of a statement on the site.

That site could have little visible conversation while its infrastructure remained busy. Measuring an audience requires defining participation, then collecting data about the people taking part.

Inspect the report's population

A security provider's telemetry comes from the customers it serves. Its classification rules decide which observed activity falls into each category. The report's publication year may follow the calendar year of collection.

Comparing providers requires checking whether they observed comparable populations. Customer mix can change a percentage, as can the chosen time window. Classification rules can also differ, so a larger headline figure needs more context before being called growth.

A bot-protection vendor has a commercial reason to describe automated threats. Its data can still be informative. The reader needs enough methodological detail to judge which claims the observations support.

A site operator can use that report to investigate automated access on their own service. A claim that human participation has disappeared would need evidence about participation itself. A suspicious account's statements can also be checked individually.

Specify what authentication establishes

Passkeys address account access through phishing-resistant authentication. The FIDO Alliance distinguishes that task from verifying identity, while discussing how digital credentials can supply identity evidence alongside authentication. FIDO's identity and authentication explanation

A securely authenticated account can publish a false statement. Its owner may authorize automated posting. A returning credential establishes continuity of access without verifying every detail in the account's biography.

A product needs to define the assurance its feature supplies. Login may require continuity of control. A regulated transaction may require an identity process, while public discussion needs moderation of abusive behavior. A single badge conceals those different checks unless the interface explains its meaning.

I'd state the check directly. Control of a credential establishes one fact; receipt of a document from an organization establishes another. Assessing a post's truth requires evidence about the post.

Read provenance as a file history

Content Credentials supply cryptographically bound records of media origin with editing history. C2PA's explainer states that those records leave truth or factuality to another assessment. It also describes incomplete provenance, including records that have been removed. The C2PA explainer

A photograph of a real event can be captioned with the wrong date. Its file history may help investigate the caption. A photograph without credentials may also be authentic, requiring other evidence for that assessment.

Provenance gives an investigation another record to examine. A badge claiming universal truth would extend that record beyond what its history can establish.

The interface should show the record's supplier beside the checks performed. If the record leaves the reader's question unanswered, that gap needs to remain visible.

Define the decision before choosing evidence

A website owner can investigate whether automated requests cause failures. Account takeover requires another investigation. An action needing stronger identity evidence gives the product team a third, separately defined task.

For a reader, I'd begin with the specific statement. Find its original source before examining the date with its context. Look for contradictory evidence. Repetition by several accounts can still trace back to a single unsupported assertion.

Requests measure activity on a service. Authentication supplies evidence about account control. Provenance records media history, with the truth of an individual claim requiring evidence relevant to that claim.

When a bot percentage appears in an argument about people, check what the underlying count measures before accepting the inference.