Perplexity searches beyond the open web

Two Help Center pages rechecked on September 14, 2026 describe more than one general search context. Source selection includes Web, Org Files during a migration, their combination, and None. A separate premium-data page adds licensed collections and connectors to databases for which a user already has access rights.

The material matters as a clarification of what is being measured, not as a new ranking factor. A Perplexity answer can be generated in a context where a public web page is not the only available source class. The documentation-modification date confirms the state of the pages; it does not disclose when every product capability launched.

Section sources:[1] Perplexity[2] Perplexity

These contexts are not one strict mutually exclusive taxonomy: some can be combined, and availability depends on the account. They are still useful as separate measurement fields because each sets a different search boundary and a different reproducibility condition.

• Web: Perplexity uses sources from the internet.
• Org Files and Web + Org Files: transitional modes using an organization's file repository; Org Files itself is being retired.
• Included premium sources: built-in licensed collections that do not require a separate partner account.
• Licensed connectors: data from a user's existing subscription to an external provider.
• None: an answer with no selected external sources.

Section sources:[1] Perplexity[2] Perplexity

Perplexity divides premium data sources into two access models. Included sources are available inside the product without a separate partner account; the current list names Wiley, PitchBook Essentials, CB Insights, Statista, and Midpage. The number of premium queries is still limited by plan. A licensed connector instead uses a subscription that the user already holds with a data provider.

For connectors, the documentation adds plan and entitlement requirements: a Perplexity subscription alone may not be sufficient, and Enterprise access may need administrator approval. Two observers asking the same question with different entitlements therefore do not necessarily have the same potential source pool. That is an inference from the documented access rules, not disclosure of Perplexity's internal selection algorithm.

Section sources:[1] Perplexity

When a researcher stores only an answer's visible links, the record covers the final output of one run, not every document the system could have considered. With a premium source or internal context active, an answer may draw on material that does not exist on the open web. A missing ordinary domain is compatible with several explanations, including a different source pool, a changed prompt, different access rights, or different source selection.

One disappearing link should not become a claim that a website lost AI visibility. First repeat the question in a Web context and preserve the selected sources. Even then, a visible source link is not proof that every relevant passage was used, that the site held a particular hidden rank, or that the user followed the link.

Section sources:[1] Perplexity[2] Perplexity

For every run, store the date, exact question, language, product mode, account type, organization context, and selected sources. Record separately whether Web was enabled, which premium source or connector was used, and whether the observer could open the underlying document. Archive the answer text and visible links without merging multiple repetitions into one row.

A useful control pair is one Web run and one run of the same question in an available premium context. Compare more than domains: count named named subjects, factual claims, displayed source links, and whether those source links are accessible without authentication. A difference shows the effect of the conditions in that test; without repetitions, it does not provide a universal score for Perplexity or establish causality.

Section sources:[1] Perplexity[2] Perplexity

A public page remains relevant to the Web context, but it does not compete with a closed database on equal terms in every scenario. Publishers have no documented method to force Perplexity to prefer an open URL over a licensed source. The verifiable work remains familiar: accessible HTML, clear authorship, primary data, precise definitions, a stable preferred page address, and direct evidence make a page usable in web search and auditable by a reader.

Reports should separate four events: PerplexityBot requested a page, a public URL appeared as a source link, a claim from the page matched the answer, and a user followed the link. The new documentation adds another required field: source context. Without it, a comparison can silently blend the open web with licensed or internal data.

Section sources:[1] Perplexity[2] Perplexity

The official pages confirm source classes and access conditions, but they do not publish weights, a records of source selection, or a link-selection formula. They do not prove that premium data always displaces the web, that public sites were demoted, or that a connector improves every answer. The premium-sources page's source link promise also does not replace independent inspection of a specific answer.

Perplexity says Org Files is being retired: files should be moved, and material left in that repository will be deleted after October 1, 2026. We will recheck the page on October 2 and distinguish the retired repository from Project files and supported file connectors, which the documentation explicitly says are not part of this retirement.

Section sources:[1] Perplexity[2] Perplexity