Skip to content
eWiki The practical AI reference

AI Tools

How to Use Perplexity Source Filters, Check Citations, and Ask Focused Follow-Ups

Open blank notebook with eyeglasses and a fountain pen on a white desk beside stacked books and a magnifying glass.

Perplexity source filters vary by product context. Learn how to narrow sources in Pro Search or Enterprise, configure Search API domain filters, verify citations, and refine results with focused follow-ups.

Identify Which Perplexity Source Filter You Need

Perplexity documents different filtering controls for different contexts. In Pro Search, search modes can narrow results to content types such as Academic, Finance, or a user’s files. Perplexity’s Enterprise interface provides a data-source control for enabling web search, academic papers, SEC filings, connected organizational files, and certain other connected services. The Search API instead uses search_domain_filter to include or exclude specified internet domains.

These controls should not be treated as interchangeable. Before searching, identify whether you want to choose a category of material, select an Enterprise data source, or restrict API results by domain. I suggest using the narrowest control that directly matches your research objective.

Stack of blank papers marked with blue and gray tabs and a dark strip beside a magnifying glass on a white desk.

Narrow Sources in Pro Search or Enterprise

For Pro Search on the web or in Comet, Perplexity instructs users to select Search in the question box. Pro Search operates automatically when the user’s plan includes it, subject to that plan’s limit. The same official guidance says a user can choose a model from the model menu or leave the selection on Best. Search modes can narrow the search to particular content types.

The documented Enterprise workflow is more specific. Its data-source button can turn web search on or off and add sources such as academic papers or SEC filings. It can also expose connected organizational files and supported external services. Availability therefore depends on the product context and configured connections.

As an editorial recommendation, choose only the source categories relevant to the question. This may make the result easier to review, but it should not be presented as a guarantee of accuracy or completeness.

Person wearing glasses seated at a desk, handling a printed sheet among folders, papers, an open notebook, pen, ruler, lamp, and potted plant.

Configure Search API Domain Filtering

The officially documented Search API domain filter has two modes. An allowlist contains unprefixed entries and includes only the specified domains. A denylist uses a minus prefix on every entry and excludes those domains.

For example, an allowlist can be expressed as:

{"search_domain_filter":["nature.com","science.org"]}

A denylist can be expressed as:

{"search_domain_filter":["-reddit.com","-pinterest.com"]}

The filter supports root domains, top-level domains, and paths. A root-domain entry can match its subdomains; a TLD entry can be written as .gov; and a path such as nature.com/articles can narrow results to that section. I suggest using an allowlist when you have a defined set of acceptable publishers and a denylist when only particular domains need to be removed.

Avoid Common API Filter Errors

Perplexity’s domain-filtering guidance says not to combine allowlist and denylist entries in one search_domain_filter array. Choose one mode for each request.

Wildcard syntax is not supported. Use .gov rather than *.gov, or provide a path prefix when a particular site section is required. An empty filter array has undefined behavior; omit search_domain_filter when the search should cover all domains.

Perplexity also suggests keeping allowlists focused, with five to ten domains usually sufficient. Treat that quantity as provider guidance rather than a universal requirement. My practical recommendation is to begin with the smallest defensible domain set and expand it only when relevant sources are being missed.

Check the Citations Behind the Answer

Perplexity says its answers link to numbered citations that users can preview or open to read the full source. Its Pro Search guidance also advises validating information against the sources linked in an answer. In the documented Enterprise interface, users can view sources associated with the query and inspect sources used for individual answer sections.

I suggest opening every citation attached to a decision-critical statement. Compare the claim with the linked page, confirm that the page addresses the same subject and scope, and separate multi-part claims so each part can be evaluated. This is an editorial verification method, not a Perplexity guarantee.

Source labels can help with initial triage, but reading the linked material is still the practical check recommended here. If the page does not clearly support the claim, treat the point as unresolved and refine the question.

Ask a Focused Follow-Up

Perplexity documents that Pro Search retains context from earlier interactions when users continue with follow-up questions. Its Enterprise guidance says users can ask their own follow-ups and can highlight a particular answer area for inclusion in a follow-up.

A useful editorial approach is to target one unresolved issue at a time. For example: “Which linked source supports the second sentence?”, “Answer only from the academic sources already shown,” or “Separate the evidence for these two claims.” These prompts are suggested wording, not documented promises about the resulting answer.

After the response, repeat the citation review. Retained conversational context can support refinement, but the provider’s own guidance still calls for validating information through linked sources.

Conclusion

Perplexity source filters are most useful when matched to the correct environment: content-type modes for Pro Search, configured data sources for Enterprise, and domain or path restrictions for the Search API. My practical recommendation is to start with a narrowly framed question, apply only the relevant filtering control, open the citations supporting the most important claim, and use one focused follow-up for anything the linked evidence does not resolve. For API work, keep allowlists and denylists separate and omit an empty filter.

Sources

Disclosure

This article was prepared with AI assistance.

Improve this guide

Report a problem or suggest the next question

Submissions enter a private editorial queue. They are reviewed and deduplicated before any public update or new page is created.