Perplexity can cite Reddit discussions, G2 reviews, and independent comparisons when a B2B buyer asks which software to choose. A study run on August 16, 2026 found at least one Reddit citation in 96 of 200 decision stage prompts and at least one G2 citation in 33. A vendor blog can still earn a citation, but repeating category keywords gives Perplexity little evidence for a claim about customer experience or competitive fit. Study and methodology.
What the "Trust Graph" describes
The Trust Graph is a working model of the sources that recur around a category's buyer questions. Nodes represent cited pages and domains; connections show which sources support mentions of each vendor. Perplexity has not published a ranking system by that name, so the graph should describe observed answers rather than claim access to its internal source scores.
Citation Building is the work of making accurate product evidence available on pages a buyer might encounter and an answer engine might retrieve. An earned review, a detailed comparison, or a useful community reply can supply a different kind of evidence from a vendor's product page. The aim is a verifiable claim in a relevant source, with a citation as an observed outcome.
Perplexity's Search API returns ranked web results, while its answer products can generate prose with citations. Search retrieval and final citation are separate steps. A retrieved page may inform an answer without receiving a visible citation, and a cited page may contain only one fact used in the response. Perplexity Search API documentation.
Why third-party pages appear in software comparisons
A Reddit thread can record implementation problems that a vendor's sales page omits. G2 and Capterra organize software reviews around products and categories, giving an answer engine pages that match comparison requests. A reported product test in a news publication can establish when a capability became available.
Those sources carry different risks. Reddit posts can be anonymous and out of date; review platforms may contain incentivized feedback; a news story can describe a product version that no longer exists. A citation shows what Perplexity surfaced, not that the source independently verified every claim.
The 200 prompt study found Capterra in 10 answers and G2 in 33. It also cited vendor domains, including product and documentation pages, across its category tables. That mix argues against a blanket claim that Perplexity ignores SaaS blogs or follows a fixed list of preferred domains. The study used perplexity/sonar-pro through OpenRouter, so teams should test the Perplexity experience their buyers actually use. Source data.
Another citation study found Reddit in 45.8% of its Perplexity answers while community and forum pages accounted for 1.1% of its total citation volume. One answer can cite Reddit alongside many other domains. Report "share of answers containing a source" separately from "share of all cited URLs" or the same data will support conflicting headlines. Study definitions and limits.
Map the competitor before buying coverage
Start with prompts taken from sales calls and lost deals. Include direct comparisons, integration constraints, and industry use cases. Record the date and model for each run because answers and source lists can change.
A competitor citation mapping dashboard should preserve each prompt, the full cited URL, its domain, the vendor named in the answer, and the claim the page supports. Group URLs by source type so the team can distinguish a review profile from an independent test. Keep the original answer beside the extracted data; a domain count alone cannot explain why a rival appeared.
For each prompt, mark four outcomes: the brand is named and cited, named without an owned citation, absent while a competitor appears, or absent with no vendor recommendation. That classification separates a source gap from a brand mention gap. It also prevents a G2 page about the category from being counted as a citation for every company listed there.
Suppose a competitor appears in a Perplexity answer about SOC 2 reporting and the answer cites a G2 comparison page. Inspect that exact page before commissioning outreach. The gap may be an absent product profile, thin review evidence, or a feature claim the brand cannot substantiate.
Earn evidence where buyers already look
On G2 or Capterra, complete the product profile and keep category placement accurate. Ask eligible customers for candid reviews under the platform's rules. A higher review count or rating offers no citation guarantee, and fabricated feedback creates a credibility problem that reaches beyond AI search.
On Reddit, answer specific questions with technical detail and disclose a company affiliation when relevant. A staff reply that explains a limitation can help a buyer more than a planted recommendation. Community moderators and users can remove promotional posts, so this work needs product knowledge and restraint.
For editorial coverage, give reporters reproducible evidence: product access, dated release notes, a customer who consents to speak, or original data with a stated method. An article that confirms a claim can support future answers. Paying for a mention without editorial review creates a different source, with different trust and disclosure issues.
Keep the vendor site in the source mix
Independent coverage cannot replace current product documentation. Publish pricing conditions, integration requirements, security scope, and feature limits on accessible pages that a cited comparison can check. Perplexity documents separate crawlers for automated indexing and user requests; confirm that the intended pages are reachable under the company's access policy. Perplexity crawler documentation.
Measure citation change against a fixed prompt panel. For 100 prompts, report the number of answers naming the brand, the number citing its domain, and the number citing a competitor on a claim the brand also makes. Store the cited URLs so an editor can audit each count.
When a competitor wins the same prompt across repeated runs, inspect the cited page's evidence and publication date. Assign the gap to the owner who can change it: product documentation, customer review operations, or editorial PR. Rerun that prompt after the source changes and retain both answers in the dashboard.
Discussion
No comments yet — start the discussion.