Perplexity SEO: How to Rank on Perplexity and Get Cited in Its Answers

Perplexity is the engine that shows its work. Every answer carries numbered citations. The API that produces them returns two lists: the URLs the answer was built from, and the search results behind them with a title, a snippet and a date.

That makes it the easiest engine to reason about, and the one where a vendor can see most directly why it was or was not cited.

This guide covers how Perplexity’s sonar model picks sources, what is and is not documented about citation order, why dates matter, and what a property-ops vendor can change. It also says what the AI Shortlist Index will record once it runs Perplexity. The commercial version, for vendors who want the work done, is how to get cited by Perplexity.

What does ranking on Perplexity mean?

Two things, and they are not the same. A vendor is cited when one of its pages is among the numbered sources behind an answer. A vendor is named when the answer text says its name.

A PM reading a Perplexity answer sees citation [1] first and is most likely to click it.

We expect that being cited at all matters more for whether a vendor is named than where in the citation list it lands. The answer names vendors from the sources it read regardless of their footnote number. That is a hypothesis to be checked against mention rate and citation position across editions.

The Index scores position from the vendor’s first mention in the answer text, not from its footnote number. The two are reported separately once runs exist.

How does Perplexity choose its sources?

The default search model is sonar. Perplexity’s model documentation describes sonar as a lightweight, cost-effective search model with grounding, suited to quick factual queries, topic summaries, product comparisons and current events. It positions sonar-pro above it for complex queries and follow-ups.

A property manager asking which maintenance software fits 300 doors is a product comparison. The Index protocol uses sonar rather than sonar-pro because the question is what a PM gets by default, not which model answers best. It sets no domain filter, no recency filter and no search mode beyond the default web search, because a PM typing into the Perplexity app does not set them either.

What comes back

Two response fields do the work. Perplexity’s API reference defines citations as the URLs of sources used to generate the response. It defines search_results as an array of objects each carrying a title, a url, a snippet, an optional publication date and an optional last_updated date.

The answer text itself carries bracketed numbers that point into those lists. Citations tell you which URLs were used.

Search results tell you what the engine saw about each one: the snippet it extracted and when the page was published or last updated. For a vendor the snippet is the useful field, because it shows which passage of your page the engine pulled.

Perplexity’s search control documentation describes a search_context_size setting with low, medium and high levels that governs how much content is extracted per result, alongside a manual token budget per page. The consumer product does not expose that setting, and the Index protocol leaves it at the API default so captured answers stay comparable to what a PM would see.

The distinction between cited and consulted matters less here than it does for ChatGPT, where the consulted sources run far longer than the inline citations. For Perplexity the two are closer together.

How is citation order decided?

The documented part is short. Perplexity’s API reference gives citations as an ordered array, and the answer text numbers its references. The documentation we reviewed does not state how the order is decided or how bracketed numbers map to array positions.

Our working hypothesis is that citation order follows the order in which each source was first used while writing. The first citation would be the source behind the opening sentence, and that order would correlate with search rank without being the same thing.

The Index will test that against the raw files once Perplexity runs exist. Until then it stores the citations array in order in every run file without asserting a mechanism.

Why do dates matter on Perplexity?

Every search result comes with a date, and the API exposes filters built on it. Perplexity’s API reference documents a search_recency_filter that limits results to the last hour, day, week, month or year, and search_after_date_filter and search_before_date_filter that bound results by publication date. The Index protocol sets none of them; it measures the unfiltered baseline.

The existence of those filters says something about how the engine treats time. Freshness is an attribute of a source that the API exposes directly.

For a property-ops vendor that has practical consequences. A pricing page last updated in 2023 carries a 2023 date into the search results. An integration list that added Rent Manager last spring but never updated its dateModified looks older than it is.

We think a vendor whose pricing and integration pages carry current, visible dates is more likely to be cited than an older listicle stating the same facts. The engine has the dates, and a reader-facing product has every reason to prefer the fresher source. That is a hypothesis, and it is cheap to act on.

Which filters exist, and why do they matter to you?

Perplexity documents a search_domain_filter that restricts results to an allowlist of up to 20 domains or excludes a denylist, with optional paths. It also documents a search_mode that switches between general web, academic and SEC-filing sources. The Index runs with none of these.

They matter to vendors for a different reason. Any tool that claims to measure “Perplexity visibility” and quietly sets a domain filter, a recency filter or a non-default search mode is measuring something a buyer never sees. When you compare Index numbers to a vendor’s own tooling, ask which filters were set; the Index’s answer is none, and its request parameters sit in every run file.

What should a property-ops vendor do?

In the order the mechanism suggests.

1. Answer the question in the first paragraph. The snippet in search results is a passage, and the passage the engine extracts from your page is the raw material for the citation. A category page that opens with “maintenance coordination software for property managers on AppFolio, Buildium and Yardi, priced per unit per month” gives the engine a matching snippet. A page that opens with a slogan does not.
2. Date your pages, and mean it. Visible publication and modified dates, matching dateModified in the page’s structured data, so the search-result date reflects when the facts last changed. The structured data guide covers the Article and Organization markup that carries those dates.
3. State the facts a comparison needs. Price, minimum portfolio, integrations by PMS name, whether there is a service layer. These are the same five facts the ChatGPT run went looking for, and a search engine that extracts a per-result snippet will find them or not.
4. Be on the pages the category conversation happens on. Marketplace listings on the PMS platforms, the property management associations, trade publications. The Index’s source map for each category, such as the maintenance and vendor network sources page, will show which of those appear in Perplexity’s citations once the runs are in.
5. Track it monthly with the same prompts. The AI visibility guide defines mention rate, position score, consistency and source dependency, the metric that will show which cited domains carry your mentions on Perplexity.

What will the Index measure on Perplexity?

Nothing yet. The September 2026 edition’s numbers come from ChatGPT, and an engine that has not answered drops out of every denominator. Nobody gets a zero for Perplexity because Perplexity was not asked.

When it runs, each prompt goes to the sonar model three times, in fresh sessions, spread across at least two days. The adapter stores the answer text, the citations array, the search results with URL, title, snippet and date, the model identifier, token usage and cost, and the complete raw response. Files are written once, and a vendor disputing a row gets the raw file.

What the ChatGPT runs suggest

The only run whose consulted list is visible is a ChatGPT answer to one maintenance prompt, captured through the OpenAI API on 3 September 2026. It was retired from the edition when capture moved to the app. In that run ChatGPT named Property Meld first and Latchel second, and its consulted sources were led by reddit.com, appfolio.com and buildium.com.

In the 25 app captures the edition now rests on, the links in ChatGPT’s maintenance answers went mostly to vendor sites, led by latchel.com, propertymeld.com and lula.life. Whether Perplexity reaches for the same sources is exactly the kind of question the Index exists to answer.

Our expectation is that Perplexity’s citations for the same prompt will skew toward pages with a clear date and a quotable snippet, and away from forum threads. Vendor pricing and integration pages and dated listicles fit that description. The first Perplexity edition will confirm or kill that.

Limits, and what changes month to month

The endpoint is changing. Perplexity’s API reference, read on 2026-09-04, stated that sonar chat completions was being succeeded by its Agent API and that sonar would be supported until 27 September 2026.

The Index will record the endpoint and model it uses on its first Perplexity run and log the choice in the Index changelog. A shift in who gets cited across that boundary is a capture-method change first and a market change second.

Models drift. The model identifier returned by the API will be stored per run and shown per edition, so month-to-month movement can be read against a model change.

Answers vary. Three runs per prompt on separate days and a published consistency metric make the variance visible without removing it.

API is not the app. The Perplexity consumer app carries its own system prompt, its own model routing and personalization. API sonar is the closest repeatable proxy, and the Index labels it as such.

US only, unpersonalized. The Index measures the baseline a new US user gets, with no account history and no location beyond the country.

Related guides

How to rank in ChatGPT is the contrast case, with a long trail of consulted sources, few inline citations, and a real run to read. The AI Overviews guide covers an engine the Index can only observe from a captured results page. Gemini and search grounding covers grounding chunks, the nearest equivalent to Perplexity’s search results.

The Index’s full protocol is on how we measure AI recommendations, and this site’s own scorecard against these guides is the reference implementation. The mechanism the engines share is on the AI search optimization pillar.

Changelog

• 2026-10-28: Reformatted with a short answer, question headings, the vendor steps as a numbered list, and an FAQ. Retitled toward “how to rank on Perplexity”. The endpoint note now gives the date the documentation was read, since the stated support date has passed.
• 2026-09-26: The September figures now cite the 25 ChatGPT app captures per category; the API run of 3 September 2026 is described as retired from the edition and archived.
• 2026-09-03: Rebuilt for the AI Shortlist Index scope. Replaced the general citation-earning checklist with the sonar capture and the citations and search_results fields. Added the documented and undocumented parts of citation ordering, the date filters, and the missing-engine status.
• 2026-06-10: First published as a general Perplexity SEO guide.