All posts
Analysis · 5 min read · 2026-09-20

The Perplexity Tracking Problem You'll Hit in October

There is a hard deadline next week that most businesses monitoring their Perplexity AI visibility are not aware of. On September 27, 2026, Perplexity retired its Sonar Chat Completions API -- the developer endpoint that powers most third-party AEO monitoring platforms. Tools built on that API and not yet migrated will start returning zero citation data for Perplexity, with no error message to indicate anything is wrong.

If you rely on any AI visibility tracking tool to see whether Perplexity mentions your business, you need to know this is happening.

What the Sonar API Is

Perplexity runs two kinds of access: the consumer product at perplexity.ai (what your customers actually use) and a developer API that allows tools to run Perplexity queries programmatically. The Sonar Chat Completions API was the primary developer endpoint until Perplexity deprecated it this fall.

The Sonar API is not a quirk that only developers care about. It's the data source feeding most of the AEO monitoring platforms and AI visibility reports that businesses pay for. When a tool tells you "Perplexity mentioned your business in 18% of queries this week," that number almost certainly came from API calls, not from a human checking perplexity.ai. Perplexity gave developers a 45-day migration window (August 13 to September 27) to move to its new Agent API.

In our investigation of this change, filed in the Sourcepull research vault on 2026-09-19, the key finding was the nature of the failure when tools don't migrate.

The Silent Zero Problem

The new Agent API returns citation data in a different location in the response structure. The old Sonar endpoint returned citations inside `choices[0].message.content`. The Agent API returns them inside `output[]`.

A tool that hasn't migrated will not crash. It will run the query, parse the response, look for citations in the old location, find nothing there, and return a result of zero citations. No error. No alert. No indication that anything changed.

A GitHub thread flagging this exact risk during the migration window describes it plainly: "A change dropping one of citations/search_results would otherwise take citations to zero without an error."

For a business watching its AI visibility dashboard, this produces a specific false signal: a sudden drop in Perplexity citation rates in early October, with no corresponding change in actual Perplexity behavior. If your business is appearing in Perplexity responses today, it will likely continue appearing after September 27. The tracking tool just won't see it.

Why Perplexity Data Has Always Been Harder to Track

The Sonar retirement is new. But Perplexity's citation data has been unreliable in third-party tools for longer than that.

Our research on Perplexity-Reddit citation patterns -- tracked continuously since early 2025 -- shows the underlying volatility. At the height of the Reddit-Perplexity relationship in mid-2025, approximately 46.5% of Perplexity responses included at least one Reddit citation. That was a direct consequence of Perplexity aggressively indexing Reddit following a cease-and-desist dispute. After Reddit filed a federal lawsuit against Perplexity in October 2025, the citation share collapsed, falling approximately 86% in a matter of weeks. By Q2 2026, our data shows that number had declined to around 7% of Perplexity responses.

That kind of swing -- 47% to 7% -- is a major structural shift in which sources Perplexity trusts. Most AEO tracking tools did not capture it in real time. Business owners using those tools saw numbers that were stale, miscalibrated, or outright wrong for months while the actual platform behavior had moved on.

The lesson from the Reddit collapse is the same lesson from the Sonar retirement: Perplexity's source data changes faster and more dramatically than any other major AI platform. Tools that assume a stable relationship between their API access and what users actually see will get burned.

Why Perplexity Specifically

There's a structural reason Perplexity data is more volatile than ChatGPT data.

Our platform architecture research, filed 2026-05-12 and drawing on the Yext 17.2M citation study -- the most comprehensive dataset on AI platform source behavior published to date -- describes the distinction:

ChatGPT primarily draws from training data. Its citation behavior reflects what was in its training corpus at a particular point in time and changes slowly. When you run the same query multiple times, you tend to get similar answers.

Perplexity uses live retrieval. It runs live web searches for every query and synthesizes from current web content. This is why Perplexity was so heavily influenced by Reddit's presence in search results, and why that influence disappeared quickly when it changed what it indexed. There is no stable snapshot; Perplexity's behavior reflects the current state of its index at query time.

An API endpoint that returns Perplexity responses is always one infrastructure change away from breaking. When the Perplexity platform changes -- its index, its retrieval layer, its data partnerships -- the API changes with it. Any tool dependent on that API needs constant maintenance to stay accurate.

What to Check Before October

**Verify manually.** Go to perplexity.ai and run a few of the queries your tracking tool monitors. Compare what you see there to what your tool reports. If your tool is showing zero Perplexity mentions but manual queries show your business appearing, the tool is likely affected.

**Ask your tool provider directly.** If you pay for an AEO monitoring platform, ask whether they've migrated from Sonar to the Agent API. A well-maintained platform will have documentation on this. If they can't tell you, treat their Perplexity data with low confidence.

**Treat October Perplexity dips with skepticism.** If any report you receive -- from a paid tool or a published industry study -- shows Perplexity citation rates declining sharply in October or November 2026, check the data source before drawing conclusions. A study using an API-based tool that wasn't migrated will show a cliff in Perplexity data that doesn't correspond to what's happening on the actual platform.

**Consumer-facing Perplexity.ai is unaffected.** This is a developer API issue. The product your customers use to search is running normally.

The Bigger Pattern

Perplexity's citation landscape has shifted more dramatically in the past 18 months than any other AI platform's. The Reddit collapse is the clearest example: a platform-level change that moved the citation needle by 86% while most tracking tools were still reporting the old numbers. The Sonar retirement is another inflection point in the same pattern.

Manual spot-checking -- running queries on perplexity.ai directly -- remains the most reliable way to understand whether your business is being cited there. Tools are useful for scale and for monitoring trends over time, but they are one API migration away from returning a clean zero.

Sourcepull's Signal Check is worth running now to establish a clean reference point before the October data landscape shifts. Whether or not the tools you use today survive the migration intact, having a baseline you trust matters when you're trying to tell the difference between a real visibility problem and a broken instrument.

See how your business scores on AI platforms.

Check your score — free