Why Your Perplexity Data Will Go Dark on September 27

Two things changed in the last 48 hours that affect whether the AI visibility data you're looking at is accurate. One is a deadline landing in eight days. The other is a setting that has been live since September 1 and may already be producing wrong numbers for some sites.

Neither one sends you a notification. Neither one throws an error. Your monitoring dashboard looks normal right either way.

Why Perplexity data matters more than you probably think

Before explaining what breaks, it's worth establishing what's at stake.

In our investigation of citation behavior across platforms -- drawing from three independent studies totaling 680 million analyzed citations -- we documented a finding that surprises most business owners: Perplexity cites businesses at a 13.05% brand citation rate, compared to 0.59% on ChatGPT. That is a 46x difference.

This is not a quirk of one study. The same number was confirmed independently by Averi (680 million citations), Whitehat SEO (118,000 responses), and Passionfruit (12% source overlap), all arriving at roughly the same result using different methodologies. The three-study convergence -- documented in our `platform-divergence-data.md` knowledge file, updated 2026-06-09 -- means this isn't noise. Perplexity is structurally a higher-citation platform because it runs a real-time web search for every query and cites sources in every response. ChatGPT answers primarily from parametric memory; citations are selective.

The practical consequence: if you're monitoring AI visibility, Perplexity data is not optional. A business that is invisible on Perplexity has a much larger problem than one that is invisible on ChatGPT, because Perplexity cites actively and frequently.

Which makes what happens on September 27 worth paying close attention to.

What Perplexity changed -- and what breaks silently

Perplexity is retiring its Sonar Chat Completions API on September 27, 2026. The migration window opened August 13. The new endpoint is the Agent API, and it uses a different response structure.

The old Sonar API returned citations inside `choices[0].message.content`. The Agent API returns them inside `output[]`. Any tool built on the old Sonar endpoint that hasn't migrated will not receive an error after September 27. It will execute normally. It will return a response. It will just return a response with zero citations.

In our 2026-09-19 investigation of this retirement (Scout session 143, `edge-cases/perplexity-sonar-api-retirement-2026-09-19.md`), we documented the specific warning that appeared in developer discussions:

> "A change dropping one of citations/search_results would otherwise take citations to zero without an error."

This is a silent data integrity failure. The tool appears to work. The citations just disappear.

The platforms at risk are any AEO monitoring tools that use the Perplexity API internally to run citation checks and haven't confirmed migration to the Agent API -- including third-party tools that display Perplexity citation rates as part of their dashboards. A sudden drop in Perplexity citation rates in any platform's data published after September 27 should be treated as a potential data artifact of this migration, not as evidence that Perplexity stopped citing the businesses in question.

One important distinction: Perplexity's own web app (perplexity.ai) and Perplexity Comet are not affected by the API retirement. A manual citation check -- running a query directly in Perplexity's browser interface -- will return accurate results regardless of what happens to API-based tools. If you see a discrepancy between a monitoring tool's Perplexity data and what you see doing a manual check, this is the likely explanation.

The other silent disqualifier that went live September 1

Separately, and filed in our methodology recommendations as a Priority-0 preflight item (Scout session 142, `methodology-recs/2026-09-18-google-ai-mode-opt-out-preflight.md`): Google's publisher opt-out controls for AI Mode and AI Overviews went globally live on September 1, 2026.

The mechanism is a single toggle in Google Search Console. If it's enabled, the site stops appearing in AI Overviews, AI Mode, and AI Overviews in Discover simultaneously -- as a linked source and as grounding data. The site otherwise looks and functions completely normally.

What makes this a monitoring problem is how easily the toggle can be accidentally enabled. A developer auditing GSC settings and toggling it without understanding the AEO consequence produces the same outcome as an intentional opt-out. An SEO agency recommendation to opt out of AI Overviews to protect referral traffic produces the same outcome. Editorial advice from media organizations encouraging publishers to opt out produces the same outcome. The site doesn't know it happened. The score doesn't change. The traffic attribution just quietly disappears.

Our methodology rec from September 18 adds this to the Phase 1 preflight checklist immediately before any audit work begins -- in the same group as robots.txt and user-agent checks, because all three are silent disqualifiers. The check takes thirty seconds: Search Console > Settings > AI Mode / AI Overviews control. If the opt-out is enabled, that is a Priority-0 fix before schema, directory, or content recommendations have any meaning.

There is also a competitive angle worth noting: if a client's key competitors have opted out, those citation slots in Google's AI surfaces are currently open. A competitor who opted out of AI Overviews to protect organic click-through rates has vacated a position that a well-optimized site can now take.

What to check in the next eight days

Three specific actions, in order of urgency:

**Check your Perplexity data against manual results.** If you're using an AEO monitoring tool that shows Perplexity citation data, run the same queries manually in perplexity.ai and compare. If the tool shows near-zero citation rates that don't match what you see manually, the tool may already be partially affected by the migration or may be using stale API data. After September 27, treat any platform-reported Perplexity citation data with skepticism until the tool confirms it has migrated to the Agent API.

**Verify the GSC AI Mode opt-out toggle.** Log into Search Console, navigate to Settings > Search Appearance, and confirm the AI Overviews/AI Mode control is not opted out. If you manage multiple clients, check each property. This has been live for three weeks and it's an easy thing to have missed.

**Check robots.txt for AI crawler blocks.** This one has been on the pre-flight checklist since July 2026, but it belongs in the same conversation: blocking GPTBot, PerplexityBot, or ClaudeBot in robots.txt produces the same silent-zero outcome as the Sonar migration failure or the GSC opt-out. If the block is there, schema and content work is irrelevant until it's removed.

Signal Check surfaces the robots.txt block and will flag the GSC opt-out status as part of its technical section. For the Perplexity monitoring data issue, the honest answer is that this is an industry-wide problem affecting API-based monitoring tools -- Sourcepull included -- and the right response is manual spot checks against perplexity.ai until each tool confirms its migration status.

The pattern across all three issues is the same: silent failures that look like data until you know what to check. The September 27 deadline is the hardest clock. Eight days is enough time to verify, but not enough to procrastinate.