Citations stop loading. You switch queries. Same result. Refresh, clear the cache, try a different browser. Still confident answers with no sources attached.
Mode conflict. Seconds to fix once you know where to look.
That’s the pattern with most Perplexity problems. Not deep bugs. Configuration mismatches, quota hits, browser quirks, or an API change you haven’t accounted for yet. Here are the ten that come up most, updated for where Perplexity stands in late 2026.
Fix 1: Perplexity Giving Wrong or Outdated Answers
Start with which search mode you’re in. Look at the mode selector on the search bar. Perplexity now has four distinct modes: Best, Pro Search, Reasoning Search, and Research.
Best is the default. It picks the most appropriate model for your query without using up your weekly Pro Search quota. That’s worth knowing. Best handles most general queries well and costs you nothing against your limits.
Switch modes deliberately based on what you’re actually asking:
- Research mode (Deep Research): Use this for complex, multi-source topics. It can take around 4 to 5 minutes and produces a comprehensive report with progress visible as it works (Source). Use it when you need synthesis across many sources, not when you want a quick answer.
- Reasoning mode: For analytical queries that require breaking down complicated problems into steps. Uses Claude 4.6 Sonnet on Pro; Max subscribers also get Claude 4.6 Opus.
- Pro Search: The general mode with deeper synthesis than Best. Uses your weekly quota.
If you’re getting wrong answers in Best mode, the issue is often source quality. Check the citations. If the sources themselves are thin, the answer will be too.
Perplexity can sound authoritative about outdated information without flagging it.
Fix 2: Pro Search Quota Exhausted
Pro Search runs on weekly limits on the Pro plan, not a fixed daily number. Perplexity describes it as “weekly limits (average use)” for Pro subscribers. When you hit the wall, you’re done until the weekly reset.
The practical fix: Best mode has no quota. Use it for follow-up questions, clarifying questions, and anything that doesn’t need deep web synthesis. Save Pro Search for the initial complex query where you need it.
Research mode (Deep Research) has a separate monthly limit. If you’re doing a lot of deep research sessions, those count against a different counter than your Pro Search quota.
If you’re on the free tier: Pro Search is limited to 3 queries per day. Research mode is 1 query per month on free. Neither goes far for regular research work.
For daily research use, the Pro plan ($17/month when billed annually) is worth comparing against alternatives (Source). The Perplexity AI pricing breakdown covers what each tier actually gets you.
Fix 3: Citations Broken or Returning 404s
Perplexity’s citations are snapshots. The source existed when Perplexity retrieved it. If the page has since moved, been deleted, or landed behind a new paywall, you’ll get a 404.
This is how the web works, not a Perplexity failure. But it’s frustrating when you need to verify a claim.
Workarounds:
- Paste the URL into web.archive.org. Wayback Machine often has a cached version.
If you’re getting 404s on most citations for a specific query, that’s usually a sign Perplexity sourced from low-quality sites that rotate URLs frequently.
Fix 4: Perplexity App Not Loading (Mobile)
The mobile app is more fragile than the web version. Three things cause most app failures:
Clear the cache first. This fixes most “app opens to a blank screen” or “stuck on loading” issues.
Check for updates. Perplexity ships mobile updates frequently. A version mismatch between your app and the current API can cause silent failures: searches that appear to run but return nothing, or a UI that loads but doesn’t respond. Check your app store.
Test on cellular vs. WiFi. Some corporate and university networks block Perplexity’s API endpoints. If the web version at perplexity.ai works on cellular but the app doesn’t work on WiFi, that’s the network. Not fixable on your end without a VPN.
If the app is crashing on launch: delete, reinstall, and don’t restore from iCloud backup. The cached state can carry forward the same issue that caused the crash.
Fix 5: Deep Research Not Working or Timing Out
Deep Research (listed as Research mode in the interface) can take around 4 to 5 minutes per report (Source). That’s normal. You should see a progress display showing which sources are being read and key findings as they appear. If the progress display is running, the feature is working.
Quota issues. Free users get 1 Research query per month. Pro subscribers have monthly limits (described as “average use” with no published count).
Model differences matter. Max subscribers use Opus 4.6 Thinking for Deep Research. Pro subscribers get 4.5 Thinking, rolling out gradually; some Pro accounts may not have it yet.
Topic scope. Deep Research works best on questions that require synthesizing multiple sources: market research, academic topics, due diligence. For questions with a quick factual answer, Best mode or Pro Search will return faster with no quality loss.
If you’re getting timeouts: try breaking a very broad research question into a more focused one. “Competitive landscape of B2B SaaS CRMs with pricing” performs better than “tell me everything about CRM software.”
Fix 6: Search Mode Not Finding Relevant Sources
You switched to a specific mode and you’re getting worse results than the default. That’s usually a scope mismatch.
Match the mode to the question type:
| Question type | Best mode choice |
|---|---|
| Research / scientific | Research mode |
| Product recommendations | Best (default) |
| Complex analytical | Reasoning mode |
| Multi-source synthesis | Research mode |
If you’re in the right mode and still getting irrelevant results, the query itself may be too vague. Perplexity handles specific queries better than general ones. “Machine learning” returns broad noise. “How does RLHF work in training LLMs” returns useful synthesis.
Fix 7: Perplexity Slow or Timing Out
Slowness breaks down into two scenarios.
Mode selection on Pro. Research mode can take around 4 to 5 minutes (Source). That’s expected. Pro Search and Best mode should return in a few seconds. If Pro Search is running slowly, try switching to Best mode for that query. If Best is slow, the issue is server load or your connection.
Connection issues. If Perplexity is slow on one device but fast on another, the issue is local. Check whether a VPN is adding latency. Confirm you’re not on a throttled mobile connection.
If queries time out entirely: reload the page, try the query again. If it consistently times out on complex queries, try breaking the query into two simpler ones. Very long input prompts can hit processing limits.
Fix 8: Perplexity Projects Not Syncing
Projects let you save searches and maintain context across sessions (Source). When they don’t sync, you lose threads you thought were saved.
The actual save behavior. If you’re relying on auto-save, check that you’re not in a private/incognito session. Those expire after 24 hours (Source).
Login state across devices. Perplexity can appear to work without being logged in (guest sessions), and that’s when people lose threads they thought were saved.
Fix 9: API Errors (For Developers)
Perplexity’s API changed significantly in 2026. Sonar Chat Completions support ended on September 27, 2026. If you’re building on the old Sonar endpoint, synchronous and streaming requests are being reformulated as Agent API requests on a rolling basis, but asynchronous requests are no longer supported. Move those to background mode in the Agent API. For any new projects, use the Agent API directly.
Current Agent API issues:
401 Unauthorized. Your API key is invalid or not being passed correctly. Keys go in the Authorization header as a Bearer token: Authorization: Bearer pplx-xxxx. Regenerate from the API portal if you’re unsure.
429 Too Many Requests. You’ve hit rate limits. Implement exponential backoff. Don’t retry immediately on a 429; that makes it worse.
HTTP 400 with max_output_tokens is required. This is specific to Anthropic models in the Agent API. When you use any anthropic/* model, you must include the max_output_tokens parameter. Omitting it returns HTTP 400. This is a breaking difference from the old Sonar interface.
Model name errors. The Agent API uses provider-prefixed model IDs: openai/gpt-5.6-sol, anthropic/claude-sonnet-4-6, perplexity/sonar. The old Sonar model identifiers like llama-3.1-sonar-large-128k-online no longer work. Check the Agent API model reference for current IDs.
Perplexity’s own Sonar model is still available in the Agent API at perplexity/sonar ($0.25/1M input, $2.50/1M output). The Agent API also gives you access to models from OpenAI, Anthropic, Google, xAI, and others through one endpoint.
Fix 10: Login and Account Access Issues
Can’t get into your account or your Pro subscription isn’t showing up.
Pro subscription not activating. Banks sometimes block AI subscription charges from unfamiliar merchants. If payment went through but features are still missing, contact support with your transaction ID.
Pro status not visible. This happens if you’re logged into the wrong account. Sign out completely, clear cookies, and sign back in on the account that holds your subscription.
When to Contact Perplexity Support
If none of the above works:
- Billing issues (charges with no service, plan that won’t activate) always need human review.
- Account lockouts: if you’ve lost access to both your primary email and your social login, contact support with your account creation date and the email you used.
- API production incidents: if your Agent API integration is down in production, open a support ticket with your request ID and response logs.
Perplexity support is at Perplexity’s Help Center. Response times are reasonable for Pro users.
If the app or website itself is completely unreachable, the issue may be broader than your account. The Is Perplexity AI down? page covers how to check server status and what to use while it’s offline.
For everything working fine technically but not returning great results, the issue is usually how Perplexity is being used rather than how it’s running. Our guide to using Perplexity AI effectively covers query structures and mode combinations that consistently return better synthesis. And if you’re weighing whether Perplexity is the right tool at all, the Perplexity vs ChatGPT and Claude vs Perplexity comparisons break down the actual use-case differences.
TechSifted is reader-supported. When you buy through links on this site, we may earn an affiliate commission at no extra cost to you. This doesn’t influence our editorial process; we evaluate tools independently.