Perplexity Review: 4/5 — Cited Answers Fast, Then One Answer Came Back in Chinese

I opened Perplexity in its own Incognito mode and typed a dense diabetes lit question before I bothered with signup. The UI said Incognito sessions are not saved and expire after 24 hours. Sidebar still showed liang leo / Free plan from an older shell, but this run never touched password, email verify, Google/Apple SSO, or a card. I just wanted sources I could click, not another chat that invents confidence.

Seven seconds on fasting and HbA1c

Exact prompt:

“What did 2023–2024 systematic reviews and randomized trials find about intermittent fasting and HbA1c in adults with type 2 diabetes? Please cite studies, report HbA1c changes and limitations, and distinguish direct trial evidence from broader reviews.”

UI: Researched 7s. First useful prose landed almost immediately while the rest kept streaming. Not diet-blog mush. It put numbers on the page: a 2023 RCT meta-analysis with SMD −0.08 (95% CI −0.20 to 0.04; P=.19); INTERFAST-2 intermittent fasting −7.3 mmol/mol (~−0.6 percentage points) vs control +0.1; EARLY 5:2 meal-replacement −1.9 points at 16 weeks vs metformin −1.6 and empagliflozin −1.5. It called out limitations and bundled components in the same breath.

Source drawer: 24 sources. Links tab showed PMC, PubMed, Diabetes Care and friends. That is the product working — I stayed on one screen instead of opening ten tabs myself.

A card also sat over the answer: “This question looks difficult” with Upgrade to Pro. The answer still finished on Free/Incognito. Soft upsell, not a hard stop. Mildly annoying. I kept reading.

The follow-up that became a table

I asked it to rebuild the evidence:

“Turn the evidence into a concise table comparing the 2023 meta-analysis, INTERFAST-2, EARLY, and the 2024 16:8/14:10 trial: sample, duration, HbA1c change, and what each actually proves. Flag any study where the intervention bundled meal replacements or medication changes.”

Researched 8s. Sources climbed to 30. The table held the caveats: insulin-dose adjustment in INTERFAST-2; meal replacements, calorie guidance, contact, and education bundled into EARLY. That was the moment I stopped testing “can it write” and started trusting it for structure. Prose-to-table without dropping the footnotes is rare.

I did not independently recalculate those medical effect sizes. I checked that the citations looked concrete and that pooled evidence was separated from direct trials. Good enough for this session; not a systematic review of my own.

Thirty-six seconds, then Chinese

Second query, different domain:

“What evidence from 2023–2025 studies supports or challenges the claim that remote work changes software-engineer productivity? Summarize at least three studies with methods and measured outcomes, and distinguish causal from correlational evidence.”

Researched 36s. Over four times slower than the diabetes ask. Fine. Waiting.

Then the whole answer rendered in Chinese even though the prompt was English. The synthesis underneath still looked serious across 25 sources: Gibbs/Mengel/Siemroth (2023) full-remote IT setting with output/hour down about 8–19%; Bloom/Han/Liang (2024 Nature) hybrid RCT, 2 days/week at home, no significant performance/code-line damage and lower attrition; BLS industry stuff labeled correlational. Useful distinctions. Wrong language. That glitch is the kind of thing that makes you reread the thread twice before you trust the next one.

I clicked the Nature citation anyway. It opened “Hybrid working from home improves retention without damaging performance” (12 June 2024). Nature’s own cookie dialog. Real link. So the retrieval path worked even when the language router tripped.

Another Upgrade to Pro card on this difficult question, plus a “Pro Free preview of advanced search enabled” banner during the follow-up / second query stretch. Still no hard wall.

Pricing as shown, and who this is for

Personal pricing modal, annual billing:

• Pro — $17/month. Expanded Computer access, 4,000 bonus credits, deep research, top models, model selection, polished documents/apps, higher limits/memory.
• Max — $167/month. 35,000 bonus credits and 10,000 monthly credits.
• Free — referenced as the base (“Everything in Free”). Account shell said Free plan. Modal did not show a separate numeric Free allowance, so I never watched a daily counter tick down.

No purchase flow. I also skipped file upload, Collections, exhausting a daily quota, and email signup/OTP.

If you need fast cited synthesis and a table rebuild on Free without creating an account first, this session delivered. If you need predictable language on every hard prompt, or a clear Free allowance before you commit to a research day, those are still open questions from what I saw.

Final rating: 4 out of 5.