Perplexity Finance vs the new AI Google Finance: what each one does, and where they slipped in dated tests

For many investors, the first AI answer about a stock now comes from a finance portal rather than a chatbot window. Two are worth knowing: Perplexity Finance and Google Finance, rebuilt around AI. Both are free at the door. Here is what they include and what dated, documented tests found, as of October 14, 2026.

Google Finance: rebuilt and out of beta

Google began rebuilding Google Finance around AI in August 2025, reached more than 100 countries in April 2026, and took it out of beta on June 25, 2026 with a new Android app, according to The Next Web. Google's May 11 announcement for its European launch lists the features:

• AI-powered research: ask about a stock or a market trend and get an AI answer with links; Deep Search for more complex questions, now globally available.
• Charts with context: technical indicators, and 'key moments' on the price chart explaining why it moved on a given day.
• Live earnings: call audio with synchronized transcripts and AI-generated, annotated highlights.

The Next Web adds portfolio setup by uploading a CSV, PDF or screenshot, after which the AI can answer questions about that portfolio, plus scheduled briefings you describe in plain English; an iOS app is promised later. We found no published accuracy test of Google Finance's AI answers specifically.

Perplexity Finance: the answer engine with a stock page

Ben Dixon's AI Reliability Tracker, which re-checked free-tier limits on August 23, 2026, ranked Perplexity Finance the best free first pass on a US-listed name among seven free tools: a stock page, an earnings hub with documents and a plain-English screener in one place. It reports that Perplexity's free Standard account includes five Pro Searches a day, with a separate limited Research allowance.

Where Perplexity slipped

The same tracker documented its failures with dates, from a Pro account between May 14 and June 18, 2026:

• A thousandfold unit error. On May 14, it turned a small company's $6.095 million revenue into '$6K' and narrated a collapse that never happened, apparently missing the filing's 'in thousands' label. The identical question returned the correct figure on June 18.
• Coverage gaps. On a smaller US company, the transcript tab never loaded and it suggested a paid service instead. On May 25, Unilever's page showed 'No data found' for analyst consensus while Apple's showed named ratings and targets.
• A clean answer too. On Next Fifteen Communications, a small UK company, it gave revenue and guidance correctly and said plainly it could not find an adjusted EBITDA headline in its sources.

The tester stresses that these are dated product sessions, not a controlled experiment, and that the underlying model varied by session.

In a larger test, Daloopa's 500-question retrieval benchmark scored Perplexity Finance at 33.8% exact match on single reported figures nContext: n=500 single-number questions · run date not shown, models of that time · Daloopa, a data vendor · margin: not disclosed, behind ChatGPT and Grok in that run. Perplexity has since launched a Finance Search tool for developers that combines licensed financial datasets with web sources, which it says improves accuracy on a time-sensitive retrieval benchmark; that is Perplexity's own evaluation, and it applies to its developer API.

Side by side

Using them well

• Treat both as a first pass. They are good at assembling the documents; the number you act on should come from the filing.
• Check units and periods. The documented Perplexity failure was a unit label, the kind of error a quick look at the filing catches.
• Notice coverage. Smaller and non-US companies are where both the tracker's gaps and the 'no data' pages appeared.
• Prefer answers that admit a gap. The UK answer that said what it could not find was more useful than a confident complete one.