Perplexity

Is Perplexity safe?

Perplexity is a safe enough way to search the web with AI, and on privacy it sits close to ChatGPT. It has no social feed, no messaging and no known breach, and it gives you a clear switch to stop training on your searches. The catches: training is on until you turn it off, the app can link your precise location to your account, and Comet, Perplexity's AI browser, has twice been shown by researchers to follow instructions hidden in web pages. We score it 57/100.

What Perplexity collects

Apple's App Store label lists no tracking across other companies' apps. Data linked to you includes precise and coarse location, purchase history, product interaction, identifiers and diagnostics. Google Play's form adds that the Android app shares the list of apps installed on your phone and device IDs with other companies to make the app work and for analytics.

Your searches are the bigger issue. Perplexity's help centre says "AI data retention" is on by default for Free, Pro and Max users, which means your questions and uploads can be used to train its models. Switching it off covers only future data; Perplexity says previously collected training data "cannot be deleted or removed".

The privacy notice, updated in July 2026, says Perplexity does not sell personal data and does not send your queries or conversations to advertisers. It also says third parties that help measure its own ads may collect data "over time and across websites" and use it to show you tailored ads. On the web you can refuse those cookies; in the app, the opt-out is your phone's tracking settings.

Comet: an AI browser that can be hijacked

Comet is Perplexity's browser, with an assistant that can read pages and act for you, including inside accounts where you are logged in. The privacy notice says that, depending on your settings, Perplexity records your browsing history when Comet is synced to your account.

Two research teams have shown how that power can be turned against users:

This class of attack, called prompt injection, is a weakness of AI agents in general, not just Comet. But it means an agent with access to your inbox should not browse unknown pages unsupervised. Amazon has also sued Perplexity over Comet shopping on Amazon accounts; a district judge blocked the agent in March 2026, and an appeals court lifted that order in August 2026, according to press reports.

Accuracy and the publisher lawsuits

Perplexity's selling point is that answers cite sources. That helps you check them, but it does not guarantee they are right. Several publishers have sued, alleging Perplexity copies their journalism and sometimes attributes text to them that they never wrote: Dow Jones and the New York Post in October 2024, Encyclopaedia Britannica in September 2025, the Chicago Tribune and The New York Times in December 2025, and CNN in May 2026. These are allegations; none had been decided on the merits when we checked. For readers, the practical lesson is to open the cited article before relying on a quote or number.

How to use Perplexity more safely

1. In your account settings, turn off "AI data retention" to stop future training.
2. Use incognito mode for health, money or legal questions. Those searches are not saved to your history.
3. Deny the app precise location unless you need local results; approximate location is enough for most searches.
4. In Comet, do not give the assistant access to email or banking, and confirm every action it takes on a logged-in site.
5. Open the cited source before acting on an answer about medicine, law or money.

How we scored it

Transparency and control (65) is Perplexity's strongest factor thanks to the training switch and incognito mode. Privacy (48) is held down by default training, precise location and ad-measurement partners. Security (58) reflects the Comet prompt-injection findings rather than any breach. For searching with fewer data trails, see DuckDuckGo.