Is Perplexity safe? Research data and the Comet browser

Is Perplexity safe for store data? See its privacy settings, what the Comet browser can read and what to keep out of your research.

It depends on the data: Perplexity is useful for public store research, while confidential files need an approved business setup and restricted access. Free and paid personal accounts can use search data for training unless you opt out. Comet adds a separate risk because its assistant can read pages and act in accounts you’re signed into.

Is Perplexity safe for store data?

Use Perplexity for public market research and drafts that contain no customer information or confidential store material. Looking up published product specifications exposes less than uploading a sales report with customer names and order values.

Perplexity’s privacy notice covers queries, uploaded content and generated output. It also describes account and device data. Content you submit leaves the store’s systems even when you keep the resulting conversation to yourself.

The privacy notice allows processing overseas, including in the United States. It sets retention according to the purpose and legal requirements, so an opt-out from training doesn’t mean immediate deletion.

A personal Pro or Max subscription changes what you can use, but it doesn’t automatically give you Enterprise privacy terms. Perplexity’s data-collection guide says Free, Pro and Max accounts have AI Data Retention enabled by default. Enterprise data has a separate no-training commitment.

For a store team, the practical distinction is the account and the permission to send the data. An approved Enterprise account can support business research under its terms. It still needs the store’s rules for customer records and supplier confidentiality.

Perplexity privacy settings

Turn off AI Data Retention on each personal account to stop future search data being collected for model training. Perplexity’s data-collection instructions give this route:

  1. Open your Account settings.
  2. Go to Preferences.
  3. Find AI data retention and turn the toggle off.
  4. Reopen Preferences and confirm the toggle stays off on that account.

The setting applies to each user. Perplexity says the opt-out covers data collected after the change and doesn’t remove previously collected training data. Data can still be processed to operate the service and meet legal obligations.

Treat history and file deletion as separate tasks. Removing a conversation from view and excluding future training answer different questions about your data.

What Enterprise changes

Perplexity’s Enterprise privacy guidance says Enterprise data isn’t used to train its models. It also says agreements prohibit third-party AI providers from training on that data.

Files attached to sessions are deleted after seven days under the documented policy. The conversation containing a file remains unless deleted. Files kept in My Files or Projects follow their own retention policies, so the seven-day rule doesn’t cover every uploaded file.

Custom session retention is available to organisations with at least 50 seats or one Enterprise Max user. An administrator sets it under Organization Settings, then Data and privacy. Perplexity’s retention instructions measure the period from the session’s last activity. A follow-up query resets that period.

Before staff use business files, confirm they’re working in the organisation’s account and that its file-sharing and connector permissions match their work. A personal browser profile signed into a personal account needs its own controls.

Perplexity’s Comet browser

Comet’s assistant can use page contents and browsing context to answer questions or perform tasks. Perplexity’s Privacy Snapshot explanation says relevant open-tab information and browsing history can be sent to its servers when you ask the assistant for help. Some browser information stays on the device, but an assistant request can transfer page context for processing.

A signed-in store admin page can contain customer addresses, private order notes or refund controls. Those are different from a public product page. A browser assistant operating with that session can reach information the public can’t see, subject to its permissions.

Prompt injection happens when content the assistant reads contains instructions that try to redirect its task. A malicious supplier page could, for example, tell the assistant to copy data from another open tab. That is a hypothetical attack, not an instruction from the staff member who asked for research.

Perplexity describes classifiers, separation of untrusted content and confirmation steps in its Comet prompt-injection explanation. The same explanation calls prompt injection an unsolved problem. Keep access limited even when those protections are present.

For store work, separate public research from the browser profile used for privileged administration. Don’t give an assistant access to the store admin just to compare public supplier pages. Review any proposed action and its destination before approving it.

Enterprise administrators can use Comet domain permissions to set Browser Control, Read Only or No Access for specific domains. Read Only still permits reading the contents. Use No Access for a domain whose customer data the assistant mustn’t read.

Inspect domain overrides as well as the global switch. Perplexity says a domain permission can permit access even when the assistant is globally disabled. Confirm the store admin domain has the intended setting before staff open it.

Sales reports and supplier files in Perplexity

Keep identifiable customer data and restricted supplier information out of personal Perplexity accounts. An approved Enterprise setup reduces some risks, but it doesn’t give staff permission to upload every file they can access.

  • Remove names, contact details and delivery addresses from order or sales exports.
  • Exclude payment details, passwords and login codes entirely.
  • Keep private wholesale prices, contract terms and unreleased ranges within their approved systems.
  • Remove complaint details that reveal health information or other sensitive facts.
  • Inspect screenshots, hidden spreadsheet sheets and document comments before sending a file.

For a hypothetical category review, upload a small table of product categories and aggregated sales totals instead of customer-level orders. Remove rare combinations that could identify a buyer. Supplier restrictions can still apply after customer details are gone.

Use PII redaction when preparing text, then review the result for context that still identifies a person. Apply the store’s AI privacy rules to files and browser permissions as well as chat prompts. Public research needs public data. Keep restricted information in the store’s approved business workflow.

Give your store its own shopping agent

Customers ask in their own words. Your agent finds, compares and checks out from your catalogue, right on your site.

See it in actionFree trial · Simple setup