Broker guide
Is Perplexity Safe for Client Data in 2026?
Perplexity searches the web with what you type and may keep it. Check its data settings and which client details should never go into a search.
- Published
- Updated
It depends: Perplexity can help with public policy research, but keep identifiable client information out of a personal account. Its Enterprise plans have different training protections and retention controls. Turning off training on a Free or Pro account doesn’t, by itself, make that account suitable for client records.
Is Perplexity Safe for Client Information?
Perplexity is suitable for researching a general policy question without identifying your client. Use an approved enterprise arrangement before letting it process confidential records, and follow the uses your brokerage has authorised.
Separate what you submit from what a web search retrieves. Perplexity receives your prompt, then uses AI to interpret it and search the web. Search terms and requests used to retrieve external content are a different data flow from the conversation Perplexity stores.
Treat those search terms as information leaving your brokerage. A citation identifies a source used in the answer. It doesn’t establish that every cited website receives your full conversation, or that personal details in your prompt stay inside Perplexity.
Perplexity’s Privacy Notice covers submitted queries and uploads, as well as device information. It also describes overseas processing, including in the United States. A work email address or a paid personal subscription doesn’t change which agreement covers your account.
Perplexity’s Comet browser adds browser information to this assessment. Its Privacy Notice describes browsing-history collection depending on your settings and synced account. Keep client portals and documents outside an unapproved AI browser workflow, even when you’ve disabled model training.
What Perplexity Keeps and Uses for Training
Perplexity’s consumer plans allow training unless you opt out, while Enterprise customer content has a contractual protection against model training. Retention is separate: content can remain stored even when it isn’t used to train a model.
Plan Differences
| Account | AI training | Retention consequence |
|---|---|---|
| Free | AI Data Retention is enabled by default, with an individual opt-out | Turning it off doesn’t delete your account or search history |
| Personal Pro or Max | The same training opt-out applies | Paying for a personal plan doesn’t supply Enterprise retention controls |
| Enterprise Pro or Enterprise Max | Enterprise terms prohibit training on customer content | Organisation controls apply, with configurable session retention available to eligible organisations |
Perplexity’s data-collection instructions call the control “AI data retention”. Open Preferences in your Account settings and turn the toggle off.
The setting stops subsequent data collection for training. It doesn’t remove data already collected for training, and it applies to the individual user.
Perplexity can still process information to operate the service after you opt out. Its search-history retention guidance says personal information stays while the account is active. It describes removal within 30 days after account deletion, while the Privacy Notice preserves legal and other retention grounds.
The Enterprise Terms and Conditions prohibit Perplexity and authorised third parties from training generative AI models on customer content. That protection includes inputs and outputs. It doesn’t mean Perplexity never processes the content or that every staff member’s personal account has Enterprise protection.
Enterprise file retention also depends on where you upload a file. The Enterprise retention guide says session attachments are deleted after seven days. Files in My Files and Projects follow their respective retention policies.
Deleting an attachment doesn’t mean the conversation containing its extracted information disappears too.
Custom session retention requires at least 50 Enterprise Pro seats or one Enterprise Max user. The period runs from the session’s last activity, so a follow-up question resets it. Administrators must choose a policy that covers the brokerage’s actual use and know which content it affects.
What a Broker Must Never Enter
Keep client identifiers and confidential documents out of a general Perplexity search. Ask the policy question using only the facts needed to understand the rule, without reproducing a real person’s distinctive circumstances.
Exclude the following information from personal-account searches.
- Names, contact details and property addresses.
- Account numbers, tax file numbers and identity documents.
- Bank statements, payslips, tax returns and credit reports.
- Health information, hardship accounts and private correspondence.
- Screenshots or copied file notes containing information about clients or guarantors.
For example, this is a fictional policy search with no client identifiers:
What evidence do Australian lenders commonly request when assessing casual employment income? Find published lender guidance and cite the source documents.
A search naming a real employer, exact salary and property address can still identify a client after you remove their name. Use a separate generic question, without attaching the client’s payslip or asking the browser assistant to read their application.
The AI privacy guides explain how other tools handle information. Keep the search focused on the policy question and leave the client file in your brokerage’s approved systems.
