Perplexity AI has established itself as one of the fastest-growing AI tools in the workplace. Positioned as an AI-powered search engine, it answers questions by searching the web in real time and synthesising results into direct, cited responses. For professionals who spend large parts of their day researching, fact-checking, or gathering competitive intelligence, it has become a genuinely compelling alternative to traditional search.
But as with every AI tool that gains rapid workplace adoption, the security and compliance implications have lagged behind the enthusiasm. Most organisations using Perplexity have done so without any formal assessment of what the tool does with the queries and context their employees submit. That gap matters, and this article explains why.
As with our previous analysis of Claude and ChatGPT, the answer to whether Perplexity is safe to use at work is: it depends on how you're using it, and whether you have the right controls in place. Here is what you need to know.
What Is Perplexity AI?
Perplexity AI is a US-based AI company founded in 2022. Its core product is an AI-powered answer engine that combines large language model capabilities with real-time web search. Unlike ChatGPT or Claude, which generate responses based primarily on their training data, Perplexity retrieves live information from the web and uses AI to summarise and present it with source citations.
This makes it particularly popular for research-intensive tasks: market analysis, competitor research, regulatory lookups, technical documentation, and current events. It is available as a free consumer product, a paid Pro tier, and an Enterprise Pro plan with different data handling terms.
Perplexity has attracted significant investment and partnerships, including a reported relationship with SoftBank. It has grown rapidly and positioned itself as a direct competitor to Google Search for professional research workflows. That growth has brought increased scrutiny, including from publishers who have raised concerns about how Perplexity handles their content, but for the purposes of this article, we are focused on the data security and compliance implications for organisations.
What Does Perplexity Do With Your Data?
This is the question most organisations haven't asked, and it is the most important one. The answer, as with other AI tools, depends heavily on which tier your employees are using.
Perplexity Free and Pro (Consumer Tiers)
Perplexity's privacy policy states that it collects the queries users submit, along with associated metadata including IP address, browser information, and usage patterns. Queries may be used to improve Perplexity's models and services. The policy also notes that Perplexity may share data with third-party service providers involved in operating the platform.
Critically, Perplexity's consumer product does not offer a data processing agreement. This means that any personal data submitted via consumer Perplexity (customer names, employee details, client information), is being processed without the contractual safeguards that UK GDPR requires when using a third-party data processor.
Perplexity also uses third-party AI models to generate responses, including models from OpenAI and Anthropic. This means that when an employee submits a query to Perplexity, that query may be passed to one or more third-party AI providers as part of the response generation process, adding an additional layer of data exposure that is not always immediately obvious to users.
High risk scenario: A business development manager uses Perplexity to research a target acquisition, pasting in details about the deal structure and the target company's financials to get a competitive analysis. That information, which may be material non-public information, is now submitted to Perplexity's servers, potentially passed to a third-party AI provider, and retained for service improvement purposes. This is a data protection issue, a potential insider trading concern, and a breach of deal confidentiality.
Perplexity Enterprise Pro
Perplexity's Enterprise Pro plan is designed for business use and comes with materially different data handling terms. Under the enterprise agreement, Perplexity commits to not using customer data to train its models, provides a data processing agreement, and offers additional administrative controls including SSO, user management, and audit logging.
Enterprise Pro also includes the ability to restrict which AI models are used to generate responses, reducing the risk of data being passed to multiple third-party providers without organisational oversight.
If your organisation uses Perplexity for anything beyond publicly available research on non-sensitive topics, Enterprise Pro is the appropriate tier. The consumer and standard Pro products are not suitable for business use involving non-public or sensitive information.
Key distinction: Perplexity Free / Pro = queries retained, may be used for model improvement, no DPA, data may be passed to third-party AI providers. Perplexity Enterprise Pro = no training on your data, DPA available, model controls, audit logging. These are not interchangeable from a compliance perspective.
Risks Unique to Perplexity
Beyond the data handling concerns common to all AI tools, Perplexity has several characteristics that create specific risks for business users.
1. The Research Context Problem
Perplexity's value proposition is research. Users naturally provide rich context to get better results, they describe their industry, their organisation, their strategic objectives, and the specific decision they are trying to inform. This context-rich interaction pattern means that employees are, by design, sharing more sensitive business information with Perplexity than they might with a general-purpose AI tool.
A lawyer researching case precedents may include client details. A strategy analyst may include confidential market assumptions. A procurement manager may include supplier pricing information. The tool encourages this behaviour because richer context produces better answers, but the data protection implications are the same regardless of how useful the output is.
2. Third-Party Model Exposure
Unlike Claude or ChatGPT, which use their own models exclusively, Perplexity routes queries through multiple AI providers depending on the query type and the user's settings. On the consumer and Pro tiers, users have limited visibility into which model is processing their query at any given moment.
This means that a single query submitted to Perplexity may be processed by Perplexity's own infrastructure and one or more third-party models, each with their own data handling policies. From a data governance perspective, this is significantly more complex than a direct relationship with a single AI provider.
3. Citation and Accuracy Risks
Perplexity's cited responses give answers an air of authority that can encourage over-reliance. Because responses include source links, users may assume the information has been verified, but Perplexity's AI summarisation can misrepresent, truncate, or incorrectly attribute source material. Several publishers and researchers have documented cases where Perplexity's summaries materially mischaracterised the content of the sources cited.
For organisations using Perplexity to inform business decisions (pricing strategies, regulatory compliance, competitive positioning), acting on inaccurate summaries without independent verification carries real commercial and legal risk.
4. No Persistent Memory: But Query Logs Remain
Unlike ChatGPT's memory feature, Perplexity does not maintain persistent memory across sessions by default. However, query history is logged and accessible within the product. On shared or personal accounts used for work purposes, this creates a record of an employee's research activity, including any sensitive business context submitted, that persists in Perplexity's systems and may be accessible to Perplexity staff.
Query log risk: An employee's Perplexity account contains months of research queries, many of which include commercially sensitive context about clients, competitors, and internal strategy. If that account is compromised, or if Perplexity experiences a data breach, that research history becomes exposed. Unlike a Google search history, which is bad enough, Perplexity queries often contain full sentences of sensitive business context rather than simple keyword searches.
5. Perplexity's Relationship With Publisher Content
Perplexity has faced significant criticism from publishers and news organisations for allegedly reproducing copyrighted content in its responses without adequate attribution or licensing. Several major publishers have taken legal action or issued formal complaints. For organisations in regulated industries or those with strict IP policies, using a tool that has an unresolved relationship with intellectual property rights warrants careful consideration.
Risk Summary
| Risk | Applies To | Level |
|---|---|---|
| Sensitive business context submitted via consumer tiers, no DPA | All organisations | High |
| Queries passed to third-party AI providers without visibility | Consumer and Pro tier users | High |
| No AI usage policy, employees using Perplexity unsupervised | All organisations | High |
| Research context pattern, richer queries = more sensitive data shared | Strategy, legal, finance, BD teams | Medium |
| Query history logged, sensitive research activity persists | All consumer and Pro tier users | Medium |
| Inaccurate summaries presented as cited, over-reliance risk | All organisations | Medium |
| IP concerns around publisher content reproduction | Media, publishing, regulated sectors | Medium |
| Perplexity Enterprise Pro with DPA, controlled deployment | Organisations on enterprise plan | Low |
How to Use Perplexity Safely in Your Organisation
Perplexity is a genuinely useful research tool, and the risks above are manageable with the right approach. Here is what responsible deployment looks like.
1. Establish a Written AI Usage Policy
An AI usage policy is the foundational control for any organisation using AI tools, and it applies to Perplexity just as it does to ChatGPT or Claude. The policy should specify which tools are approved, what data must never be submitted to any AI tool, and how AI-generated outputs should be verified before being acted upon. Without a policy, you have no framework and no defensible position.
2. Move to Enterprise Pro for Business Use
If Perplexity is being used for anything involving non-public business information, client data, or commercially sensitive research, your organisation needs to be on Enterprise Pro with a DPA in place. The consumer and standard Pro tiers are not appropriate for business use involving sensitive data.
3. Define What Can and Cannot Be Submitted
Given Perplexity's research context problem, the natural tendency to provide rich business context to get better answers, it is particularly important to give employees clear guidance on what kinds of context are and are not acceptable to include in queries. General industry questions are fine. Client names, deal details, internal financial data, and unreleased strategy are not.
4. Verify Outputs Before Acting on Them
Perplexity's cited responses do not guarantee accuracy. Source citations indicate where information was retrieved from, they do not confirm that the summary accurately represents that source. Any Perplexity output used to inform a material business decision should be verified against the original source before being relied upon.
5. Classify Your Data
Data classification, categorising information by sensitivity, is the control that makes AI governance enforceable. If employees understand what constitutes confidential data and know that it must not be submitted to external AI tools, your risk exposure drops significantly. Without classification, your AI policy is difficult to apply in practice.
6. Train Your Staff
A short, practical briefing on AI risks (covering what these tools do with submitted data, what must never be shared, and how to verify AI-generated outputs), is a necessary component of your security awareness programme. Perplexity in particular warrants specific mention given how naturally its research workflow encourages the sharing of sensitive context.
The short version: Perplexity is safe for business use if you are on Enterprise Pro with a DPA, have a written AI usage policy, have defined what context employees can and cannot submit, verify outputs before acting on them, and have trained your staff. If none of that is in place, your employees are conducting sensitive business research through an uncontrolled third-party channel right now.
What Should You Do Right Now?
If your organisation doesn't have an AI usage policy and your employees are using Perplexity on consumer or Pro accounts, the immediate steps are straightforward:
- Find out if Perplexity is in use. Ask your team leads or run a quick survey. Perplexity's growth has been rapid and organic, it may be more widely used than you realise.
- Issue interim guidance immediately. Tell staff not to submit client data, personal data, or confidential business information into any AI tool, including Perplexity, until a formal policy is in place.
- Check your DPA position. If Perplexity is being used commercially, verify whether you have a DPA. If you don't, you are likely processing data without adequate contractual safeguards.
- Assess whether Enterprise Pro is appropriate. If Perplexity is genuinely valuable to your team's research workflows, the right answer is not to ban it, it is to deploy it properly on the enterprise tier with appropriate controls.
- Review your broader AI tool landscape. Perplexity is one of many AI tools your employees may be using. ChatGPT, Claude, Gemini, Copilot, Grammarly, Notion AI, each has its own data handling implications. A single AI governance review covering all tools in use is significantly more efficient than addressing them one by one.
Not sure which AI tools your organisation is using: or whether they're being used safely?
KAVIX works with London businesses to audit AI tool usage, identify data protection gaps, and put the right policies and controls in place. Whether it's Perplexity, ChatGPT, Claude, or any other AI tool your team has adopted, we can help you understand your exposure and close the gaps. Get in touch for a free conversation.
Talk to Us About AI SecuritySummary
Perplexity AI is a capable and genuinely useful research tool. Its real-time web search with cited responses fills a genuine gap that general-purpose AI assistants don't address as well. For organisations willing to deploy it properly (on the enterprise tier, with a DPA, and with clear usage policies), it can be a valuable addition to professional workflows.
The risk, as always, is the gap between the tool's capability and the governance frameworks most organisations have in place. Perplexity's research-oriented workflow is particularly prone to sensitive data submission, its multi-provider AI architecture adds complexity to data governance, and its consumer tiers offer no contractual data protection for business use.
The organisations that address this proactively (with proper enterprise agreements, clear policies, and trained staff), will get the productivity benefits without the compliance exposure. Those that don't are conducting sensitive business research through an uncontrolled channel, and may not realise it until something goes wrong.