Comet Enterprise: Perplexity now logs the AI answers too
With the Comet Enterprise release in March 2026, Perplexity didn’t just roll out a managed AI browser for organisations — it also expanded its audit logs significantly. Logs now capture more than just the queries employees submit to Perplexity: they include the generated answers, the model used, the answer mode and the cited sources. For data protection, compliance and IT, that’s a clear step forward — and a new mandatory question in any AI tool evaluation.
What actually changed
- Audit logs now include answers. They additionally capture the streamed answer text, the model used, the mode and the sources — not just the question.
- Granular feature controls. Admins can selectively allow or restrict API access, the visibility of org-wide Spaces, available AI models and individual Comet features per user group.
- Silent Comet rollout via MDM. macOS and Windows can centrally distribute Comet, configure hundreds of browser policies and control which actions the AI agent is allowed to take.
- No model training on company data. All activity inherits the Enterprise settings for retention, audit and permissions; data never goes into training sets.
What was true before
Comet launched in summer 2025 as a personal AI browser for individuals and Pro subscribers. Audit logs in the Enterprise variant existed — but mainly captured account and admin actions plus the submitted search queries. Whatever Perplexity returned to the employee remained outside the log. For compliance officers that left a gap: you knew what was asked, not what the AI assistant answered. For sensitive questions or regulated industries, that’s not enough.
In parallel, the line between Pro and Enterprise was blurry: an employee with a Pro account could effectively use the same models and features as an Enterprise licence holder. Centralised control over which parts of the tool are even allowed inside the company was missing.
What is true now
1. Audit logs cover the full answer event. According to the Perplexity changelog from March 13, 2026, logs now additionally contain the streamed answers, including model, mode and cited sources — not just the question. Compliance teams can reconstruct complete AI interactions: what was asked, what was answered, which model produced it, which sources were used. For industries with documentation duties — banking, health, public sector — Comet only becomes truly deployable from this point on.
2. Granular feature controls enable real tool governance. Organisation admins can configure API access, visibility of the org-wide Spaces list and available AI models per user group. That’s more than an on/off switch: sales teams might get different models from engineering, API access is reserved for a small power-user group. Exactly the kind of governance IT only knew from classic SaaS tools so far.
3. MDM rollout and browser policies make Comet IT-grade. Comet can be silently deployed via Mobile Device Management on macOS and Windows. Hundreds of browser policies are available, plus fine-grained control over which actions the AI agent may take in the browser at all — for example whether it can fill forms, open tabs or download files. That’s the difference between “experimental tool” and “managed work asset”.
How to read it
Extending audit logs to AI answers sounds small but is strategically the moment AI browsers become deployable in regulated environments. Anyone working in banking, insurance, healthcare or the public sector must be able to reconstruct an AI answer later — otherwise the tool isn’t certifiable for many use cases. Perplexity addresses exactly this hurdle before it becomes a show-stopper.
What few in the German-speaking market notice: an AI browser without answer logs cannot be audited from a data-protection point of view. With this update Comet becomes the first AI browser product where a DPO can technically reconstruct the actual answer — without “looking over the shoulder” of the production system. Competitors such as Edge Copilot or Chrome AI offer no comparable out-of-the-box auditability at this level today.
What you can do now
If you’re evaluating Comet: Ask for a live demo of the audit log structure, not just a doc excerpt. Answer, model, mode, sources — all four fields must actually show up in exported logs. That’s the central compliance question.
If you’re in a regulated industry: Clarify with your DPO whether audit data can be exported locally and how long Perplexity retains it. The default retention follows Enterprise settings, but that is not the same as an on-premise solution.
If you’re already rolling Comet out: Set the feature controls before the rollout. Once employees get used to features that later have to be turned off again, adoption suffers. Better start narrow and expand later.
Entdecke mehr
GEO — Generative Engine Optimization
GEO is the discipline of shaping content for visibility in AI answer engines — AI Overviews, Perplexity, ChatGPT Search, Claude. The goal is not the classic SERP click, but appearing as a cited source inside the generated answer.
LexikonGEO — Generative Engine Optimization Explained
What GEO is, how generative engines cite sources, and which factors raise citation likelihood. Honestly framed, without the hype.
NewsPerplexity drops advertising entirely — and goes all-in on subscriptions
In February 2026 Perplexity removed sponsored answers for good. What it means for publishers, GEO visibility and the trust question.