The comparison
| Perplexity (Free/Pro) | Ask Safely | |
|---|---|---|
| Trains on your data | Yes, by default. "AI Data Retention" is on for Free, Pro, and Max tiers. Opt-out toggle exists in Account Settings, but the July 2026 policy rewrite removed the written description of it from the Privacy Notice while keeping the toggle in the app. | Never. Two layers: AWS Bedrock contract + Ask Safely's own policy. |
| How long your data is kept | Unclear. The original policy said 30 days after account deletion. The July 2026 rewrite replaced that with "as long as reasonably necessary" and no specific timeframe. | Auto-deletes on a timer: 8 hours (default), 7 days, or 30 days. Database TTL index. When the timer hits zero, it's gone. |
| Third-party tracking | A 2026 federal class action alleged Perplexity embedded Meta Pixel, Google Ads, and Google DoubleClick trackers that sent conversation content to Meta and Google, including in Incognito Mode. The case was dismissed without prejudice (can be refiled). Perplexity denied wrongdoing. | No third-party ad trackers. No Meta Pixel. No Google Ads. No advertising identifiers. |
| Memory | Conversation history and context retained for personalization. Details on user controls are limited in public documentation. | Off by default. Every fact visible, editable, deletable. Declined facts leave no trace. |
| Encryption | SOC 2 Type II and HIPAA gap assessment documented in Trust Center. Specific encryption details (algorithms, per-user keys) not publicly disclosed. | AES-256-GCM with per-user data encryption keys, KMS envelope encryption. Memory system is fail-closed. |
| Data collected | Search queries, full conversation content, uploaded files, voice audio, device/browser/IP, location, demographic and inferred-interest data, and (since July 2026) locally stored browser history from Perplexity's Comet browser. | Message content (encrypted, auto-deleted), account email (KMS-encrypted). No location, no device fingerprinting, no browser history. |
| Policy stability | Three significant changes in 2026 alone: removed deletion timeline, removed opt-out language from policy text, removed commitment to notify users of changes. | Privacy commitments are legal commitments. The deletion architecture is a database TTL index, not a policy promise. |
| Price | Free / $20/mo (Pro) / $200/mo (Max) | Free (Essential) / $12/mo or $120/yr (Expert) |
What Perplexity actually does with your data
Perplexity positions itself as a search tool, not a chat app. But under the hood, it collects more data than most chat apps do.
Training is on by default, with a disappearing opt-out
Perplexity trains on your searches by default. The "AI Data Retention" toggle is on for Free, Pro, and Max tiers. An opt-out exists in Account Settings, but the July 2026 privacy policy rewrite removed the written description of that opt-out from the Privacy Notice itself. The toggle still works in the app. The policy no longer mentions it. If you're looking at the privacy policy to understand what Perplexity does with your data, you won't find the opt-out described there anymore.
The tracker lawsuit
In March 2026, a federal class action was filed in the Northern District of California (Doe/Noel v. Perplexity) alleging that Perplexity embedded Meta Pixel, Google Ads, and Google DoubleClick tracking pixels in its product. The complaint alleged these trackers sent conversation content, search queries, and session data to Meta and Google's advertising infrastructure, including when users were browsing in Incognito Mode. The case was dismissed without prejudice in May 2026, meaning it can be refiled. Perplexity denied wrongdoing. In July 2026, Perplexity rewrote its privacy policy.
The deletion timeline vanished
The original Perplexity privacy policy stated data would be deleted within 30 days of account deletion. The July 2026 rewrite replaced that with "as long as reasonably necessary." No specific timeframe. No upper bound. The change happened without the advance notice the previous version of the policy had promised, because the rewrite also removed the commitment to notify users of policy changes.
The publisher scraping history
Before the tracker lawsuit, Perplexity had already faced credibility questions about how it handles other people's content. In 2024, WIRED published an investigation showing Perplexity was scraping content from publishers, including WIRED itself, and presenting it as AI-generated answers with minimal or no attribution. Forbes reported similar findings. Perplexity's crawler was found ignoring robots.txt directives. The New York Post and Dow Jones subsequently sued Perplexity for copyright infringement. The pattern matters here because it speaks to how the company treats data boundaries in general: when the default is "take first, address concerns later," user data gets the same treatment.
It collects more than a chat app
Perplexity collects search queries, full conversation content, uploaded files, voice audio, device and browser information, IP addresses, location data, and demographic and inferred-interest data. Since July 2026, Perplexity's Comet browser also collects locally stored browser history. This is a broader data footprint than most AI assistants. It's closer to what Google collects through Chrome than what a focused chat tool needs.
What Ask Safely does differently
Ask Safely runs on Anthropic's Claude models but wraps them in a different data architecture.
Conversations disappear. You pick the timer at signup: 8 hours, 7 days, or 30 days. A MongoDB TTL index handles deletion at the database level. When the timer expires, the conversation is permanently removed. This isn't a policy promise. It's a database constraint. The data can't overstay because the infrastructure won't let it.
No ad-tech trackers. No Meta Pixel. No Google Ads pixel. No DoubleClick. No advertising identifiers of any kind. Ask Safely uses Plausible (a privacy-focused, cookie-free analytics tool) for aggregate traffic data. No conversation content, no user identifiers, no data sent to advertising infrastructure.
The privacy policy matches the product. Ask Safely's deletion architecture is a database TTL index, not a line in a privacy policy that can be rewritten in a quarterly update. The no-training commitment has two independent layers: AWS Bedrock's contract prohibits it, and Ask Safely's own policy prohibits it. Removing one still leaves the other. This is architecture, not a promise.
Where Perplexity is better
Perplexity has capabilities Ask Safely doesn't. Being honest about this makes the rest of the comparison worth trusting.
Search is the core product. Perplexity was built as an AI search engine. It searches the web in real time and returns cited, sourced answers. Ask Safely is a conversational AI assistant that uses Brave Search when needed, not a search-first product.
Model selection. Perplexity lets you choose between GPT, Claude, Gemini, and its own Sonar models. Ask Safely routes across Anthropic's Claude family automatically based on the question.
Deep Research. Perplexity's Deep Research has more time in market and handles complex multi-step research workflows. Ask Safely's Researcher expert does similar work but is newer.
SOC 2. Perplexity has SOC 2 Type II certification. Ask Safely's SOC 2 is expected early 2027.
Enterprise tier. Perplexity offers enterprise contracts with zero-data-retention guarantees. Ask Safely doesn't have an enterprise tier.
The bottom line
If you want a powerful AI search tool and you're comfortable with the data practices described above, Perplexity is a capable product. Use it.
If you don't want your conversations sent to Meta and Google's advertising infrastructure, or you want them permanently gone after you're done with them, Perplexity doesn't guarantee either of those things. The tracker lawsuit was dismissed without prejudice, not on the merits. The deletion timeline is gone from the policy. The opt-out description is gone from the policy.
A product whose privacy policy got weaker three times in one quarter has negotiable commitments. Ask Safely's aren't. Deletion is architectural. No-training is contractual. The privacy policy describes what the infrastructure already enforces, not what the company intends to do.