How AI assistants handle your questions, compared

How we compare

We compare what each provider says on its own published pages, not what we infer about how its systems work.

• Sources. Every statement about ChatGPT, Gemini, Claude, Copilot or Perplexity quotes or closely paraphrases a page published by OpenAI, Google, Anthropic, Microsoft or Perplexity. We opened each page ourselves and give the date we checked it. Where none of a provider’s pages we checked addresses a question, we write “Not stated” and say no more.
• Scope. We compare consumer versions used with a personal account. Each provider handles business, education and API plans under different terms.
• Brello. Facts about Brello describe Brello 1.0, version 1.0.0 (build 1), a prototype Android app that runs language models entirely on the phone, as of 5 October 2026.
• Wording. We don’t rank the assistants or call any of them the most private, and we use no logos and no adjectives about other products.
• Checks. We re-check every source at least every 90 days, and whenever a provider changes a policy, and log each check in the version history of the comparison it affects.

The full process is in our editorial standards. For the concepts behind this page, such as retention, training and human review, read how AI assistants handle your data.

Where processing happens

Brello 1.0 processes questions on the phone. The other five assistants process them on their providers’ systems.

In Brello 1.0, the prompt goes only to the on-device runtime, which runs one of three downloaded models: Brello Pro, based on Gemma 4 E4B by Google; Brello Vision, based on Gemma 4 E2B by Google; or Brello Core, based on Qwen3 1.7B by Alibaba. Brello isn’t affiliated with or endorsed by Google or Alibaba. There is no Brello server, proxy or relay. The exception is web search: when you turn it on, or approve it for one question, the phone sends a cleaned-up search text directly to a search engine and fetches up to four result pages, so the search engine and those sites see a normal request, including the phone’s IP address. How Brello handles your questions, photos and answers covers this in full.

OpenAI collects your prompts and uploads as “Content”,1 and says it stores content on its own and its service providers’ systems “in the US and around the world”.2 Google relies on its own services and infrastructure for the Gemini app, and grounds some answers in Search results.7 Anthropic transfers personal data to its servers in the US, or to other countries outside the EEA and UK.10 Microsoft may store and process personal data in your region, in the United States and wherever it or its providers operate facilities.15 Perplexity collects queries and uploaded documents directly from you.20

On-device AI is not unique to Brello. Android’s documentation says Google’s Gemini Nano runs in the AICore system service, so apps that use it can offer generative AI “without needing a network connection or sending data to the cloud”.9 The comparison with the Gemini app covers how that differs from Brello’s approach.

Storage and deletion

Brello 1.0 keeps chats only on the phone. The other assistants keep them on their providers’ systems, under different rules for how long and for what happens when you delete them.

ChatGPT keeps saved chats until you delete them, then schedules them for permanent deletion within 30 days, unless they were already de-identified or must be kept for security or legal obligations.4 Claude removes a deleted chat from your history immediately and from Anthropic’s back end within 30 days; with model improvement on, Anthropic may keep chats de-identified for up to 5 years.13 Gemini Apps Activity is auto-deleted after 18 months by default, and you can choose 3 or 36 months or turn auto-delete off; with Keep Activity off, chats are still saved with your account for up to 72 hours.8 Chats that Google’s service providers have reviewed are kept for up to 3 years, even if you delete your activity.7

Microsoft says the current Copilot app keeps prior conversations so you can revisit them, and you can delete one or all of them.17 Its FAQ for the older app says: “By default, we store conversation activity for 18 months.”19 Perplexity says it keeps personal data “for only as long as necessary to fulfil the purposes in this Notice”, without a fixed period.20

Brello 1.0 stores chats in one file in its private app storage. Android backup is turned off for the app, and every storage domain is excluded from cloud backup and device-to-device transfer. Nothing is synced, so if the phone is lost or reset, the chats are gone.

Model training defaults

ChatGPT, Claude and Perplexity may use consumer chats for training unless you opt out, and the Gemini app does while Keep Activity is on, which it is by default for adults.8 Microsoft says the current Copilot app doesn’t use prompts, responses or files to train foundation models.

Each opt-out has limits. OpenAI may still train on a whole conversation if you rate a response with a thumbs up or down.5 Anthropic still uses conversations flagged for safety review, and material you report through feedback.10 With Gemini’s Keep Activity off, Google doesn’t use future chats to train its models unless you send feedback.7 Perplexity says opt-outs “only apply to data collected after the opt-out date”, and that “previously collected training data cannot be deleted or removed”.21 Microsoft’s FAQ for the older Copilot app describes training on conversations unless you opt out.19

Brello 1.0 has no account system, no analytics and no crash reporting, and no Brello server receives your questions, so we have no chats to train on.

Human review

Google, OpenAI and Microsoft describe people seeing some content, Anthropic describes safety review of flagged chats, and the Perplexity pages we checked don’t address it. Brello 1.0 has no server, so there is nothing to review.

Google says “a subset of chats are reviewed by human reviewers (including Google’s trained service providers)”, and advises: “Please don’t enter confidential information that you wouldn’t want a reviewer to see or Google to use to improve our services, including machine-learning technologies.”7 OpenAI says authorised staff and trusted service providers may access content for a limited set of reasons, including to improve model performance unless you have opted out, and asks you not to “enter sensitive information that you would not want reviewed or used”.2

Microsoft’s privacy statement says its processing of personal data “involves both automated and manual (human) methods”.15 Its FAQ for the older Copilot app says “some Copilot conversations are subject to both automated and human review”, and that “an opt-out of human review is not available”.19 Anthropic keeps the inputs and outputs of chats its automated systems flag for up to 2 years.13

Accounts

Brello 1.0 has no accounts. ChatGPT and Gemini document signed-out use, Claude may ask you to create an account, and the current Copilot app’s privacy pages assume a Microsoft account.

OpenAI explains how to turn off training “when using ChatGPT without signing in”.3 Google says that when you use Gemini signed out, it processes data such as your IP address and browser type under the Google Privacy Policy.7 Anthropic’s consumer terms say “we may ask you to create an Account”, with a minimum age of 18.11 Perplexity says accounts let you, among other things, “save preferences, and save query history”.20

Brello 1.0 has no sign-up, and its models download from Hugging Face without an account or token.

Offline use

Brello 1.0 works offline after a one-time model download. None of the other providers’ pages we checked describes offline use of their assistant apps.

Brello Core is a 977 MB download, Brello Vision 2.59 GB and Brello Pro 3.66 GB. After that, Brello works fully offline, and only the optional web search needs a connection. Android’s documentation says on-device generative AI such as Gemini Nano “enables offline functionality” for the apps that use it.9

Where Brello is weaker

Brello 1.0 is a prototype, and next to every assistant on this page it has smaller models and fewer features.

• Availability. Brello 1.0 (version 1.0.0, 4 October 2026) is on Google Play and the App Store.
• Phones only. It runs on iPhone and on 64-bit ARM phones with Android 8.0 or newer. There is no web or desktop app.
• Smaller models. On-device models are far smaller than frontier cloud models. They can be wrong, have a knowledge cutoff and are weaker at long or complex reasoning.
• A 4,096-token window. Each answer sees only the last six messages of the current chat, clipped.
• One photo per message. No files, documents or PDFs.
• No voice. No speech input or output.
• No sync. No cloud backup, export or multi-device sync; chats live only on one phone.
• A large first download. 977 MB to 3.66 GB, plus a speed cache of similar size, and up to a minute for a model’s first launch.

Sources and the dates we checked them

We opened each page ourselves on 5 October 2026. Where a page shows its own date, we give it too.

Facts about Brello describe Brello 1.0, version 1.0.0 (build 1), as of 5 October 2026. Product names are trademarks of their owners; Brello isn’t affiliated with them.