What Dia won in ZDNET’s test
ZDNET contributor Jack Wallen reports that a Dia Skill successfully summarized five open tabs, while his attempt to create an equivalent Skill in Gemini in Chrome produced a refusal claiming it could not access browser tabs. He also preferred Dia’s arrangement of separate AI conversations into tabs and pinnable groups.
That is a useful hands-on finding, but not a universal capability verdict. The review supplies no browser builds, model versions, controlled timings, or repeated trials. Wallen’s impression that Dia responded faster than Gemini in Chrome and Opera Neon should likewise be read as an observation, not a benchmark.
The distinction is narrower—and more interesting—than “Dia has Skills and Google does not.” Wallen actually credits Gemini with easier Skill creation and the ability to stack Skills. Dia’s reported strength is how its reusable instructions connect with multi-tab research.
For someone comparing documentation, release notes, and troubleshooting discussions, that integration could matter more than another clever answer in a sidebar. The useful question is whether the assistant consistently brings the right pages into the task.
Gemini can read multiple tabs, with explicit sharing
Google’s current Chrome Help documentation says Gemini can use multiple open tabs and compare or consolidate information across pages. Users can share up to 10 open tabs, including by entering @ in a prompt and selecting a tab. That does not establish that a saved Skill behaves exactly like Dia’s, but it rules out treating Wallen’s refusal message as proof that Gemini in Chrome cannot perform cross-tab analysis.
For eligible desktop users, Google documents this procedure:
- Open Ask Gemini in Chrome.
- In the prompt box, choose Add tabs.
- Select the pages to include.
- Ask Gemini to compare or summarize them.
Shared tabs display a check mark in the selection interface, and a tab shows a glowing underline when Gemini uses its page content. If the feature is missing, Google identifies account eligibility, region, language, browser version, and organizational settings as possible restrictions. Its supported desktop platforms are Windows, Mac, and Chromebook Plus—not every operating system on which Chrome runs.
The practical distinction: successful tab sharing and successful reusable automation are separate tests. An assistant’s refusal describes that interaction; it is not a substitute for checking the browser’s documented permissions and context controls.
Windows users no longer need to wait for Dia
The review’s description of Dia as Mac-only is outdated. The Browser Company’s official release notes document Dia for Windows 1.2.0 on October 15, 2026, explicitly adding Skills, including custom Skills, with an experience the company describes as matching macOS. The same release brought page attachments and mentioned tabs directly into chat context.
That establishes a concrete Windows connection to the feature under discussion. It does not independently reproduce Wallen’s five-tab result or prove identical behavior across platforms and accounts.
In Wallen’s macOS setup, he enabled Advanced Chat under Settings > Apps, restarted, and created a Skill through Settings > Skills > +. He then invoked it by typing / in a new chat. Those are his reported steps, not a verified current Windows setup procedure; the review does not identify the build used.
For a useful comparison of your own, start with public, nonsensitive pages and give both browsers the same task. Require one short summary per page, an explicit list of any unreadable pages, and a final comparison. Judge coverage and accuracy—not merely whether the answer sounds polished.
Privacy: “not used for training” needs qualification
The Browser Company’s security documentation supports the review’s claim that Dia’s external AI providers are contractually restricted from training on user data. However, it also says AI requests send the necessary prompt and context through its servers and Atlassian’s AI Gateway to those providers. This is cloud processing, not an entirely local assistant.
There is another important distinction: the company says it uses some content data by default to improve Dia and may use content to train small on-device models. It documents an opt-out at Settings > Privacy > Share content data to help improve Dia. Provider-training restrictions and the browser maker’s own improvement practices are different controls.
Dia’s privacy policy also says Skills data can include their prompts and descriptions, and that sharing to the Skills Gallery makes certain information public. Avoid putting confidential instructions, credentials, or customer details inside a reusable Skill intended for sharing.
The security documentation further says a Dia chat begins without access to other tabs and requires review before granting consequential capabilities. That qualifies the review’s broad suggestion that Skills automatically traverse everything: do not assume every open tab or connected tool is automatically available.
None of these facts alone establishes that Dia is categorically more secure than Perplexity Comet. A training-policy comparison is not a complete security assessment.
Pricing also changes the comparison
Dia’s official plans page differs from the review’s quoted pricing. It lists:
| Plan | Published monthly price | Main distinction |
|---|---|---|
| Better Browser | Free | Browsing without AI features |
| Better Answers | $10 | AI assistance, Morning Brief, and connected tools |
| Better Days | $100 | Everything in Better Answers, with 20 times more asks |
The same page says paid-plan rollout starts with most new Mac users, while existing users retain their current access for now and Windows users are not yet subject to that rollout. Those exceptions matter: the published subscription tiers should not be presented as an immediate charge applying to every Windows installation.
The real takeaway
Wallen’s review offers a compelling example of Dia making a five-tab task feel straightforward. It does not establish an enduring monopoly on multi-tab AI.
For Windows users, the stronger conclusion is that Dia’s custom Skills are already available to evaluate, while Gemini in Chrome has documented tab-sharing controls worth trying before switching browsers. The deciding factor should be repeatable results with the pages you actually use, clear visibility into what the assistant can access, and acceptable data-handling terms. A convincing summary is welcome; a convincing summary that quietly skipped two tabs is considerably less helpful.