Memo 2026: A message from Partoo’s co-CEOs
As every year, this memo invites us to pause and revisit the successes and lessons…
Table of contents
On 21 October 2025, the US company OpenAI officially launched its ChatGPT Atlas web browser. This browser is being lauded due to its native integration with ChatGPT, making it the first “AI-powered” browser.
While it may raise the browser stakes, the use of such tech, however, requires taking precautions and sets new requirements if local companies want to ensure a solid online presence. We’ll tell you all you need to know!
Developed by OpenAI, ChatGPT Atlas is a web browser that natively integrates ChatGPT’s AI, an LLM (Large Language Model) that forms its system’s core. It’s innovative tech that combines classic navigation (similar to Chrome or Firefox) and an AI agent into one tool. This AI integration allows the browser to, among other things, analyse the session context (history and open tabs) and provide the user with an ultra-personalised experience
Good to know
ChatGPT Atlas is currently only available on macOS, with Windows, iOS, and Android versions in the planning.
ChatGPT Atlas is currently only available on macOS, with Windows, iOS, and Android versions in the planning.
Visually, the interface is minimalist and similar to ChatGPT. It has all the usual elements found on a classic web browser: the address bar, tabs, windows, and a history. The homepage is simple, with an input bar where users can type in a URL or ask ChatGPT a question.
You can download ChatGPT Atlas for free, but must create an account to use the browser.
OpenAI also has a paid version where subscribers can access the “Agent Mode” feature.
The “Ask ChatGPT” sidebar is a button on web pages that opens a side panel where you can ask ChatGPT to do various things, such as summarise, translate, or explain the page’s contents. This means you can instruct the AI to undertake tasks without switching tabs.
This feature stores and uses past searches to personalise future user interactions. Atlas AI “understands” the context and can “remember” a user’s navigation to provide ultra-personalised responses. For example: looking for a pair of shoes? Atlas will remember the pair you looked at the previous day. Are you gluten intolerant? The AI will take factor this in when searching for a restaurant.
This is ChatGPT Atlas’s next-gen feature! Only available with the paid option, Agent Mode performs complex tasks directly on the navigation interface.
Unlike classic features, Agent Mode takes the wheel and will navigate for you. For example, the user will see the cursor:
At the same time, the AI will explain what it’s doing step by step . Whether browsing or buying, AI can handle all the steps of an online purchase.
ChatGPT Atlas must seek the user’s consent at each key step, such as selecting a size, adding it to the basket, and confirming the basket’s contents.
So, with Agent Mode, a user can ask the AI to book their hotel and plane tickets, order their nephew’s birthday present, or buy tickets to a gig without barely lifting a finger.
Good to know
Agent Mode is still in an experimental phase. OpenAI has also warned users that it may fail when working on overly complicated tasks.
Agent Mode is still in an experimental phase. OpenAI has also warned users that it may fail when working on overly complicated tasks.
With Agent Mode, ChatGPT Atlas has taken the lead in the ultra-competitive world of web browsers. But while it may be a super-smart breakthrough, it’s not without risks.
The full integration of AI into a web browser raises serious concerns about the safety of user data and customers’ information.
The arrival of AI has led to new types of cyberattacks. One of the most common and hardest to tackle is the prompt injection attack.
A prompt injection attack is a malicious attack that tricks the AI by tampering with its prompt instructions. The AI receives instructions to produce content that it should not normally generate.
For example, these malicious prompts can be hidden on a website’s page, often in data areas that users don’t see, but the AI will read (such as during a web browsing session).
This type of attack leans on the fact that AI doesn’t distinguish between instructions given by a user or a malicious source.
The results can be disastrous:
When launching ChatGPT Atlas, OpenAI warned that prompt injection cyberattacks were possible and that users should be extra-vigilant.
OpenAI stated that ChatGPT Atlas does not store sensitive data such as passwords or bank details. However, tests carried out by an international NGO that works to protect digital rights found that the AI browser kept traces of medical and personal searches.
And the reason is simple: ChatGPT has a hard time telling the difference between “classic data” and sensitive user data (personal, banking, healthcare information, etc.). To prevent this type of attack, users should stick to secure sites and never share sensitive data with AI browsers.
The arrival of ChatGPT Atlas has ramped up web browser competition, but its impact on local businesses’ online presence is not as trailblazing as one might expect.
Just like a classic web browser, ChatGPT Atlas crawls data found on the web to provide results aligned with users’ AI-enhanced search habits.
So the optimisation basics remain the same:
Download our free guide and discover how AI is reshaping local search, online reputation, and conversational commerce!
As every year, this memo invites us to pause and revisit the successes and lessons…
On 21 October 2025, the US company OpenAI officially launched its ChatGPT Atlas web browser….
On 3 November, 2025, Google Maps began to replace the Q&A section with Ask Maps,…
Receive our best articles and practical guides directly in your inbox every month