Agentic AI in B2B: ChatGPT Atlas and Perplexity Comet, and what changes [2026]

Until now AI has been an adviser: it received a question and answered. The person made the decision and clicked. That is changing. ChatGPT Atlas from OpenAI and Perplexity Comet are web browsers with built-in AI agents that can visit sites, fill in forms, compare offers and carry out multi-step tasks without constant supervision. That is a turning point for B2B visibility.

Picture a scenario that is already possible: a procurement director at a manufacturer says to Perplexity Comet, "Find me the three best industrial automation suppliers, compare their pricing and send a request for quotation to each." Comet visits the sites, reads the content, compares - and contacts the suppliers. If your site does not meet the criteria the agent is looking for, there is no second chance. The agent simply moves to the next company.

What is agentic AI?

Agentic AI describes systems capable of planning and executing multi-step tasks autonomously to achieve a goal, without a human confirming each step. Unlike a standard chatbot answering one question at a time, an AI agent can:

• Break a complex task into sub-tasks and carry them out sequentially or in parallel
• Use tools: a browser, a calendar, email, spreadsheets, forms
• Retain context between sessions and continue a task after an interruption
• Make decisions from previously gathered data without further questions
• Delegate sub-tasks to specialised sub-agents

ChatGPT Atlas - what it does

ChatGPT Atlas is OpenAI's web browser with a built-in agent. It runs on a Chromium engine with full web access - it can sign into services with the user's permissions, fill in forms, click buttons, copy content and carry out instructions on web pages.

Example B2B uses already being tested: automatically comparing supplier prices, submitting requests for quotation through contact forms, searching tender databases, collecting contact details for outreach, and monitoring competitors' prices, offers and site changes.

Perplexity Comet - a research assistant with autonomy

Perplexity Comet takes a slightly different direction from Atlas. It positions itself as a browser with an AI research assistant. Its strength is integrating search with automatic summarising and action. Comet has been available free, which sharply lowers the barrier to entry.

From an AI visibility perspective, Comet matters because it builds on Perplexity, already a leading research tool; because when performing research tasks it cites and visits pages using criteria similar to standard Perplexity; and because free access means fast adoption, particularly among freelancers and small companies.

Atlas versus Comet for B2B

How agentic AI changes the B2B buying journey

The traditional B2B buying journey looked like this: an employee researches in Google, gathers links, opens tabs, compares offers manually and passes a shortlist to the decision-maker. Every stage is a point of contact with your brand - a chance to impress with design, sales responsiveness or case studies.

With agentic AI it looks different. The agent gathers information from sites automatically and aggregates it. The human receives a processed result: "here are the three best options, with reasoning". Points of contact with your brand shrink drastically. Instead of many touchpoints, there is one assessment by the agent. If your site did not give the agent the information it was looking for, in a clear and structured format, you drop off the list.

What an AI agent checks on your site

From tests with ChatGPT Atlas and Comet on B2B sites, it is possible to identify what agents check first:

• A legible offer on the home page - the agent has to understand within seconds what you do and for whom. Vague slogans without specifics lead to being skipped.
• Contact details visible without clicking - a contact form buried in the navigation is a barrier. Put an email address and phone number in the header or footer.
• Pricing or price ranges - B2B agents often look for price information to compare. A site with no pricing information at all is hard to assess.
• Case studies and references - agents look for evidence of experience. Specific project descriptions with the client's sector, the outcome and the scale.
• A contact form that works without JavaScript - agents do not always render JS. Provide an HTML fallback or a visible email address.
• Fast loading, under two seconds - agents have timeouts. A slow site gets skipped.
• No pop-ups blocking content - an agent cannot dismiss a cookie banner or modal as easily as a person. A blocking pop-up makes the page unreadable.

A new dimension: machine-readable content

Machine-readable content is an emerging standard in GEO and AI Visibility that takes on particular significance in the agentic era. It means preparing content not only for people and search engines but specifically for AI agents that must process a large volume of information quickly.

The key principles for B2B:

• Facts in the first sentences - do not build a narrative before giving the key information. The agent is looking for facts: what, for whom, how much, how it works.
• Data in tables rather than paragraphs - comparisons, technical specifications and summaries. An HTML table is easier to process than narrative text.
• Schema.org for every content type - Product, Service, Organization, Offer. The more structured data, the more the agent can absorb without parsing narrative.
• llms.txt - an emerging configuration file for AI agents, analogous to robots.txt for search engines. It points agents to the most relevant sections of the site.
• Unambiguous calls to action - "Request a quotation", "Download pricing", "Book a demo". The agent has to identify the right action to take.

When will agentic AI become standard in B2B?

Forecasts are difficult, but some signals are clear. ChatGPT Atlas was available in beta for Pro users from mid-2025. Perplexity Comet released a public free version in late 2025. According to Gartner, 30% of B2B transactions in the technology sector globally will be assisted by AI agents by the end of 2026.

Adoption will be slower in some markets than in the US or UK, but companies that prepare earlier will build an advantage that is hard to close. The pattern is always the same: SEO was ignored for five years by smaller firms - until it turned out a competitor had invested earlier and held a durable advantage. AI Visibility in 2026 is where SEO was in 2010.

Risks for companies ignoring agentic AI

Beyond the opportunity there is real risk. Companies that do not prepare their sites for interaction with agents may:

• Drop off agent-generated shortlists - if an agent cannot process your site in three seconds, it moves to the next one
• Lose the advantage of response speed - an agent can gather offers from ten suppliers within minutes. Firms with an unclear online offer get skipped even with a better product
• Be represented inaccurately - an agent may build a description of your company from incomplete data. If you do not control what the agent sees, you do not control how you are presented
• Lose contact-data visibility for AI prospecting - firms without structured contact details can be skipped in automated B2B outreach by agents

A preparation plan

Specific actions, prioritised by effort and effect:

1. 1Immediate (week 1): check robots.txt - unblock GPTBot, ClaudeBot and PerplexityBot. Add or update an llms.txt describing the key sections of the site.
2. 2Short term (month 1): rebuild the home page for clarity to agents - a clear service description in the first 100 words, visible contact details, and schema.org Organization and Service.
3. 3Medium term: build a dedicated section for AI agents - a machine-readable summary on every service page, tabular comparisons of the offer, and a contact form that works without JS.
4. 4Long term: monitor agent behaviour on your site in analytics. Test ChatGPT Atlas and Perplexity Comet regularly on representative category queries.