Computer-Use Agents and SEO: What Changes for Organic Visibility

A computer-use agent is an AI model that can operate a computer or browser by looking at the screen and taking actions. Anthropic introduced computer use for Claude, OpenAI launched Operator and then built agent features into ChatGPT, and Google has shown Project Mariner and agent capabilities in Gemini. Browsers with built-in agents, such as Perplexity’s Comet, are another form.

The user gives a goal, such as “find a dentist near me who takes new patients on Saturdays and book the earliest slot”. The agent searches, visits sites, reads opening hours, opens booking forms and completes the steps, asking the user to confirm sensitive actions.

Agents mostly start from a search engine or their provider’s own index. That means traditional SEO still decides which sites enter the shortlist. If your page does not rank or is not cited for the query, the agent may never visit it.

What changes is the selection step. A person scans results and picks by instinct; an agent reads pages and checks facts against the goal. Pages that state clear, specific information, such as prices, availability, location and conditions, give the agent what it needs to choose.

Agents read pages through screenshots, the page’s accessibility tree, or both. Many of the practices that help screen reader users also help agents.

• Use real HTML buttons and links with clear labels, not clickable divs with icons only.
• Label every form field with a visible label element, not only placeholder text.
• Keep key information in text, not inside images or canvas elements.
• Avoid content that only appears on hover or after complex interactions.
• Make sure pop-ups and cookie banners can be closed with an obvious button.
• Keep layouts stable while loading, so elements do not shift as the agent clicks.

These overlap with Web Content Accessibility Guidelines and with Core Web Vitals such as Cumulative Layout Shift. Work that improves accessibility and stability helps people and agents alike.

Agents acting for a user often browse with a normal browser user agent, and some providers identify their agents in headers or user agent strings. AI crawlers that collect training or search data, such as GPTBot, ClaudeBot, Google-Extended or PerplexityBot, are controlled through robots.txt.

The decision depends on your business. A shop or booking service usually benefits from letting agents complete purchases. A publisher may want to limit some crawlers while staying visible in answer engines. Decide per crawler, document it, and review the choice as providers change their bots and policies.

Bot protection and CAPTCHAs

Aggressive bot protection may block legitimate agents along with scrapers. That can mean lost bookings or sales. At the same time, agents should not bypass security checks, and well-behaved agents hand control back to the user for CAPTCHAs. Review your bot rules with your security provider so real customer journeys are not broken.

Structured data gives machines facts in a predictable format. Product schema with price and availability, LocalBusiness with opening hours, Event with dates, and FAQ content in clear headings all help agents and search engines confirm what a page says. Keep markup consistent with visible content; Google’s guidelines treat mismatches as a violation.

Clear, current information in Google Business Profile, Merchant Center and major directories also matters, because agents often cross-check several sources before acting.

Measurement is still immature. Some agent visits look like normal browser sessions, while others are identifiable by user agent or IP range published by the provider. Start with what you can see.

• Check server logs for known AI agent and crawler user agents.
• Watch GA4 for referral traffic from chatgpt.com, perplexity.ai, gemini.google.com and similar sources.
• Track form completions and bookings with unusual patterns, such as very fast completion.
• Ask customers in booking or checkout forms how they found you, including an option for AI assistants.

Over time, providers may share more information about agent sessions. Set up tracking now so you have a baseline. Even rough numbers collected today will make later changes in agent behaviour much easier to read and explain to stakeholders.