A People-First SEO/GEO View of Optimizing for Agents

I’m a people-first SEO/GEO practitioner.

People-first SEO means that all optimizations are done primarily for the benefit of users first with search engine visibility as a secondary benefit.

But as far as GEO is concerned, where does that leave us with AI visibility from agents?

It’s a good question: how does optimizing for agents fit into a people-first SEO/GEO approach?

The answer is that agents should get treated the same as human users, meaning optimizations should be done for agents as well.

What are Agents?

Agents are AI-powered software systems that can interpret a user’s goal, decide what steps to take, gather and evaluate information, use tools, and take actions on the user’s behalf.

Essentially, an AI agent is a digital representative of a user.

A person gives their agent an objective, and the agent may then search the web, retrieve information, compare options, reason about those options, interact with websites or APIs, and potentially take an action to accomplish that person’s goal.

You can build agents in AI systems like Claude Code or Codex from OpenAI, while platforms such as Anthropic’s Agent SDK let developers build custom agents.

A People-First Approach to Optimizing for Agents

Optimizing for agents doesn’t have to mean choosing machines over people. When agents act as representatives of people, making information easier for agents to understand and actions easier for them to complete can be another form of people-first optimization.

Here are tips for optimizing for agents, from a people-first SEO/GEO point of view:

Optimize around the user’s goal (not around the agent itself).

Since an agent is acting on behalf of a person, start with what that person is trying to achieve. They may be looking to learn something, compare products, book an appointment, get a quote, or purchase an item. Google describes agents as systems performing tasks on behalf of people and still recommends helpful, reliable, people-first content for AI-driven search experiences.

Make information explicit (rather than forcing inference).

This aligns with best practices for extractability, mainly the idea of clearly stating information with specificity (entity naming).

Instead of:

“It’s available there for $20.”

Write:

“The Acme Widget is available at the Manhattan store for $20.”

Details to include can be prices, availability, locations, features, restrictions, dates, eligibility requirements, or shipping information, basically any decision-making details.

Presenting information this way becomes important when an agent has to compare options or make a recommendation.

The same principle of reducing ambiguity is visible in machine-readable product data. OpenAI’s product-feed specification, for example, calls for structured attributes, such as title, description, availability, price, brand, links, and images, so ChatGPT can accurately understand and display products.

Use semantic HTML.

Google’s web.dev guide explains that browser agents may understand websites through screenshots, HTML/DOM structure, and the accessibility tree. That guide specifically recommends using semantic elements such as <button> and <a> rather than turning generic <div> or <span> elements into controls.

Good semantic HTML isn’t just an accessibility issue. It’s potentially an agent usability issue too.

Have a meaningful accessibility tree.

Google’s agent-friendly website guidance calls the accessibility tree a kind of semantic map of the page and recommends properly associating labels with form inputs. W3C likewise recommends explicit labels so user agents can determine the purpose of controls programmatically.

Proper labels, roles, names, and states in your accessibility tree give agents a clearer representation of what elements actually do.

From a people-first perspective, accessibility improvements simultaneously help humans, assistive technologies, and AI agents.

Make buttons and actions unambiguous.

Rather than using vague text like “Continue,” use descriptive controls, such as “Check Availability,” “Add to Cart,” “Schedule an Appointment,” or “Request a Quote” when the context isn’t obvious.

Google’s agent guidance recommends making necessary actions clearly reflected in the interface, while OpenAI confirms that ChatGPT agent can visually navigate pages, click buttons, and fill out forms.

Keep important user journeys predictable and stable.

This means avoiding interfaces where buttons constantly move, overlays that unexpectedly cover controls, or where the same action behaves differently across similar pages.

Google’s agent-friendly guidance warns that unstable layouts and transparent overlays can confuse agents using screenshots or visual analysis.

Think of this step almost as agent UX:

Human UX: “Can I figure out how to accomplish this?“

Agent UX: “Can my representative reliably figure out how to accomplish this for me?“

Review access for AI crawlers, retrieval systems, and agents.

Make sure your robots.txt isn’t inadvertently blocking AI search crawlers that agents may rely on to retrieve information.

If you want your pages discoverable through ChatGPT Search, for example, OpenAI recommends allowing OAI-SearchBot, which is separate from GPTBot (associated with model training). You can allow search discovery while still blocking GPTBot, for example.

Use structured data (realistically).

Structured data reduces ambiguity and describes information consistently, but adding schema doesn’t somehow guarantee agent visibility.

By making information machine-readable, structured data can help systems understand entities and attributes. That said, Google says there’s no special schema markup required for generative AI search in general and also warns against overfocusing on structured data as an AI hack.

Keep machine-readable information consistent with human-visible information.

If your page says a product costs $99 but your feed, API, or schema says $89, an agent has to reconcile conflicting evidence.

OpenAI’s agentic-commerce documentation emphasizes accurate, up-to-date product feeds so ChatGPT can understand current pricing and availability.

Data consistency extends beyond ecommerce, though. Hours, addresses, pricing, policies, inventory, specifications, event dates, and other attributes should also be consistent across your website and secondary data sources.

Provide agents with APIs or tools when the task goes beyond reading.

If you want agents to do more than read your webpages, consider exposing capabilities through APIs, MCP servers, or other agent-compatible interfaces. Having an MCP allows servers to expose tools that language models can discover and invoke for things like querying databases, calling APIs, or performing actions.

Think beyond webpages to product feeds and agentic commerce protocols.

This is specifically for ecommerce brands. OpenAI supports structured merchant product feeds that provide ChatGPT with current catalog information, such as product titles, descriptions, images, prices, and availability. (Direct feed onboarding is currently available to approved partners, while other merchants can apply for access.)

Takeaways

Taking a people-first approach to SEO/GEO doesn’t mean focusing only on humans. It also includes the agents that can perform tasks on their behalf.

Optimizing for agents is thus a smart addition to a holistic SEO/GEO strategy.

The shift toward an agentic web is already measurable. HUMAN Security observed traffic from AI agents and agentic browsers grow 7,851% year over year in 2025 across its customer dataset.

This shows that AI systems are increasingly becoming intermediaries between people and the web, and if we want to optimize for users, we need to include agents in that mix.

If you want a hand with agentic readiness, or SEO/GEO in general, I’m available as an independent consultant. Get in touch with me to learn more.

Until next time, enjoy the vibes:

Thanks for reading. Happy optimizing!

Related posts