If you feel like the SEO world is shifting under your feet, you’re not imagining things. The rise of AI-powered search – think Google’s AI overviews in Search Generative Experience, ChatGPT’s answers, or Perplexity’s cited responses – is altering how people find information.
Instead of ten blue links, users now get conversational answers with citations and context. This is where Generative Engine Optimization (GEO) comes in.
GEO means optimizing your content so that AI engines (ChatGPT, Claude, Google’s Gemini, Perplexity, etc.) include your brand’s insights in their answers
In essence, it’s the process of making your digital content visible and valuable to generative AI when people ask about topics related to your expertise.
And just like with traditional SEO, the goal is clear: if the AI consistently sources you in its answers, your brand stays top-of-mind (and yes, that can lead to more clicks and engagement).
We’ve seen this trend already.
Google’s own data shows that when an AI Overview in Search includes links, those links get more clicks than a regular search listing
Users get a quick answer and still visit the source for depth – a double win. As a marketer, I find that encouraging: AI isn’t stealing our traffic outright; it’s reshuffling it.
The task for us now is to ensure our content is the one being quoted, cited, or used by these AI assistants.
It’s a bit like the early days of featured snippets – except the “snippet” is an AI-generated paragraph and we need to train the AI what to say.
This is the heart of GEO. But how exactly do we “train” or influence AI models to prefer our content?
Enter Google’s new Agent2Agent protocol, a tech innovation that might just become SEO’s new secret weapon in the era of AI.
Meet Google’s Agent2Agent (A2A): A Glimpse into SEO’s Future
Leave it to Google to drop a game-changer for AI interoperability in April 2025
Agent2Agent (A2A) is an open protocol designed for AI agents to communicate with each other seamlessly, almost like a universal language for bots.
Google launched A2A with over 50 tech partners on board – everyone from big enterprise software companies to AI framework developers – signaling that this could be the foundation for a whole new ecosystem of connected AI.
So, what does that mean in plain English for us SEO and marketing folks?
Think of A2A as a set of rules that lets one AI system talk to another.
For example, one AI “agent” could be Google’s search AI trying to answer a question, and another could be an AI that your company built to provide details about your products.
A2A would be the protocol that lets Google’s AI ask your AI for the info – and get a straight answer.
No scraping web pages, no hoping it interpreted your content correctly.
It’s a direct line to the source. And it all starts with something called the Agent Card.
The Agent Card: Your AI Business Card
At the heart of A2A is the concept of an Agent Card – basically a JSON file that serves as a digital business card for an AI agent
In developer terms, it’s a public metadata file (usually hosted at /.well-known/agent.json on your server) describing the agent’s capabilities, skills, identity, and how to talk to it.
In marketer terms, it’s like a rich, structured profile of what your brand’s AI assistant can do. It might include things like:
Description: What information or tasks your agent can handle (e.g., “This agent provides up-to-date details on Acme Corp’s products, pricing, and store locations”).
Capabilities: The specific knowledge or services it offers (maybe “can answer FAQs,” “can process orders,” etc.).
Endpoints: How to contact it (an API URL where the AI agent lives).
Authentication: Any keys or permissions needed to use it.
Think of it this way: if today’s SEO relies on HTML tags and schema markup to tell Google what’s on your site, the Agent Card is a future way to tell AI agents what your digital assistant (or data source) can provide.
It’s a one-stop JSON snapshot of your brand’s knowledge and capabilities.
Google’s A2A announcement suggests that a client AI (like a search engine AI) can fetch this Agent Card to discover what a remote agent (like your brand’s AI) knows or can do.
In other words, if Google’s AI is trying to answer a user query, it could look at Agent Cards to find the best source of info. This could be huge for SEO: rather than crawling dozens of pages to piece together an answer, the AI may go straight to the most relevant agent.
If your Agent Card is well-optimized and up-to-date, you might be the chosen one for that answer. In a sense, your Agent Card becomes the “source of truth” for that topic, as far as the AI is concerned.
Google isn’t doing this out of charity to webmasters. Of course, they smell money and opportunity.
It’s solving a problem: AI agents need accurate, structured info to be reliable.
From my perspective, though, this is a golden opportunity. We always talk about “structured data” in SEO – adding schema.org markup for recipes, products, etc.
A2A takes that to the next level. It’s like Google saying: “Give us an official JSON endpoint for your knowledge, and our AI will use it.”
As an SEO old-timer, I can’t help but see parallels to the advent of XML sitemaps or RSS feeds – channels where we directly feed search engines our content in a format easy for them to digest.
The Agent Card could become the go-to feed for AI models seeking factual, source-approved information.
How Agent Cards Could Become an AI “Source of Truth”
Let’s drill down on that “source of truth” idea.
Right now, large language models (LLMs) like GPT-4 are trained on vast swaths of the internet.
They know a lot, but they’re not always up-to-date or accurate on specific, niche, or real-time info.
That’s why we see efforts like Bing’s and Google’s AI results citing recent sources – the LLMs are being augmented with real-time search data. If you’ve used tools like Bing Chat or Perplexity, you’ve noticed they always show footnotes.
Those are essentially the AI saying, “Here’s where I got this info.” Now, imagine instead of trawling the web, the AI could just call an expert agent for the info.
This is where A2A shines. The Agent Card tells the AI what an agent knows.
If your company’s agent has an entry that it knows “current stock levels and prices for Product X,” a search AI could see that in the Agent Card and decide, “Great, that’s my best source.”
It’s a bit like a librarian calling up an expert on the phone to answer a question, rather than pulling books off the shelf at random.
Here’s a hypothetical scenario. It’s not far-fetched – it’s where all this could be heading in a year or two:
Use Case: A Day in the Life of an AI-Assisted Search
A user asks, “What’s the best electric car under $40k I can buy right now?”Google’s AI (powered by Gemini) breaks this down. It might consult a few A2A agents:
Step 1: It pings Tesla’s public agent (because Tesla’s Agent Card says it can answer detailed questions about models, pricing, and availability) and gets data on the Model 3.
Step 2: It pings a Ford agent for info on the Mustang Mach-E, a Chevrolet agent for the Bolt, etc. Each of these agents responds via the A2A protocol with the latest specs and prices (maybe even inventory).
Step 3: It might also ping a trusted automotive review agent (say, Car&Driver’s agent or a Consumer Reports agent) for an expert rating or safety info.
Step 4: Now Google’s AI has a trove of fresh, authoritative data: straight from the manufacturers and experts. It composes an overview: “Based on the latest data, the Tesla Model 3 (starting at $35k) and the Ford Mustang Mach-E (around $38k) are top contenders, with Tesla offering longer range and Ford boasting more cargo space. (Data via Tesla & Ford agents.)” It cites the sources (perhaps linking to the Agent Cards or the websites). The user gets a rich answer, and can click through to those carmakers for details.
In that scenario, notice how the brands ensured they were part of the conversation: they maintained A2A agents with good Agent Cards. The AI knew exactly who to “ask” for car prices because the Agent Cards advertised that capability
It’s not hard to imagine that in the near future, having an Agent Card could be as critical as having a website indexed on Google is today.
No Agent Card might mean the AI overlooks you as a source, the same way a site with poor SEO might not rank on page 1 today.
From an optimization standpoint, this is both exciting and challenging. On one hand, it simplifies things: ensure your Agent Card (your AI’s “profile”) is complete and accurate, and your info can be pulled into many experiences.
On the other hand, it introduces new things to optimize. For instance, the fields in the Agent Card (like name and description) could influence whether the AI picks your agent.
In fact, early security research on A2A noted that the “name” and “description” fields in the agent card will carry the most weight to the host agent LLM when it’s deciding which agent to use
Sounds a bit like optimizing meta titles and descriptions, doesn’t it?
The difference is, instead of enticing a human click, you’re enticing an AI to “choose” your data over others. In practical terms, Agent Card Optimization might become a new subset of GEO.
We’d be thinking about questions like: Do we clearly state our agent’s expertise in the description?
Are we listing all the relevant capabilities (so we don’t miss out on queries where we could have been the answer)?
Is our Agent Card accessible (at the standard /.well-known/agent.json location) for any AI to fetch?
This is akin to ensuring our XML sitemap is in place and our schema markup is correct – it’s just a more dynamic, interactive version of it.
One Schema to Rule Them All: Consistency Across AI Platforms
One of the most compelling promises of A2A is the “write once, serve all” potential for content providers.
As a brand or publisher, you might only need to maintain one well-crafted schema of your info (the Agent Card + the underlying agent service), and multiple AI platforms could tap into it.
Let’s compare this to what we do today:
Current state (2023-2024)
You create content on your website for Google’s index. Maybe you also structure some of it in schema.org for things like FAQs or how-tos to get rich results.
Separately, you might create a chatbot on your site, or feed data to a voice assistant, or build a plugin for ChatGPT, etc. It’s fragmented and repetitive.
You’re basically trying to be present wherever the user might search – Google, Bing, voice assistants, forum answers, you name it – each often requiring a different format or strategy.
Future state (with A2A and GEO)
You maintain an Agent Card and an agent service that knows your content. Now any AI-powered platform can discover and interact with it.
If a user is on Bing Chat and asks about your product, Bing’s AI could find your Agent Card and query your agent for the answer.
If a user is on Google, Google’s AI could do the same. If they’re using a personal assistant in their car, same thing.
In theory, you’ve centralized your “knowledge feed” in a way that all these channels can use.
This is a bit utopian – there will still be unique nuances per platform – but the efficiency gains are hard to ignore.
For marketers, this could reduce the need to re-optimize the same info for different AI systems. Your Agent Card becomes the canonical source of truth across platforms.
It’s like having a single, canonical URL that everyone respects, instead of scraping content and possibly getting it wrong.
Example: The Restaurant Chain Case Study
To make this less abstract, consider a mid-sized restaurant chain, “Tasty Eats,” in the near future.
Tasty Eats maintains an A2A agent that provides information on its locations, hours, menu specials, and even accepts reservations (through an API).
The agent’s card (tastyeats.com/.well-known/agent.json) clearly outlines these capabilities in the description and skills list.
Now see how it plays out:
On Google Search
A user searches, “Is there a Tasty Eats open near me right now?” Google’s AI overview checks for a Tasty Eats agent. It finds the Agent Card, sees a capability like “storeLocator” or a skill “provide opening hours”.
It calls the agent with the user’s location and gets an answer: “Yes, the Tasty Eats on 5th Street is open until 10 PM” – which it displays to the user, citing Tasty Eats as the source.
On ChatGPT
Someone asks ChatGPT (connected to the internet), “What’s the special at Tasty Eats this week?” ChatGPT, via A2A or a plugin, fetches the Agent Card, and sees that this agent can answer menu queries.
It asks the agent, gets the latest special (say, “2-for-1 taco Tuesdays”), and includes that in its answer.
It might respond with something conversational but accurate like, “Tasty Eats is running a 2-for-1 Taco Tuesday deal this week (source: Tasty Eats official agent).”
On a voice assistant in a car
The driver asks, “Navigate to the nearest Tasty Eats that’s open.” The assistant uses the same agent to find the nearest open location and even get a menu teaser, all consistent with what the website and Google Search are saying.
In all these cases, Tasty Eats didn’t have to optimize three different pieces of content or rely on web crawlers guessing the right answer from its site.
By maintaining one well-structured agent schema, it served all platforms. From an operational standpoint, that’s a big deal.
Marketers could spend less time worrying about every new AI app that comes out, and more time just feeding correct information into their agent.
Of course, competition doesn’t disappear – you still need to be the preferred source. But at least you won’t be omitted just because you didn’t code a custom skill for Alexa or missed some new markup for Google.
A2A aims to level that playing field by standardizing how AIs retrieve info.
What This Means for SEO Agencies and CMOs
The rise of A2A isn’t just a technical evolution — it’s a strategic one. SEO agencies now have an opportunity to build client visibility not just for SERPs, but for AI conversations.
CMOs, meanwhile, should see Agent Cards as a brand’s official voice in AI-driven discovery. Whether it’s product info, FAQs, or expert positioning — feeding structured, real-time data directly into AI ecosystems will soon be a marketing baseline.
It won’t be long before SEO audits start with: “Where’s your Agent Card?” And in a landscape ruled by generative AI, brands that speak directly to the machines — not just to the index — will lead the narrative. Those who get A2A-ready today won’t just be found. They’ll be chosen.
Dileep Thekkethil
AuthorDileep Thekkethil is the Director of Marketing at Stan Ventures, where he applies over 15 years of SEO and digital marketing expertise to drive growth and authority. A former journalist with six years of experience, he combines strategic storytelling with technical know-how to help brands navigate the shift toward AI-driven search and generative engines. Dileep is a strong advocate for Google’s EEAT standards, regularly sharing real-world use cases and scenarios to demystify complex marketing trends. He is an avid gardener of tropical fruits, a motor enthusiast, and a dedicated caretaker of his pair of cockatiels.