AAO (Assistive Agent Optimization) - When AI agents buy on your behalf

In February 2026, Jason Barnard - one of the world's leading experts on Knowledge Graphs and AI visibility - formalized in Search Engine Land the framework of Assistive Agent Optimization (AAO). His thesis: search optimization moves through four stages where each stage absorbs the previous one (Search Engine Land).

The word that changes is "engine" to "agent". It is not a semantic shift, it is a structural pivot. We move from systems that recommend to systems that act. AI no longer just lists options for a human to pick from. It selects, negotiates, and buys autonomously.

Barnard identifies an Algorithmic Trinity universal across all these systems: LLMs (language models), Knowledge Graphs, and traditional search engines. The balance varies by platform (ChatGPT leans LLM, Google leans Knowledge Graph), but the trinity is universal. Optimizing for AAO requires covering all three (Barnard, Search Engine Land).

The starting point of this optimization is the Entity Home, a central concept in Barnard's framework. Concretely, it is a canonical page on your site (often the "About" page or a dedicated page) that defines your brand identity unambiguously. It is the page that Google, ChatGPT, or Perplexity consult to answer the question "Who is this company and what do they do?". It directly feeds the Knowledge Graph and provides the foundation on which AI agents build trust. Without a clear Entity Home, an agent cannot identify you with certainty - and without identification, no selection.

The distinction between AEO/GEO and AAO is not just vocabulary. It redefines the metrics, the data required, and the very nature of the outcome.

Barnard models the full path between content and an AI conversion through a 10-gate pipeline (DSCRI-ARGDW).

• DSCRI (infrastructure): Discovered, Selected, Crawled, Rendered, Indexed. Is your content technically accessible to agents?
• ARGDW (evaluation): Annotated, Recruited, Grounded, Displayed, Won. Is your content judged trustworthy enough to act?

The key principle: trust is multiplicative, not additive. A drop at any gate compounds downstream. A slow site, JavaScript-heavy, with inconsistent data, reaches annotation with low trust even if the content is excellent. This is the model of Cascading Confidence (Barnard).

AAO is not a theoretical projection. The agentic commerce infrastructure is already in production at the three AI giants.

Google and the Universal Commerce Protocol (UCP)

Announced by Sundar Pichai at NRF in January 2026, UCP is an open standard for inter-agent commerce. Co-developed with Shopify, Etsy, Wayfair, Target, and Walmart, endorsed by 20+ partners including Visa, Mastercard, American Express, Stripe, and Adyen . It orchestrates three protocols: A2A (agent-to-agent), AP2 (agent payments), and MCP (model context). Concrete result: an AI agent can discover a product, negotiate a price, and finalize the purchase across platforms, without a human stepping in.(Google Blog).

In February 2026, Google completed the ecosystem with WebMCP, a browser API (navigator.modelContext) that lets sites publish a "Tool Contract" defining the available actions. Agents can call functions directly, for example buyTicket(destination, date).

OpenAI and Instant Checkout with Stripe

Deployed in late 2025, Instant Checkout lets ChatGPT users buy directly from the conversation. Merchant partners include Target, Instacart, DoorDash, and over a million Shopify merchants. In February 2026, ChatGPT reached 900 million weekly users and 50 million paying subscribers.

Amazon and large-scale agentic commerce

Amazon is moving on three fronts at once. Rufus AI (300 million users) assists product discovery. Alexa+ monitors prices and triggers automatic purchases when thresholds are hit. Buy for Me goes further, letting users buy from Amazon's competitors directly inside the Amazon app.

The shift is not a hypothesis. Five independent studies published between late 2025 and early 2026 converge on the same finding: B2B buyers already use AI as their first vendor research tool.

AI has surpassed Google as the entry point

• 47% of CIOs, CISOs, and CTOs initiate vendor research with AI assistants, versus 43% with Google Search (Treble/Censuswide, December 2025).
• 79% of software buyers say AI search has changed their buying process. 29% start with LLMs more often than with Google (G2 Buyer Behavior Report 2025).
• 2/3 of B2B buyers use generative AI as much or more than traditional search. 94% use AI tools during the buying process (Responsive, 2025).

Source: Treble/Censuswide (MarTech Cube) | G2 Buyer Behavior Report

The pre-contact favorite wins 80% of deals

6sense research reveals a phenomenon decisive for AAO. The favored vendor identified before any human contact wins roughly 80% of deals. 95% of winning vendors are already on the "Day One" shortlist. Buyers fill 3.6 of the ~5 spots in their shortlist before any sales contact (6sense, B2B Buyer Experience Report).

What this means for AAO: when AI agents build these shortlists on behalf of humans, not being in the agent's data is equivalent to not existing. There will be no "second chance" in a sales call.

The trust paradox

Forrester (State of Business Buying 2026) observes that generative AI tools are the most cited type of interaction for purchase research. But 20% of buyers feel less confident due to inaccurate AI information, while 36% feel more confident thanks to AI. This paradox pushes buyers to turn to peers, experts, and vendors to confirm AI results, reinforcing the importance of third-party corroboration (Forrester, State of Business Buying 2026).

In parallel, 41% of consumers trust AI results more than traditional advertising (BoF-McKinsey, State of Fashion 2026). Trust shifts from marketing channels to algorithmic recommendations, exactly what AAO optimizes.

One of the most dangerous misconceptions about AI is that rankings are stable. Rand Fishkin (SparkToro) research published in February 2026 proves the opposite, and the consequences for AAO are major.

The inconsistency of AI recommendations

Fishkin ran 2,961 prompts on ChatGPT, Claude, and Google AI, across 12 categories. Result: there is less than 1 chance in 100 that two runs produce the same list of brands, and less than 1 chance in 1,000 that they appear in the same order (SparkToro).

Fishkin's conclusion: "ranking position" in AI is an absurd indicator. Only the visibility frequency (the percentage of times a brand appears across multiple runs) is statistically significant.

Concentration accelerates exponentially

Paradoxically, this inconsistency comes with brutal concentration. Authoritas data (Laurence O'Toole), analyzed by Barnard, covers 143 digital marketing experts (World Content Strategy study). Result: the top 10 experts went from 30.9% citability in December 2025 to 59.5% in February 2026. The HHI index (concentration measure) grew by 293% in less than two months (Barnard, Search Engine Land).

Winners are winning faster and faster. Brands that appear frequently in AI responses build an algorithmic trust moat that laggards cannot close. In AAO, where the agent only selects one brand, this winner-takes-all effect is amplified to the maximum.

Barnard models the full journey of content toward an agentic action across 10 gates. Each gate is a trust filter. A failure at any gate eliminates the content for every gate that follows.

DSCRI - Technical infrastructure

• Discovered. Is your content discovered by AI crawlers? (GPTBot, ClaudeBot, PerplexityBot allowed in robots.txt)
• Selected. Among the discovered pages, is your page selected for a deep crawl?
• Crawled. Can the content be downloaded and parsed? (no JavaScript-only, no opaque paywall)
• Rendered. Is the content understandable once parsed? (semantic HTML, schema markup)
• Indexed. Is the content stored in the agent's knowledge base?

ARGDW - Evaluation and action

• Annotated. Is the content enriched with metadata (entities, relations, categories)?
• Recruited. Is the content recalled to answer a relevant query?
• Grounded. Is the content used as a factual source in the answer?
• Displayed. Is the citation or recommendation actually shown to the user or the agent?
• Won. Is your brand selected for the final action (purchase, booking, contact)?

In GEO, the goal stops at Displayed. In AAO, you have to reach Won, which requires structured data complete enough for an agent to make an action decision without human intervention (Barnard, Search Engine Land).

AAO adds specific technical requirements to SEO and GEO fundamentals. Here are the six pillars to implement.

1. Entity Home and Knowledge Graph

Your brand needs an Entity Home, a canonical page that defines your identity unambiguously. Barnard considers it the highest-ROI and fastest intervention. It must include:

• Disambiguated identity - who you are, what you do, for whom, and how you are different
• Consistent entity data across all sources (site, social media, directories, Wikipedia)
• Third-party corroboration from independent sources
• Historical reliability, a track record of consistent content over time

2. Schema Markup and structured data

According to a Stackmatix data aggregation confirmed by Schema App, content with appropriate schema is 2.5x more likely to appear in AI responses. The JSON-LD format is preferred by every major AI system.

• Critical schemas: Product, Offer, Organization, LocalBusiness, FAQ, HowTo, BreadcrumbList
• Complete Product schema with name, brand, price, availability, GTIN, reviews, features. ISO 4217 currency codes.
• Shipping and return policies in structured format
• AggregateRating matching the visible content on the page
• Variant-level differentiation for configurable products

3. llms.txt

The llms.txt file is a plain-text Markdown file at the root of your domain. It provides a curated map of your most important content for LLMs and reduces token usage by serving clean content versus HTML parsing. Adoption is growing: Yoast, Webflow, Anthropic (Claude), and Shopify (built into every store by default) already implement it.

4. Third-party corroboration

The SE Ranking study on 129,000 unique domains confirms that referring domains (backlinks) are the most powerful citation predictor. Sites with 32,000+ referring domains are 3.5x more likely to be cited by AI. Agents trust corroborated signals more than self-proclaimed authority. The most effective corroboration sources remain Wikipedia, G2 reviews, sector reports, and Reddit threads (Search Engine Journal).

5. Actionable interfaces

This is the AAO-specific pillar versus GEO. An agent that wants to act needs transactional data.

• Exposed APIs for agent interaction (catalog, prices, availability)
• Booking calendars and demo request endpoints
• WebMCP "Tool Contracts" defining the actions available on your site
• Machine-readable purchase flows (not just HTML forms)

In AAO, transactional interfaces are as important as descriptive content.

6. Crawlability for AI bots

Critical point: most AI agent bots do not render JavaScript. Content behind client-side rendering is invisible to many agents. Favor server-side rendering or static generation. Allow GPTBot, ClaudeBot, and PerplexityBot in your robots.txt. GPTBot requests grew by +305% in 2025 and AI crawlers generated more than 50% of web traffic in 2025 according to Cloudflare and Vercel reports.

The evolution from AEO to AAO reinforces the need for a platform that does not just measure but guides action. Each Hikoo module corresponds to an AAO pillar.

• Analyzer for the baseline audit. Entity clarity, schema completeness, AI bot crawlability, machine readability score. It is the diagnosis of the first 5 gates of the pipeline (DSCRI).
• Spotlight for Share of AI Voice tracking. Not position, but visibility frequency across prompt runs. Monitoring of citability concentration vs competitors, on ChatGPT, Claude, Gemini, Perplexity, and Mistral simultaneously.
• Elevate for concrete optimization. Entity Home, schema markup, third-party corroboration, llms.txt creation, prioritized recommendations to clear the 10 gates of the pipeline.
• Battlemap for competitive intelligence. Who is winning the AAO race, which competitors appear in AI responses, their schema completeness, their entity strength.

Monitoring alone is a thermometer. AAO needs a doctor - someone who diagnoses, prescribes, and supports execution. That is the difference between tracking symptoms and treating causes.

The shift from "being cited" to "being chosen" is the most structural change in digital marketing since the migration from print to web. AAO is not a trend to watch, it is an operational reality with infrastructure in production (UCP, Instant Checkout, WebMCP), B2B buyers who have already migrated (47% start with AI), and a concentration effect that is accelerating (+293% in 60 days).

The window to build algorithmic trust is now. Every week of delay widens the gap with brands that already occupy the space. The Cascading Confidence model is unforgiving: trust accumulated at the early gates of the pipeline amplifies the results at the next ones. A well-structured, fast, semantically clean, third-party corroborated site reaches the final selection with a compound advantage that laggards cannot quickly replicate.

At Hikoo, we support companies through this transition from AI visibility to agentic selection. Analyzer diagnoses, Spotlight measures, Elevate optimizes, Battlemap watches the competition. Because in a world where the agent only chooses one brand, second place does not exist.