How Agents Evaluate

Each AI platform uses a different approach to discover, analyze, and recommend products and services. Understanding these differences is critical to optimizing your presence across agents.

This page covers the seven major platforms: what signals they prioritize, how they discover your content, and which content formats matter most.

ChatGPT (OpenAI)

Primary behavior: Browses websites directly, reads full page content, evaluates structured data, and synthesizes recommendations from multiple sources.

Discovery

ChatGPT discovers your content through two mechanisms:

GPTBot crawler — indexes pages for the ChatGPT training pipeline and retrieval system. Respects robots.txt directives under the GPTBot user agent.
ChatGPT Browse — a real-time browsing agent that visits your site during a user’s conversation. Renders JavaScript, reads visible content, and extracts structured data.

Signals It Prioritizes

Signal	Weight	Notes
Product page content	High	Clear descriptions, specifications, and pricing
Structured data (JSON-LD)	High	`Product`, `Review`, `FAQPage`, `HowTo` schemas
`llms.txt`	Medium	Used for site-level understanding
Customer reviews	Medium	Aggregated ratings and review text
Page load performance	Low	Impacts Browse agent’s ability to render content

Content Format That Matters Most

Structured product pages with JSON-LD markup. ChatGPT’s Browse agent extracts structured data first, then falls back to page content. Pages with complete Product schema (including offers, aggregateRating, and review) are consistently favored in recommendations.

Commerce Integration

ChatGPT powers Instant Checkout with over 1 million Shopify merchants. When a user asks ChatGPT to buy something, it can present products with a one-click purchase button. This flow relies on the Shopify Agentic API and ACP (Agent Commerce Protocol).

User: "Buy me a portable charger under $40"
  → ChatGPT Browse fetches product pages
  → Evaluates structured data + reviews
  → Presents top options with Instant Checkout buttons

Inception Agents optimization: Serves enriched structured data to GPTBot and ChatGPT Browse. Generates feature-led and concise-led content variants. Ensures ACP compatibility for commerce sites.

Claude (Anthropic)

Primary behavior: Performs deep document analysis, evaluates long-form content, and excels at comparative research. Heavily used for B2B/SaaS evaluation and technical purchasing decisions.

Discovery

ClaudeBot crawler — indexes pages for retrieval-augmented generation. Respects robots.txt under the ClaudeBot (or anthropic-ai) user agent.
Tool use / web search — Claude can search the web and read pages during conversations when users enable web access.

Signals It Prioritizes

Signal	Weight	Notes
Long-form documentation	High	Whitepapers, comparison guides, technical docs
Detailed product comparisons	High	Head-to-head feature analysis with evidence
`llms.txt` / `llms-full.txt`	High	Claude reads and processes these thoroughly
Pricing transparency	Medium	Clear pricing pages with tier breakdowns
Third-party reviews and analysis	Medium	Independent evaluations carry significant weight

Content Format That Matters Most

Detailed, evidence-based documentation. Claude processes long documents effectively and rewards depth. Comparison pages that methodically evaluate alternatives — with specific data points, not just marketing claims — perform well. Technical documentation, API references, and architecture overviews are strong signals for B2B products.

Typical Use Case

User: "Compare Datadog vs. New Relic for a 50-person engineering team"
  → Claude searches for comparison content and documentation
  → Reads pricing pages, feature matrices, and case studies
  → Synthesizes a detailed recommendation with trade-offs

Inception Agents optimization: Serves comparison-led and trust-led content variants to ClaudeBot. Generates comprehensive llms-full.txt for deep analysis. Enriches pages with evidence-backed claims and structured feature data.

Perplexity

Primary behavior: Real-time web search combined with LLM synthesis. Fetches multiple sources, extracts key information, and generates cited answers. Increasingly dominant in consumer product research and shopping.

Discovery

PerplexityBot crawler — indexes content for Perplexity’s search index. Respects robots.txt under the PerplexityBot user agent.
Real-time search — during a user query, Perplexity fetches and reads live web pages, extracting content for its synthesized answer.

Signals It Prioritizes

Signal	Weight	Notes
Citable facts and figures	High	Specific numbers, stats, and claims that Perplexity can attribute
Recency	High	Fresh content is strongly preferred over stale pages
Clear page structure	High	Well-organized headings, lists, and tables
Product availability and pricing	Medium	Current stock status and pricing data
Author and source credibility	Medium	Bylines, publication reputation, domain authority

Content Format That Matters Most

Citation-friendly, fact-dense content. Perplexity’s answers include numbered citations linking back to sources. Content that contains specific, extractable facts — pricing, performance benchmarks, feature lists, release dates — is more likely to be cited. Clear heading hierarchy and scannable formatting help Perplexity parse your pages accurately.

Audience

Perplexity reports that 92% of consumer prompts on its platform trigger shopping-related results. Its user base skews high-income, with 65% of users earning over $100K annually. This makes Perplexity particularly valuable for premium and considered-purchase categories.

User: "Best noise-cancelling headphones for flights"
  → Perplexity searches multiple sources in real time
  → Extracts product specs, prices, and expert opinions
  → Returns a synthesized answer with numbered citations

Inception Agents optimization: Serves concise-led and feature-led variants to PerplexityBot. Ensures pages contain citable data points. Adds structured pricing and availability data. Prioritizes content freshness signals.

Google Gemini

Primary behavior: Evaluates structured data first, then page content. Powers AI-driven shopping experiences across the Google ecosystem — Google Search (AI Overviews), Google Shopping, and Google Assistant.

Discovery

Google-Extended crawler — a dedicated crawler for Gemini’s training and retrieval pipeline, separate from Googlebot. Controlled via robots.txt under the Google-Extended user agent.
Googlebot — standard web crawler that feeds into both traditional search and AI features.
Google Merchant Center — product feed ingestion for Shopping results and AI-powered product recommendations.

Signals It Prioritizes

Signal	Weight	Notes
JSON-LD structured data	Very High	`Product`, `Organization`, `LocalBusiness`, `FAQPage`
Schema.org completeness	Very High	Entity relationships, `sameAs`, `isRelatedTo`
Google Merchant Center feed	High	Product availability, pricing, shipping, and promotions
Page experience signals	Medium	Core Web Vitals, mobile-friendliness, HTTPS
Reviews and ratings	Medium	Both on-site and Google Business Profile reviews

Content Format That Matters Most

Schema.org structured data with full entity relationships. Gemini’s evaluation begins with structured data. Sites with comprehensive JSON-LD markup — covering products, organizations, FAQs, and inter-entity relationships — have a measurable advantage. If your structured data is incomplete, Gemini may not evaluate your content as deeply as competitors with full markup.

Google Ecosystem Integration

Gemini powers AI features across multiple Google surfaces:

AI Overviews in Google Search — summarized answers that can include product recommendations
Google Shopping — AI-curated product selections with visual comparison
Google Assistant — voice-based product research and purchasing

User (Google Search): "best standing desk for home office"
  → Gemini evaluates structured data from top-ranking pages
  → Cross-references Google Shopping product feeds
  → Generates AI Overview with product recommendations

Inception Agents optimization: Generates comprehensive JSON-LD markup across all page types. Fills structured data gaps identified during content analysis. Ensures entity relationships are properly declared. Syncs with Google Merchant Center feed where applicable.

Microsoft Copilot

Primary behavior: Accesses product catalogs via APIs and structured feeds. Integrates across the Microsoft ecosystem — Bing, Edge, Windows, and Microsoft 365. Known for driving shorter buyer journeys and higher conversion rates.

Discovery

BingBot crawler — indexes pages for Bing search and Copilot’s retrieval system. Respects robots.txt under the bingbot user agent.
Microsoft Copilot Catalog — a product catalog API that enables direct product listing and purchase through Copilot.
Bing Shopping feed — product data ingestion for shopping-related Copilot queries.

Signals It Prioritizes

Signal	Weight	Notes
Product catalog API access	Very High	Copilot Catalog and Bing Shopping feeds
Structured product data	High	Price, availability, shipping, variants
Transactional capability	High	Ability to complete purchases in the agent flow
Brand authority	Medium	Established brands with consistent web presence
Review aggregation	Medium	Third-party review signals from Bing’s index

Content Format That Matters Most

API-accessible product catalogs. Copilot’s strength is in transactional interactions. It accesses product data via APIs and structured feeds rather than crawling page content. Sites integrated with the Microsoft Copilot Catalog or Bing Shopping feed receive direct product placements in Copilot responses.

Performance Data

Microsoft reports that Copilot-assisted shopping produces:

33% shorter buyer journeys — fewer steps from search to purchase
53% more purchases completed within 30 minutes of initial query
73% higher click-through rates compared to traditional search results

User (Copilot): "Find me a 4K monitor under $500 with USB-C"
  → Copilot queries product catalog APIs
  → Filters by price, specs, and availability
  → Presents matching products with purchase links

Inception Agents optimization: Generates Copilot Catalog-compatible product feeds. Enriches product data with variant information, shipping details, and promotional pricing. Ensures transactional readiness through UCP integration.

Amazon Rufus

Primary behavior: An AI shopping assistant embedded within the Amazon ecosystem. Answers product questions, makes recommendations, and guides purchasing decisions for Amazon’s 300M+ active customers.

Discovery

Rufus primarily evaluates content within Amazon’s ecosystem:

Amazon product listings — title, bullet points, A+ Content, backend keywords
Customer reviews and Q&A — review text, ratings, answered questions
Brand stores — Amazon Brand Store pages and content
External signals — limited off-Amazon content evaluation for brand context

Signals It Prioritizes

Signal	Weight	Notes
Product listing quality	Very High	Title optimization, bullet points, A+ Content
Customer reviews	Very High	Volume, recency, sentiment, and specificity of reviews
Q&A completeness	High	Answered questions on the product listing
Sales velocity	High	Recent sales volume and conversion rate
Competitive positioning	Medium	Price competitiveness, category ranking

Content Format That Matters Most

Optimized Amazon product listings. Rufus operates almost entirely within Amazon’s data. Off-Amazon content has limited direct impact, but your llms.txt and structured data can influence how Rufus understands your brand in a broader context. The primary optimization lever is listing quality — clear titles, detailed bullet points, complete A+ Content, and a strong review profile.

Key Consideration

Amazon reports that users engaging with Rufus are 60% more likely to complete a purchase. However, internal data suggests that approximately 83% of Rufus recommendations favor Amazon-sold or Amazon-fulfilled products. For third-party sellers, strong listing optimization and FBA (Fulfillment by Amazon) enrollment significantly improve recommendation likelihood.

User (Amazon): "What's a good tent for family camping?"
  → Rufus evaluates product listings, reviews, and Q&A
  → Weighs review sentiment, sales data, and listing quality
  → Recommends products with conversational explanations

Inception Agents optimization: Generates Amazon-optimized content recommendations. Identifies listing gaps (missing A+ Content, thin bullet points, unanswered questions). Monitors competitive positioning signals. For DTC brands, ensures off-Amazon content reinforces brand authority for Rufus’s external signal processing.

Grok (xAI)

Primary behavior: Real-time data access with an emphasis on recency and social signals. Grok has direct access to X (Twitter) data and evaluates social proof alongside traditional web content.

Discovery

Grok crawler — indexes web content. Check robots.txt for the xAI user agent.
X (Twitter) integration — Grok has native access to posts, engagement metrics, and trending topics on X.
Real-time web search — fetches and reads live web pages during conversations.

Signals It Prioritizes

Signal	Weight	Notes
Recency	Very High	Latest information is strongly preferred
Social signals (X/Twitter)	High	Mentions, engagement, sentiment on X
Real-time availability	High	Current pricing and stock status
Web content quality	Medium	Page content and structured data
Brand mentions and sentiment	Medium	Cross-platform brand perception

Content Format That Matters Most

Fresh content paired with strong social signals. Grok weights recency and social proof more heavily than other platforms. Products and brands that are actively discussed on X — with positive sentiment — receive stronger consideration. Keeping your web content current (updated pricing, recent blog posts, fresh reviews) and maintaining an active X presence both contribute to Grok’s evaluation.

User (Grok): "What's the best new productivity app right now?"
  → Grok searches real-time web content and X posts
  → Evaluates recent discussions, reviews, and launches
  → Recommends products with social context and recency emphasis

Inception Agents optimization: Ensures content freshness signals (last-modified headers, publication dates, changelog entries). Enriches pages with social proof data. Generates concise-led variants with emphasis on recent updates and current availability.

Cross-Platform Summary

Platform	Primary Discovery	Top Signal	Best Content Format
ChatGPT	GPTBot, Browse agent	Structured data (JSON-LD)	Product pages with schema markup
Claude	ClaudeBot, web search	Long-form documentation	Comparison guides, technical docs
Perplexity	PerplexityBot, live search	Citable facts	Fact-dense, well-structured pages
Gemini	Google-Extended, Merchant Center	Schema.org completeness	Full JSON-LD with entity relationships
Copilot	BingBot, Copilot Catalog	API catalog access	Product feeds and catalog APIs
Rufus	Amazon ecosystem	Listing quality + reviews	Optimized Amazon listings
Grok	xAI crawler, X integration	Recency + social signals	Fresh content with social proof

Inception Agents detects which platform is visiting and automatically serves the content format and variant most likely to result in a recommendation. You can review per-platform performance in Analytics > Platform Breakdown.

Next Steps

Core Concepts — understand detection, optimization, and learning
Quickstart — connect your site in under 10 minutes
Content Variants Guide — customize how your content appears to each platform
Commerce Protocols Guide — enable one-click purchasing through AI agents