Free vs paid AI: why your GEO strategy splits by platform tier

A free ChatGPT user and a ChatGPT Plus user do not see the same AI. A free Claude user and a Claude Pro user do not see the same AI. A Google AI Overview query and a Google AI Mode query are not the same surface. As of 2026-06-18, the market keeps saying "optimize for AI search" as if there is one AI. There is not.

The practical split is simple: free-tier AI is mostly memory; paid-tier AI is memory plus retrieval, shopping feeds, and agentic actions. That means a brand needs two tracks. Track A builds training-data presence. Track B builds real-time discoverability. Most GEO checklists mix the two and then wonder why the results do not match.

The two tracks

Track A is slow. You cannot ship a `robots.txt` rule and retroactively change model weights. Track B is faster. You can fix bot access, product schema, shopping feeds, and agent endpoints in days. The mistake is treating Track B as a substitute for Track A. Real-time retrieval still needs an entity for the system to decide whether to fetch you.

Platform behavior by tier

The platform table below is directional. Product behavior changes quickly, and Google explicitly says its AI Overviews and AI Mode are rooted in core Search ranking and quality systems. ¹ The strategic point remains: each platform has a different mix of memory, retrieval, and commerce.

CiteMetrix found only 11% domain overlap between ChatGPT and Perplexity top-cited domains. ⁷ That number should be treated as a warning label: a brand that wins one AI engine can be invisible in another. Tier and retrieval architecture are part of the strategy.

Track A: training-data presence

Track A is the brand-memory layer. It answers the question: does the model know you without looking anything up? This is where free-tier AI, older model snapshots, and no-tools prompts live. It is also where many brands discover that their website is not the center of their AI presence. The center is the web footprint around the brand.

• Wikipedia and Wikidata presence where the brand qualifies.
• High-authority press mentions and analyst references.
• Reddit, LinkedIn, YouTube, and review-platform presence.
• Durable content that explains the category, product, and differentiator.
• Author and organization schema with sameAs links where applicable.

Track A is not "PR for AI." It is entity formation. The brand has to exist as a recognizable node across independent sources. If the model only knows the brand from owned content, the model has a weak entity. If the model sees the brand across Wikipedia, press, community, review, and product surfaces, the entity becomes stable.

Track B: real-time discoverability

Track B is the live-action layer. It answers the question: when an AI system does look, can it reach and use the brand quickly? This is where bot policy, server-rendered content, product feeds, and agent protocols matter.

• Allow the right AI retrieval bots in `robots.txt` and avoid CDN blocks.
• Ship server-rendered HTML for the pages that matter.
• Keep sitemap.xml current and product sub-sitemaps discoverable.
• Use accurate structured data that matches visible content.
• Publish `llms.txt` as low-cost optionality for non-Google agent surfaces, while recognizing that Google says it does not need it for Search ¹⁵.
• For commerce, support ACP, UCP, OpenAI Merchant Feed, and product-data freshness where relevant.

Track B is where Hidden Layer audits are strongest. The audit can show whether AI crawlers can reach the site, whether product pages are complete, whether agent endpoints exist, and whether the site exposes the signals an agent needs. It cannot, by itself, create ten years of brand memory. That is why the two tracks must be read together.

Commerce changes the tier split

Commerce is the clearest example of why tier matters. OpenAI announced expanded product discovery in ChatGPT through the Agentic Commerce Protocol and Shopify Catalog integration. ² Google and Shopify launched UCP as an open standard for agentic commerce across Google AI surfaces. ³⁴ These are not generic SEO signals. They are commerce rails for agents that can compare, recommend, and transact.

A retailer that only optimizes Track A will be known but not transactable. A retailer that only optimizes Track B may be fetchable but not remembered. The winning setup is both: entity memory plus real-time commerce readiness.

The paid-tier trap

Paid-tier AI looks attractive because it can browse, cite, and act. It also makes brands lazy. If the model can fetch the site, teams assume the site will win. But retrieval does not remove the need for entity memory. It only changes the path to the answer.

The Semrush ghost-citation study shows the problem clearly: 61.7% of AI citations are source links without brand mentions. ⁶ A brand can be retrieved and still be forgotten. Paid-tier access makes the page available; it does not make the brand memorable.

What to do by business type

The operating checklist

1. Identify which AI platforms your buyers actually use.
2. Classify each platform by memory, retrieval, and commerce weight.
3. Run Track A checks: entity presence, Wikipedia/Wikidata, press, Reddit, LinkedIn, YouTube, reviews.
4. Run Track B checks: bot access, server-rendered content, sitemap, structured data, `llms.txt`, and agent endpoints.
5. For commerce, add feed and protocol checks: ACP, UCP, Merchant Feed, AP2/x402.
6. Track citation rate, mention rate, ghost citation rate, and recommendation rate by platform.

Run a Hidden Layer audit to measure Track B: access, discovery, entity signals, structured data, agent integration, and product-page completeness. Then pair it with a Track A brand-presence audit. The combined view tells you whether AI can know you, find you, use you, and recommend you.