GEO and Rich Results: Why AI Shows Your Price but Recommends Your Competitor

Why AI shows your price but recommends your competitor

Modern AI engines (ChatGPT, Perplexity, Gemini, Claude) separate two distinct tasks: structured data extraction and recommendation generation. Your Product or Offer schema is parsed correctly, your prices and ratings appear in the AI answer, but the algorithm recommends buying or acting with a competitor. This separation is not a bug, it is the very architecture of language models: they process metadata and natural content separately, and the final ranking does not depend solely on the visible structured data.

The critical distinction: extraction vs recommendation

When you install a JSON-LD schema on your site, you make structured data extraction easier. AI crawlers retrieve your Product schema with the name, price, average rating, and availability. This is the layer that fills the table or card displayed in the answer. However, the final recommendation ("choose this product," "buy from this seller") relies on other signals: textual relevance, domain authority, external mentions, consistency between user intent and natural content, and the model's training patterns.

A competitor with less structured data but more informative content, a better external reputation, or a better match with the query can therefore be favored despite the absence of enriched schema. AI does not "read" your rich results like a directory: it uses them as signals among dozens of others.

How AI models treat rich results differently

Product schema: visible but not decisive

The Product schema (name, image, description, price, offers) is extracted and displayed. ChatGPT and Gemini often mention it in their contexts: "You will find this product at X for Y euros." However, this schema does not directly improve recommendation ranking. It mainly serves to enrich the display and provide basic information. A site without a Product schema but with rich content on product features and detailed reviews can outperform a competitor that has installed the schema but offers nothing else.

Review schema: the impact on perceived credibility

The Review schema (author, rating, reviewBody, datePublished) carries a different weight. Perplexity and Claude cite structured reviews to justify their recommendations: "this item has a rating of 4.8/5 from 200 users." This schema therefore indirectly influences the recommendation, because it provides the social proof that the model articulates in its answer. A product with 3 structured 5-star reviews will systematically be cited more favorably than one without review data.

Offer schema: availability and price information

The Offer schema includes the price, currency, availability (in stock, out of stock, preorder), and delivery terms. AI extracts this information for immediate display. If your Offer schema indicates "in stock" and a competitive price, the AI answer will show it. But if a competitor has a better price/availability/delivery time ratio, that competitor will be recommended even if your structured data is impeccable.

The hidden signals that go beyond rich results

Domain authority and entity history

AI models incorporate an internal representation of domain authority. A site like Amazon or Cdiscount, regardless of its rich results, benefits from structural weight: it is known, its historical data is massive, and the models have seen it recommended millions of times. A new site with excellent rich results but no external history will not close that gap immediately.

External mentions and brand citations

When an independent blog, a forum, or a media outlet cites your product or recommends your brand without a link, AI models incorporate that mention into their representation of your entity. A competitor mentioned 50 times in recommendation contexts will be favored, even if it has not invested in rich results.

Consistency between context and user query

If a user asks "best laptop for programming," a site with a basic Product schema but detailed content on Python programming, CPU specifications, and RAM will be recommended ahead of a site with a complete Product schema but generic content. AI seeks the maximum match between intent and overall content.

Why Perplexity, ChatGPT, and Gemini do not recommend the one with the best data

Models optimize for relevance, not metadata richness

The loss functions used to train LLMs reward relevance and usefulness, not the presence of structured schemas. A model learns to recognize a good product through the surrounding content, external quality signals, and well-constructed natural language. The absence of a JSON-LD schema is never a direct penalty: it is the absence of quality content that penalizes.

Source diversity in the context window

Perplexity and Claude generally include several sources in their context. If your site ranks number 2 or 3 but number 1 has better natural content, the recommendation will go to number 1. Even if you have the best Product schema in the industry, you remain in the same category of sources as the others, with no decisive advantage.

Training patterns outweigh current data

Generative models were trained on billions of pages before GEO (Generative Engine Optimization) became a topic. Their internal patterns about what constitutes a "good source" for recommendations come from an era when rich results were not an SEO strategy. Although more recent models incorporate technical signals more heavily, this historical bias persists.

How to convert data extraction into recommendation

1. Enrich rich results with related natural content

Do not install your rich results in isolation. Around each Product schema, write content that explains why this product is relevant, its differentiating advantages, and its use cases. If the model finds your schema in a rich, coherent context, it will be more inclined to recommend it. Example: if you sell headphones with a Product schema, also write a section on acoustics, types of use (sport, studio, travel), and honest comparisons with alternatives.

2. Structure Review schemas with detailed justifications

Do not just put "5 stars" in your Review schemas. Add a descriptive reviewBody that justifies the rating. A sentence like "Excellent value for money, 48-hour battery, very comfortable for long sessions" will be more useful to the model than a simple numeric star. AI extracts and cites these review texts: the more specific they are, the more they influence the recommendation.

3. Build an external presence with topical mentions

Rich results alone are not enough: you need relevant external mentions. Encourage blogs, niche forums, and technical influencers to talk about your product. Answer questions on Reddit, Stack Overflow, and Quora by recommending (soberly, without spam) your solution. Each external mention strengthens the signal that AI incorporates into its final recommendation.

4. Implement authority and attribution schemas

Use Organization and Author schemas to establish a strong, credible entity. An Author schema indicates who wrote a review or piece of content; an Organization schema indicates your legitimacy, your certifications, and your presence in directories. These signals complement Product and Review schemas to build an image of trust.

5. Optimize the Offer schema for comparison

In your Offer schema, fill in all fields: price, priceCurrency, availability, seller, shippingDetails, hasMerchantReturnPolicy. The more complete your Offer, the more AI models can cite it in detail and compare it with others. A minimal Offer enables extraction, but a rich Offer facilitates differentiation and a recommendation in your favor (if your offer is genuinely better).

The signals that weigh more than your rich results

Landing page quality

A fast, mobile-friendly product page with a clear structure, high-quality images, and unique content weighs more than a perfect Product schema on a slow, generic page. AI implicitly evaluates user experience: if your site is friction-heavy, even the best metadata will not save the recommendation.

Backlinks and internal linking

External links to your product from relevant, authoritative sites remain one of the strongest signals. A competitor with 10 authoritative backlinks will probably outperform a site with impeccable rich results but no external links.

Evergreen and original content

A comprehensive guide, a detailed FAQ, video tutorials, or case studies around your product are part of the content that AI models value for recommendations. Rich results make parsing easier, but content creates perceived value.

Complete strategy to dominate AI recommendations

Step 1: Audit your existing rich results

Check that all your Product, Review, and Offer schemas are in place, valid, and complete. Use Google's Rich Results Test to confirm there are no errors. If your rivals do not have these schemas, that is an initial win: you will at least have the advantage of enriched display.

Step 2: Produce specialized natural content around each offer

For each major product, create a page with evergreen content: product history, uses, honest comparisons, FAQ, detailed reviews. This content must be unique and answer the questions AI implicitly asks: "why this product rather than another?"

Step 3: Multiply authentic Review schemas

Encourage customers to leave detailed reviews. Import these reviews into structured Review schemas. Each review becomes a piece of content that AI models can extract and cite to justify a recommendation. 50 structured reviews of 4+ stars will weigh more than any self-produced content.

Step 4: Build a targeted external presence

Identify the forums, niche blogs, YouTubers, and influencers relevant to your sector. Make your product known without aggressive spam. Answer questions on Reddit or Quora by recommending your solution when it is genuinely relevant. External mentions increase your credibility in the eyes of AI models.

Step 5: Monitor AI recommendations and adjust

Ask ChatGPT, Perplexity, and Gemini the same questions your customers ask. Note who is recommended and why. If a competitor is ahead of you despite your better rich results, analyze their natural content, backlinks, and external presence. Adapt your strategy to close those gaps.

Common pitfalls to avoid

Do not assume rich results are enough

Many sites install perfect Product schemas and then expect AI recommendations. That is false security. Rich results are a necessary but not sufficient condition. They facilitate extraction and display, but they do not guarantee recommendation.

Do not neglect consistency between schema and content

If your Product schema indicates a price of €99 but your page shows €120, or if the Product schema mentions a feature absent from the content, AI models will detect the inconsistency and penalize your credibility.

Do not ignore negative Review schemas

If there are structured 2- or 3-star reviews with well-founded criticism, AI will see them and mention them. Do not delete negative reviews from your schemas: instead, respond to them publicly in your natural content to show that you take feedback seriously.

Do not spam external mentions

Manipulation attempts (fake reviews on other sites, hidden paid mentions, bot comments) will be detected by modern AI models. Relevant, authentic mentions are worth infinitely more than a hundred suspicious ones.

The future: the growing integration of rich results into AI recommendations

AI models will likely evolve toward better use of structured schemas, particularly for Product and Offer. As GEO becomes a standard practice, models will learn to reward sites that offer rich, consistent structured data more heavily. This means your investment in today's rich results will not be in vain: it prepares your site for tomorrow.

However, this will not eliminate the need for excellent content, external presence, and domain authority. Rich results will amplify these factors, not replace them.