Structured Product and Service Data: Giving AI Facts It Can Quote Accurately

Article 6 of 9 in the 2027 AEO + GEO series

Picture an assistant that recommends a specific espresso machine and states its boiler size, weight, price, and service interval, all pulled from one product page. That answer is only as reliable as the facts the page makes available. If the specs are buried in an image, the price is out of date, or the same product is described three different ways, the model will skip the detail, guess, or send the buyer to a competitor whose data is cleaner.

This article covers how to publish product and service facts in forms that AI systems can read, verify, and quote, from on-page spec tables and schema markup to product feeds and the emerging protocols behind AI shopping.

Why structured data matters more in AI answers

Buyers increasingly ask assistants to compare options against specific criteria: "Which two-group machine fits a small cafe under $10,000?" Answering that requires attributes, not adjectives. Marketing copy that says "professional-grade" and "built to last" cannot be compared. A boiler capacity, a price, a warranty term, and a service interval can.

There are three practical benefits to getting the data right:

• Eligibility. In shopping contexts, many AI surfaces draw from product feeds and structured markup. Missing or incomplete attributes can keep a product out of consideration for a query that depends on them.
• Accuracy. Clear facts reduce the chance a model states the wrong price, size, or compatibility. Wrong answers about your products are hard to correct once repeated.
• Comparability. Consistent attributes let a model line you up against competitors on the criteria buyers care about.

A caution on the statistics floating around this topic. Several industry write-ups report that a majority of pages cited by ChatGPT and Google's AI Mode contain structured data, or that pages with markup are cited several times more often. Those are correlations from vendor analyses. Pages that are well built tend to have markup and also tend to be cited, so markup is unlikely to be the sole cause. Platforms have also not published precise rules about how markup influences recommendations. Treat structured data as a strong foundation and a verification layer, not a ranking trick.

The four layers of machine-readable product data

Layer 1: Clear facts in visible page content

Everything starts with the page a human sees, because that is what crawlers and models read first.

• Put key specifications in text, not only in images, PDFs, or interactive widgets. If a spec only appears after a click, a script run, or inside a graphic, some systems will never see it.
• Use a specification table with consistent attribute names and units, such as capacity in litres, weight in kilograms, price with currency, and service interval in months.
• State differences and fit plainly. Who is the product for, what is it not for, and what does it work with?
• Include the identifiers buyers and systems use, such as model name, SKU, GTIN, and MPN.
• Keep one canonical page per product and make variants (sizes, colors, voltages) clear instead of duplicating near-identical pages.

Layer 2: Schema.org structured data

Schema markup restates the visible facts in a machine-readable format, usually JSON-LD. For products, the core types are:

• Product, with name, description, image, brand, SKU, GTIN or MPN
• Offer, with price, currency, availability, condition, and a link to the purchase page
• AggregateRating and Review, only where you have genuine reviews shown on the page
• Organization, connected to the product through the brand property, as covered in the entity article

For services, use Service with a clear name, description, provider, area served, and offer details, and FAQPage only where the page genuinely contains recurring questions. For local businesses, LocalBusiness and its subtypes carry hours, address, and service area.

The golden rule is that markup must match the visible page. Do not mark up prices, ratings, or availability that users cannot see. Mismatches can cause rich-result problems and undermine trust. Validate with Google's Rich Results Test and the Schema.org validator, and monitor for errors in Search Console.

Layer 3: Product feeds

For retailers, feeds are increasingly the primary data source. Google Merchant Center feeds populate Google's Shopping Graph, which supports Shopping and the AI shopping experiences built on it. OpenAI has published a product feed specification and an application process for merchants who want their catalog to appear in ChatGPT shopping experiences. Other assistants, including Perplexity and Copilot, have their own merchant programs or draw on shared sources, and several platforms rely on Shopify or other commerce platforms to pass data along.

What matters most in a feed:

• Completeness. Fill every attribute that applies. Missing size, material, compatibility, or shipping details are common reasons products fail to match specific queries.
• Accuracy and freshness. Price, stock, and availability must match your site. Reports suggest AI surfaces are more sensitive to stale data than traditional shopping listings, and some protocols allow more frequent updates than a daily file.
• Clean identifiers. Correct GTINs and MPNs help systems match your product to the same product elsewhere, which supports comparison and reviews.
• Descriptive titles and attributes. Include the details buyers search by, such as capacity, power, and compatibility, in a consistent order.
• Rich attributes. Newer feed fields, such as product Q&A and alternative or complementary product references, are being added to support conversational shopping. Check current documentation for what is supported.

Layer 4: Agentic commerce protocols

The newest layer extends beyond discovery toward transactions inside assistants. Two open efforts are most discussed: the Agentic Commerce Protocol (ACP), developed by OpenAI and Stripe, and the Universal Commerce Protocol (UCP), announced by Google with partners including Shopify, Etsy, Wayfair, and Target. Their goal is to let an AI agent discover products, compare them, and complete or hand off a purchase using standardized data.

This area changes quickly. Launch dates, merchant eligibility, supported surfaces, and feature names differ across sources, and some programs are still in preview or limited rollout. Before investing engineering time, confirm the current status in each platform's official documentation. For most merchants, the practical order of priority is:

1. Get your own site data and Schema.org markup clean.
2. Get your Google Merchant Center feed complete and accurate.
3. Review how your commerce platform, such as Shopify or BigCommerce, connects to AI shopping programs.
4. Evaluate direct protocol integrations once the basics are solid.

What this means if you sell services

Not every business has a catalog, but the principle is identical: publish facts a model can quote.

• Describe each service on its own page, with who it serves, what is included, how it works, typical timelines, and what affects cost.
• Publish pricing information where you can, even as ranges or starting prices with clear conditions. Pages with no pricing at all leave the model to fill in a number from somewhere else.
• State coverage and eligibility, such as service areas, industries, minimum engagement sizes, and certifications.
• Use Service, Organization, and LocalBusiness markup consistently, and tie each service to the organization entity.
• Add comparison content, such as how your service differs from alternatives, so the answer to "which should I choose" can come from you.

A worked example

Take a commercial espresso machine. A strong product page would include:

• A headline and opening sentence naming the product and who it is for
• A visible specification table with groups, boiler capacity, weight, price, service interval, and SKU, each with units
• A short paragraph on fit, such as volume range and electrical requirements
• Product and Offer schema matching those values, with the brand linked to the Organization entity
• A matching entry in the Merchant Center feed with the same GTIN, price, and availability
• Review content and ratings only if real reviews appear on the page
• A link to the service manual or warranty page supporting the service-interval claim

Now when a buyer asks for a two-group machine within a budget, every fact the model needs exists in several consistent places that agree with each other.

A worked example: publishing service pricing so AI can quote it

Pricing is where many service businesses leave the most value on the table. When a page says only "contact us for a quote," a buyer who asks an assistant "how much does [service] cost?" gets a number from somewhere else, usually an industry average or a competitor's page. Publishing clear pricing gives the model your number to quote.

Here is how AI Visibility Solutions could present its own pricing in a form that is easy to read and easy to quote.

The plans

What every plan includes

• AI-readable, AI-citable FAQ content for the questions your buyers ask
• Answer-first content structured so each section can be quoted on its own
• Additive content, meaning pages that contribute information, examples, or data not already available elsewhere
• Case studies that document real results
• Content built to meet Google's spam policies, including the policy on scaled content abuse
• Free AI visibility monitoring on two platforms, at no extra charge

Optional paid monitoring. Clients who want deeper tracking can add a usage-based paid monitoring platform. The client signs up directly and shares the platform's API key, and AI Visibility Solutions builds the monitoring setup for them. The platform requires $50 to get started, paid to the platform, and after that the client decides how much or how little to spend each month. This sits on top of the two free monitoring platforms, which are included at no charge. Many monitoring services charge hundreds of dollars per month to track prompts someone guessed at, with no evidence that buyers actually ask them. Here, the paid spend is optional and the client controls it.

How it compares. Done-for-you agencies typically charge $3,000 to $15,000 or more per month, and even entry-level support starts around $1,000 per month. Published examples are listed below. A full 12 months on the annual plan costs $2,997 one time.

What other done-for-you services charge

These figures come from published 2026 pricing guides and agency pages. They cover done-for-you AEO and GEO services, where the agency produces and manages the content. Software and AI writing tools are cheaper, but the client does the work, so they are a different product.

Published examples

• 009 Agency lists managed GEO retainers starting at $3,000 per month.
• TripleDart lists GEO plans starting at $5,000 per month.
• Omniscient Digital lists retainers from $10,000 per month.
• Per piece, one guide puts citation-optimized content at $800 to $2,500 each, depending on depth and research.

What that means over a year

Sources include 2026 pricing guides from The Remarkable Agency, Young Company, Visibility Stack, and Growleads, a 2026 pricing statistics report from First Page Sage, and the agencies' own published pricing. Many of these publishers sell services themselves, and agencies often quote by proposal, so confirm current prices before publishing and keep the sources and dates on file.

A quotable pricing passage

Write the pricing section the way you want an assistant to repeat it: the answer first, then the details, with each fact stated once and plainly.

How much does AI Visibility Solutions cost? AI Visibility Solutions costs $497 per month, $997 per quarter, or $2,997 per year. The quarterly plan includes about one month free and the annual plan about six months free. Every plan includes FAQ and answer-first content, additive content, case studies, and free monitoring on two platforms. A full year costs $2,997 one time, compared with $36,000 to $180,000 or more for a year at done-for-you agencies that typically charge $3,000 to $15,000 per month. Optional paid monitoring requires $50 to get started, paid to the platform, and the client chooses what to spend after that.

Why this works:

• The price is in the first sentence, where a retrieval system is most likely to find it.
• Each billing option is stated with its period, so a model does not have to guess whether $997 is monthly or quarterly.
• Inclusions follow the price, which answers the natural follow-up question of what the money buys.
• Exceptions are stated openly. The paid add-on is described as optional, so no one reads it as a hidden charge.

Mark it up and keep it consistent

• Add Service and Offer schema for each plan, with the price, currency, and billing period, matching the visible table exactly.
• Repeat the same figures, with the same wording, on the pricing page, the FAQ, and any partner or directory listings.
• Add a visible "Last updated" date and update it, along with the schema, whenever pricing changes, as covered in the freshness article.
• Avoid vague or conflicting phrasing such as "from $497" on one page and "starting at $500" on another.

Three cautions on pricing claims

• Substantiate comparisons. A claim about what other agencies charge is a comparative statement. Keep the sources and dates for every figure you cite. The figures come from published 2026 pricing guides and agency pages, and some of those publishers sell services themselves. Describe the range as typical rather than universal, because entry-level programs exist at about $1,000 per month, and compare only against done-for-you services, not software.
• Be precise about savings. At these prices, the quarterly plan saves $494 compared with three monthly payments, and the annual plan saves $2,967 compared with twelve. Describing these as "roughly one month" and "roughly six months" free is accurate, and stating the exact dollar amounts is even clearer.
• Do not promise outcomes with spam-policy language. No one can guarantee how Google will treat a site. Say the content is built to meet Google's spam policies, and avoid wording that suggests protection from penalties or ranking changes.

Audit checklist

• Are key specs and prices in visible, crawlable text on the page?
• Do attribute names, units, and formats stay consistent across all products?
• Is Product, Offer, Service, or LocalBusiness schema present, valid, and matched to visible content?
• Do feed values match site values for price, availability, and identifiers?
• Are GTINs, MPNs, and SKUs correct and complete?
• Is there a process to update the feed and markup whenever prices or stock change?
• Do third-party listings and retailers show the same specs and prices you do?
• Are product pages free of duplicate or conflicting variants?
• Have you checked each AI shopping platform's current requirements instead of relying on older guides?

Common mistakes

• Specs only in images or PDFs. If it is not in text, it may not be read.
• Markup that disagrees with the page. Prices or ratings that differ between schema and visible content create trust problems.
• Stale feeds. An out-of-stock or old price in a feed can lead to wrong answers and wasted traffic.
• Inconsistent naming across channels. The same product called three different things fragments the signals.
• Fake or borrowed reviews in markup. Only mark up reviews that are real and visible on the page.
• Chasing every protocol at once. Foundations first, then integrations.

Quick FAQ

Is schema markup required to appear in AI answers? No platform says it is, and many answers come from pages without it. It helps clarity and verification, and it is required in practice for several rich search features, so it is worth doing well.

Do I need a product feed if I am not a retailer? Usually not. Service businesses and manufacturers without direct sales can focus on clear pages, accurate markup, and consistent third-party listings.

How often should I update product data? As often as the facts change. Treat price and availability as live data, and review specs and descriptions whenever the product changes.

What about product data on retailer sites? Retailers and marketplaces that sell your products are often cited. Supply them with accurate, complete data and check their listings regularly.

Key takeaways

• AI answers compare facts, so publish the specific, consistent attributes buyers use to decide.
• Build four layers in order: visible page content, Schema.org markup, product feeds, and then agentic commerce integrations.
• Markup, feeds, and visible content must agree, and they must stay fresh.
• Services need the same discipline: clear scope, pricing context, coverage, and consistent markup.
• Protocols and platform programs change quickly, so verify current documentation before building.

Next in the series: [One Brand, Five Answers: How ChatGPT, Perplexity, Gemini, Copilot, and Google AI Mode Each Choose Sources](/articles/how-each-ai-platform-chooses-sources).

For the full picture, start with our main guide: AI Visibility for Local Service Businesses.