The 10 GEO Checks Most Websites Fail
Last updated: 24. August 2026 · 10 entries
To appear in the answers of ChatGPT, Perplexity, Claude, or Google AI Overviews, you don't need a new discipline, but verifiable, machine-readable information. We analyzed where this fails in practice: 23,696 completed audits on 13,327 domains, tested with the AI Trust Audit. These ten criteria fail most frequently — and some of them might surprise you.
Key Takeaways
- Database instead of gut feeling: All percentages in this article come from 23,696 completed audits on 13,327 domains (as of August 2026), not from a survey or estimate.
- The classics are done, the evidence is missing: Privacy policies, robots.txt, and sitemaps sit between 73 and 94 percent. In contrast, the markup of reviews is at 5 percent.
- The biggest failure is the most unknown: The
sameAslink — the connection of your own brand to its profiles on the web — is completely missing on two out of three pages tested. - Effort and impact rarely align: Six of the ten points can be implemented in a few hours because they are about markup, not new content.
How we selected
The basis is all completed audit runs. For each criterion, we have data on how often it was fully met, partially met, or not met. We included criteria with high weighting and high failure rates — in other words, where high impact meets low implementation.
Conspicuously absent from the list are the no-brainers: Privacy policy (73 percent fulfill it completely), accessible robots.txt (94 percent), readability (87 percent), XML sitemap (81 percent), legal notice (68 percent). The title tag also only fails on 5 percent of pages, and only 6 percent actively block GPTBot. These are important foundations — but no longer major construction sites.
Product-related criteria only run on detected shop pages; their rates refer to 4,747 runs. The numbering is alphabetical and not a ranking.
Comparison at a glance
| Check | Category | Fully Met | Effort | Impact on AI Answers |
|---|---|---|---|---|
| 1. Image Alt Texts | Content | 35% | Medium | High |
| 2. FAQ Schema | Structure | 7% | Low | Very high |
| 3. Organization Schema | Structure | 46% | Low | Very high |
| 4. Product Identifiers | Product data | 27% | Medium | Very high |
| 5. About Us Page Quality | Content | 25% | Medium | High |
| 6. Review Schema | Reputation | 5% | Medium | Very high |
| 7. sameAs Entity Linking | Structure | 22% | Low | High |
| 8. Date of Publication | Content | 27% | Low | High |
| 9. Complete Offer Data | Product data | 34% | Medium | Very high |
| 10. Website Schema with Search | Structure | 36% | Low | Medium |
1. Image Alt Texts
Category: Content
What is checked: Whether the content-relevant images on a page have descriptive alternative texts — and not just filenames or empty attributes.
Success rate: 35 percent meet the check completely, 26 percent not at all. This makes it the weakest among the high-weighted criteria.
Why it matters for AI answers: Text-based crawlers don't see an image. What is in the alt text is all that can make it from the image into the answer — for product photos, this means color, material, variant.
Best Practice: alt="Espresso machine portafilter, 2 brew groups, stainless steel" instead of alt="Product image". Decorative images get an empty alt="" so they are intentionally skipped.
Most common mistake: The shop system automatically fills the alt text with the product name. Formally present, but content-wise without additional information.
2. FAQ Schema
Category: Structure
What is checked: Whether questions and answers on the page are marked up as FAQPage.
Success rate: 7 percent. More than nine out of ten pages waste the format that is closest to the answer format of AI systems.
Why it matters for AI answers: A question-answer structure provides exactly the block a model can quote: a distinct question and a self-contained answer — without having to reconstruct the context from running text.
Best Practice: Place FAQ blocks where the question arises — on product, shipping, and return pages, not just on a central FAQ page. Each answer in two to four sentences, understandable on its own.
Common mistake: The questions appear as an accordion on the page, but the markup is missing. Visible to buyers, but just a bunch of div elements for machines.
3. Organization Schema
Category: Structure
What is checked: Whether the website identifies its operator as an Organization — name, logo, contact method, address.
Success rate: 46 percent. On more than half of the pages, the basic identity is missing, even though this check is among the highest weighted.
Why it matters for AI answers: Without a clear organization, a model cannot assign statements, reviews, and products to a brand. Everything else builds on this.
Best Practice: Provide an Organization object globally in the page header, with name, url, logo, contactPoint, and address — identical on all pages and identical to the legal notice and Google Business Profile.
Common mistake: Two contradictory company names — one in the markup, one in the legal notice. This weakens the assignment instead of strengthening it.
4. Product Identifiers
Category: Product data
What is checked: Whether products carry unique identifiers: GTIN or EAN, MPN, SKU, and the brand.
Success rate: 27 percent of the shop pages checked meet the check completely, 62 percent not at all.
Why it matters for AI answers: Shopping agents must recognize that your offer concerns the same product as a competitor's. Without an identifier, your item is simply a different item for a price or availability comparison.
Best Practice: Output gtin13, mpn, sku, and brand in the Product object. For private labels without GTIN: maintain MPN and brand consistently.
Common mistake: The identifiers are in the ERP system and the Google Merchant Feed, but not in the HTML of the product page. The feed only reaches Google, the product page reaches everyone else.
Surprising fact: This point is the clearest harbinger of agentic shopping — and yet it is the most poorly maintained product dataset.
5. About Us Page Quality
Category: Content
What is checked: Not whether an About Us page exists, but whether it contains verifiable facts: founding date, location, team, core assortment, contact.
Success rate: 25 percent. In comparison: 89 percent have an About Us page.
Why it matters for AI answers: When asked "Is this provider reputable?", the About Us page is the densest source of facts on the domain. Marketing prose provides nothing for this.
Best Practice: Be specific: founding year, legal form, headquarters, number of employees, contact person with names, selection of brands carried.
Common mistake: Three paragraphs about passion and quality, without a single verifiable piece of information.
Surprising fact: The gap between "present" and "meaningful" is larger for this criterion than for any other.
6. Review Schema
Category: Reputation
What is checked: Whether customer reviews are marked up as Review or AggregateRating.
Success rate: 5 percent. This is the most striking value in the entire dataset — and it's not because the shops don't have reviews.
Why it matters for AI answers: Reviews are the most frequently cited trust feature in AI answers about providers. A grade without origin and quantity is just an unproven number.
Best Practice: Deliver and mark up the grade, quantity, and individual review texts server-side. Trusted Shops reviews — our own offer — come from verified orders and can be integrated so that they are in the source code.
Common mistake: The review widget is reloaded via JavaScript. Buyers see the stars, but a large part of AI crawlers only read the pure HTML — where nothing is listed.
Surprising fact: The distance between visible reviews and marked-up reviews is the largest single visibility gap we measure.
7. sameAs Entity Linking
Category: Structure
What is checked: Whether the brand links its official profiles — Google Business Profile, LinkedIn, Wikidata, industry directories — to its own domain via sameAs.
Success rate: 22 percent. With a 67 percent clear failure rate, this is the biggest failure of the entire evaluation.
Why it matters for AI answers: sameAs tells a model: The brand on this website and the profile over there are the same entity. Without this bridge, external evidence remains unlinked noise.
Best Practice: Maintain a sameAs array with the official profile URLs in the Organization object. Only include profiles that actually belong to you and are active.
Common mistake: Linking social buttons in the footer instead of machine-readable links in the markup.
Surprising fact: Effort: around fifteen minutes. Yet it is the least fulfilled structural check.
8. Date of Publication and Update
Category: Content
What is checked: Whether content carries a machine-readable creation and modification date.
Success rate: 27 percent, 73 percent do not meet the check at all.
Why it matters for AI answers: For time-sensitive questions — prices, legal status, availability, years — models demonstrably prefer dated sources. Without a date, your content remains excluded from any timeliness check.
Best Practice: Keep datePublished and dateModified in the markup and also output the date visibly. Only change dateModified if the content has actually changed.
Common mistake: The date is automatically updated with every deployment. This creates a sense of timeliness that isn't real — and devalues the signal.
9. Complete Offer Data
Category: Product data
What is checked: Whether price, currency, availability, as well as shipping and return conditions are available in machine-readable form for the product.
Success rate: 34 percent of the shop pages checked, 64 percent fail.
Why it matters for AI answers: "How much does it cost including shipping, and is it available?" is the typical purchase intent question. If one of the details is missing, a model cannot include your offer in the comparison — it leaves you out.
Best Practice: Output price, priceCurrency, availability, shippingDetails, and hasMerchantReturnPolicy in the Offer object and keep them synchronized with the information on the page.
Common mistake: Price and availability are marked up, but shipping and returns are not. Yet these two are decisive in a comparison.
10. Website Schema with Search
Category: Structure
What is checked: Whether the domain provides a WebSite object including a SearchAction.
Success rate: 36 percent.
Why it matters for AI answers: This tells a system how to search specifically on your site instead of navigating through the menus. This is becoming increasingly relevant for agentic access.
Best Practice: Deliver WebSite with potentialAction and your own search URL in the page header — once globally, not differently per page.
Common mistake: The search URL in the markup points to an old parameter that leads nowhere.
Frequently Asked Questions
What is GEO anyway?
GEO stands for Generative Engine Optimization: the optimization of a website to appear in answers from generative AI systems. Technically, it overlaps heavily with SEO but weights machine-readable evidence and clear entities much more highly.
Isn't classic SEO enough?
For the basics, yes — a clean structure, accessible sitemap, and good content benefit both. The difference lies in the markup: what a human sees on the page must also be available as a structured detail, otherwise it doesn't exist for an HTML crawler.
Where should I start if I only have one day?
With the structural checks: Organization, sameAs, and date information can be done in a few hours and affect the entire domain. The markup of reviews and product data follows after that.
How often should I check this?
After every major relaunch and otherwise quarterly. Experience shows that structured data is lost most quickly during template changes.
Where your website stands
All ten criteria are part of the AI Trust Audit. The check is free, takes a few minutes, and shows for each point whether it is met, partially met, or open — including specific implementation instructions for your domain.
Methodology
All completed audit runs in the AI Trust Audit (as of August 2026) were evaluated: 23,696 runs on 13,327 unique domains. The evaluation basis differs for each criterion because individual checks were introduced over time and product-related checks only run on detected shop pages — the respective basis therefore ranges from around 4,700 to 23,300 runs. Detailed results older than twelve months are deleted for data protection reasons and are not included in the rates.