Why AI assistants misunderstand what you offer

When AI assistants such as ChatGPT, Gemini and Perplexity misunderstand what you offer, the usual reason is that your website doesn’t say it clearly enough. A language model doesn’t see the design; it reads text, headings, lists and structured data, and when those don’t say enough, it fills in the gaps itself, often incorrectly. If your services are described in generic terms, if the site doesn’t answer specific questions, if schema.org data is missing and if there’s no proof that the company really does what it claims, the system gets your offer wrong or skips it. The good news: the causes are predictable and fixable. Below are the five most common reasons we find in site reviews, and the order in which to fix them.

The problem isn’t AI, it’s an unclear website

When a company first asks ChatGPT what it offers and gets a wrong or empty answer, the first reaction is usually that AI “doesn’t work”. In reality, the system did exactly what it could: it summarized what it found. If it found generic phrases about “comprehensive solutions” and a “personalized approach”, it produced a generic description that could apply to anyone.

Generative systems handle content differently from a classic search engine. For years, Google ranked pages by signals such as links and keywords, and the user read the page and decided whether it was the right one. A language model has to understand the page on the user’s behalf. It needs clear entities (who, what, where), clear definitions (X is …) and clear relationships (company A offers service B to customers of type C). Where these elements are missing, the model either leaves the company out or describes it with errors.

A quick reminder of how the approaches differ. SEO (search engine optimization) is optimization for search engines, GEO (generative engine optimization) is optimization for generative AI systems, and AEO (answer engine optimization) is optimization for answers to specific questions; GEO and AEO together cover AI search visibility. GEO doesn’t replace SEO; it builds on technical SEO, content quality, trust and structure. A site that AI assistants struggle to understand usually has problems with Google too; they were just less visible there.

Unclear services

The first and most common cause: the services aren’t clearly defined. A typical site has one “Services” page with five icons and two sentences under each. No system can work out from that what a service includes, who it’s for, how it works and how it differs from similar ones.

A language model needs a standalone page for each service that it can treat as an entity. A page for ecommerce development has to say that it covers ecommerce stores on WooCommerce or a custom platform, what’s in scope (catalog, payments, shipping, ERP integrations), who it makes sense for and how the project runs. A page for B2B ecommerce portals has to explain what a B2B portal actually is, because most customers don’t use the term day to day.

A few signs that your services are unclear:

• All services are on one page, and none has its own URL.
• The service descriptions don’t contain a single sentence of the form “X is …”.
• The same service is named differently in different places.
• The site talks about values and approach but doesn’t say what the customer actually gets.
• It isn’t clear which technologies you use (WordPress, WooCommerce, custom code) and why.

The fix isn’t cosmetic. The company has to decide exactly what it offers and write it down plainly. That’s also one of the most useful exercises for the business, not just for the website.

Too few concrete answers

The second cause: the site doesn’t answer the questions customers actually ask. A user doesn’t type “website development Ljubljana” into an AI assistant. They ask: “How long does it take to build an ecommerce store?” or “Can I connect my online store to my ERP system?” The system then looks for content that answers that question directly.

If your site says nothing about project duration, the system builds its answer from other sources, and your company won’t be mentioned. If the site has an FAQ section saying that a typical ecommerce store takes anywhere from a few weeks to a few months to build, depending on the size of the catalog and the integrations, the system has something to cite.

That’s the core of AEO. Content is organized around questions, and each answer stands on its own: it makes sense even when read without the rest of the page. In practice, this means:

1. Collect questions from emails, meetings and calls with customers. These are better than any keyword tool.
2. Write each question the way a customer asks it, not the way an expert would phrase it.
3. Answer in two to four sentences, with the answer in the first sentence.
4. Place the questions on the service pages where they belong, not on one shared “FAQ” page.
5. Mark up the FAQ section with FAQPage schema.

The same applies to longer content. An AI system is more likely to summarize an article correctly when the first paragraph says what the reader will learn and what the answer is. An article that only gets to the point after five introductory paragraphs will be summarized inaccurately or not at all.

Missing schema data

The third cause: the site has no structured data, or the data is incomplete. Schema markup is code that follows the schema.org standard, usually in JSON-LD format, and tells machines what the page represents. For a company that offers services, the key types are Organization or ProfessionalService (who you are), Service (what you offer), Article (what you’ve published), FAQPage (which questions you answer) and BreadcrumbList (where the page sits in the hierarchy).

Without this data, the system relies on text alone. With it, the system gets confirmation: this is a company called 4tech, run by sole proprietor Aljaž Polh (registered name Spletne storitve, Aljaž Polh s.p.), operating in Slovenia and offering a service called “ecommerce development”. That way the company name, address, contact details and list of services match across the website, the Google Business Profile and directories.

Common mistakes we see in reviews:

• The SEO plugin generates WebPage, WebSite and basic Organization schema, but not the Service type on service pages and not the connections between entities.
• The schema describes reviews or services that aren’t on the page; Google treats this as misleading.
• The FAQ section looks like an FAQ but has no FAQPage markup, so the system sees it as ordinary paragraphs.
• The data in the schema (phone, address, name) differs from the data in the site footer.

Schema isn’t a magic fix. It’s confirmation, and it only works when the content is already clear. But when the content is clear and the schema is missing, adding it is usually the quickest fix to make, with a result you can verify.

Weak project examples and missing proof

The fourth cause: the site doesn’t prove that the company really does what it claims. Providers don’t disclose how generative systems choose their sources, but available guidelines (such as Google’s E-E-A-T criteria) and our own observations suggest that trust counts: who the author is, whether the company has a verifiable identity, whether there are examples of its work, and whether its details are consistent across the web.

A site with no project examples, no author names on articles and no clear business details is anonymous to the system. The system can read it but gives it little weight. The same goes for a site whose portfolio is just a row of logos with no description: from a logo, the system can’t tell what you did and for whom.

Proof that systems understand:

• Project descriptions. What the challenge was, what you did, which technologies you used, what changed for the client. Qualitatively, with no made-up numbers. On the 4tech work page, we list the project type, technologies and a link to the live site; describing the challenge and solution is the next step, and one we recommend for our own site too.
• Authorship. Articles have a named author with a description of their role and experience, linked to the Organization schema via the Person type.
• Consistent business details. Name, address, company registration and tax numbers, contact details; identical on the website, in the schema and in public registers.
• An about page. Who is behind the company, how long it has been in business and in which areas. Written concretely, not as a mission statement.
• External mentions. Publications, directories, profiles and partnerships that confirm the company exists beyond its own website.

An inaccessible or slow site

The fifth cause is technical and often overlooked: the system can’t read the site at all. If robots.txt blocks the crawlers AI assistants use for search and citation, if the site returns errors, or if the content is only assembled after long JavaScript execution, the model gets an empty or truncated page and skips it. The same applies to a site without an up-to-date sitemap and to pages that take several seconds to load, because crawlers have little patience for slow responses.

Check three things: that robots.txt allows the crawlers used for search and citation, that the content is in the HTML the server returns without depending on JavaScript, and that the sitemap is up to date. Only once the site is accessible can the system read your clear services, answers, structured data and proof at all.

What to fix first

The order matters because each step builds on the previous one. Schema without clear content doesn’t help, and an FAQ without service pages has nowhere to go. We recommend this sequence:

1. Check the technical foundation. The site must be indexed, fast and accessible to crawlers. Without that, nothing else counts.
2. Define your services. Give each service its own page with a definition, scope, process and target customer. This is the most work and has the biggest effect.
3. Add FAQs to service pages. Questions from real conversations with customers, answered in a few sentences.
4. Fix your schema data. Organization or ProfessionalService, Service on every service page, FAQPage, Article, BreadcrumbList. Check that the data matches the content.
5. Strengthen your proof. Project descriptions, authorship, an about page, consistent details in external sources.
6. Check the result. Ask ChatGPT, Gemini and Perplexity the questions your customers ask, and compare the answers with how things were before the changes.

The last step isn’t a one-off. It’s worth checking AI assistants’ answers regularly, because the systems and their sources keep changing. A properly structured site doesn’t guarantee that a system will recommend you, but it makes it more likely that the system will understand you and describe you correctly when it does mention you. For companies that want to go a step further, there are also AI integrations, where a company puts its own knowledge and data to work in its processes, not just on its website.

To see how Google and AI assistants understand your offer, get a free check.