Language models do not select content based on whether it was written by a human or an AI, but rather on the basis of quality, clarity, and usefulness. From an AI’s perspective, poor content is content that cannot be easily summarized, has a chaotic structure, or adds nothing new. As a result, it not only performs worse for users but is also cited less frequently in AI responses.
Table of Contents
Anti-pattern 1 – A wall of text without structure
The most common mistake is a long wall of text without a meaningful division into sections. Such text may be full of paragraphs, but the headings are random or missing. From an AI’s perspective, this is difficult material because it cannot be easily broken down into specific answers to the user’s questions. Models prefer content in which each section has a clearly defined topic.
If the text lacks a logical H1–H2–H3 structure, doesn’t start sections with a clear answer, and can’t be “scanned at a glance,” it usually can’t be easily processed by AI systems either. Paragraphs that are too long, a lack of formatting, and mixing multiple threads in a single block are a surefire recipe for content that models will skip in favor of a better-organized source.
Anti-pattern 2 – Repeating What Everyone Else Has Already Written
Many texts are created as a compilation of the top search results, without adding anything original. For humans, this can be uninspiring, but for AI, it’s a real problem. Models look for content that provides new information, examples, data, or perspectives. If a text is merely a jumble of general statements that already appear on dozens of other websites, it becomes practically invisible to AI.
In this sense, poor content is content that repeats obvious definitions without providing practical applications, real-world experiences, or specific figures. Even a well-written article that doesn’t stand out in any way has little chance of being cited as a source. Models prefer to cite texts that include unique examples, case studies, or original data.
Anti-pattern 3 – Content “optimized for keywords” rather than for questions
The old approach to SEO relied on cramming keywords densely into the content. As a result, texts were produced that are difficult to read, full of repetitions, and unnatural phrasing. From an AI perspective, such content is of little use because it doesn’t directly answer specific questions; instead, it constantly repeats the same phrase in different variations.
Language models look for answers tailored to queries, not mechanical repetitions of keywords. If the content lacks clear, simple sentences explaining how to do something, what it is, how it works, when it’s worth it, or who it’s for, the text is less valuable to AI. Excessive “keyword stuffing” can actually reduce the chances of being cited, as it makes it harder for the model to extract a meaningful answer.
Anti-pattern 4 – Failure to Answer the Question at the Beginning of a Section
Even a well-structured text can be problematic when each section begins with generalities rather than specifics. From an AI’s perspective, a better structure is one where the heading suggests a question, and the first 1–2 sentences under the heading provide a direct answer. If the author buries the answer halfway through a long paragraph, models will often skip that section in favor of a clearer source.
Poor content in this category is that in which the user has to “wade through” introductions, digressions, and marketing jargon before figuring out what the point is. AI faces a similar problem-it’s easier for it to choose a text where the answer is provided at the beginning of a section and the rest serves as elaboration, rather than one where the most important sentences are scattered throughout the article.
Anti-pattern 5-Beating around the bush instead of getting to the point
Models place a great deal of trust in content that includes specific data, examples, definitions, and clear steps-rather than just general statements. Poor content is dominated by empty phrases such as “modern solution,” “high quality,” or “innovative approach,” without explaining what these mean in practice. For AI, such text is difficult to use because it provides no concrete points of reference.
If the text lacks numbers, examples, descriptions of real-life situations, or even simple definitions, the model has nothing to cite. It is precisely this specific information that builds the credibility and usefulness of the content. Without it, an article may appear long and “rich,” but from an AI’s perspective, it remains an empty form.
Anti-pattern 6 – Overusing jargon without explanation
In many industries, it’s natural to use specialized language. The problem arises when the entire text consists of jargon, acronyms, and internal terms that no one outside the company understands. For AI designed to assist non-expert users, such content is less useful because it’s difficult to incorporate directly into a response.
Poor content in this regard isn’t the mere presence of industry terms, but the lack of a simple explanation of what they mean. Models will make better use of text where a difficult term is introduced and immediately defined in a straightforward way. If an article contains many incomprehensible acronyms and internal jargon, AI may have difficulty correctly interpreting exactly what the company or store does.
Anti-pattern 7 – Information Locked in Images and PDFs
A major problem is content where key information is available only in images, scans, or large PDF files. From the user’s perspective, this can sometimes be inconvenient, but from the AI’s perspective, it can be a major obstacle. Models prefer to cite content they can easily read from HTML rather than documents that require additional processing.
In this sense, “bad content” refers to content in which important tables, price lists, instructions, or rules are embedded only as images or attached files and have no equivalent in plain text on the page. In such situations, models often turn to other sources where similar information is stored in text form and is easier to extract.
Anti-pattern 8 – Outdated or Self-Contradictory Content
Language models dislike information chaos. If a page contains conflicting information-such as different versions of prices, delivery terms, or product descriptions-the AI has trouble determining which version is correct. The same thing happens when the content contains outdated dates, obsolete data, and references to services the company no longer offers.
In this context, poor content refers to content that hasn’t been updated in a long time or has been overwritten multiple times without deleting previous versions. Models approach sources that appear outdated with greater caution and are more likely to cite material that is fresh, consistent, and unambiguous. For a brand, this means regularly reviewing key pages.
Anti-pattern 9 – Lack of Structured Data and Contextual Cues
Although content is the foundation, from an AI perspective, technical signals that help understand what is what are also important. Pages without meaningful headings, without structured data, and without a clear title and meta description are more difficult for models to interpret. Even good text can get lost in the sea of other content if it isn’t properly marked up.
In this context, poor content is content that provides no context. The absence of an unambiguous page title, a clear introduction, or information about the author and the company the content pertains to makes it difficult for models to associate the information with a specific brand. As a result, AI may quote the content in a way that’s disconnected from your company or choose another, better-described source.
Anti-Pattern 10 – Content That Shows No Evidence of Experience
Signals of experience and credibility are becoming increasingly important. Poor content isn’t just content full of errors, but also content that appears to have been written “from the internet’s memory” without any trace of real-world experience. A lack of examples from your own work, no named author, and no information about your expertise make the text easy to replace with other, more credible material.
Influencers are more likely to cite content that demonstrates concrete experience: references to real-world situations, a clearly described industry context, data from their own analyses, or tested procedures. If a text is completely detached from practical experience, it remains just another general description that can be skipped.