AI writes faster, but not necessarily better

“Can’t you just have ChatGPT write the text? Then we’ll be done faster.” We hear this question in almost every other initial consultation these days. It’s an understandable question. AI-generated texts are ready in seconds, cost nothing extra, and look decent at first glance.

The problem doesn't become apparent until later, once the site is live and no one can relate to it. The text sounds professional, but somehow it also sounds just like the text from ten other providers. This isn't a coincidence—it's a direct result of how these models generate text, and it's an issue we encounter time and again in client projects.

This article shows where AI really helps when writing for a website, where it consistently falls short, and why this distinction ultimately determines whether a text builds trust or quietly erodes it.

Where AI Actually Saves Time

There are certain text-related tasks where, to be honest, AI is truly excellent. It can consolidate research on a topic, suggest an initial rough outline for a long page, and turn a rambling customer interview into bullet points. AI also does a great job laying the groundwork when brainstorming different headline options or rephrasing a overly technical sentence into more understandable language.

These tasks have one thing in common: they are preparatory. They provide raw material or structure, but not yet a finished message intended to speak on behalf of a company. It is precisely in this role that AI has become firmly established in our day-to-day editorial work—not as a replacement for a text, but as a tool that speeds up the preparatory work.

The benefits are also clear when it comes to multilingual websites. AI can produce an initial translation of an existing German text into English or French with reasonable quality and in a fraction of the time it would take for a traditional translation. However, a final human review remains essential, especially for idioms and industry-specific terms that cannot be translated literally. Nevertheless, it is enormously valuable for saving time on the first draft.

Another area where AI works reliably is summarization. A long white paper, a comprehensive briefing document, or a 40-minute client meeting can be distilled down to its key points in a matter of minutes. This not only saves time, but also helps ensure that you don’t overlook any important points during a long conversation that might later be relevant to the text.

The Difference Between a Text and Your Text

This is the real crux of the matter. An AI model has been trained on millions of texts and uses that data to generate a statistically probable, average text on a given topic. The result is structurally sound but interchangeable in terms of content. It could almost be word-for-word identical to text on a competitor’s website, and that is exactly the case with many “About Us” pages and service descriptions these days.

A company profile, however, is meant to do exactly the opposite. It should show why this company operates differently from the ten others offering similar services. This differentiation stems from specific details: a real-life customer case study, a unique perspective on a controversial industry topic, or a turn of phrase that could only have come from this one person. AI has no access to these details unless someone provides them. Nor can it invent them without sounding like it’s making a claim.

We've tested this ourselves several times: A purely AI-generated draft for an "About Us" page reads smoothly but feels interchangeable. If you replace the company name with another one, it's hardly noticeable. It is precisely this lack of specificity that is the opposite of what an "About Us" page is actually supposed to achieve.

Where AI-Generated Text Stands Out Immediately

Certain patterns in AI-generated text have become so well-known that many readers unconsciously pick up on them as warning signs. Phrases such as “in today’s digital world,” “holistic approach,” or “tailored solution” now almost automatically raise an eyebrow among seasoned readers.

Added to this is a certain style of rhetoric that AI models seem to favor: “Not only X, but also Y” as a standard phrase structure, three-part lists where the third point is a broad, vague claim, and paragraphs that are all exactly the same length and end with a short, snappy one-sentence conclusion. Taken individually, none of these patterns is a problem. But when they occur in clusters, they act like a fingerprint that many people now recognize, even if they couldn’t technically explain exactly what it is that gives it away.

For a company with a premium positioning, this effect is particularly risky. A company that promises high-quality work but publishes text that reads like a hastily generated AI output creates exactly the opposite impression of what the website is actually meant to convey. The disconnect between expectations and execution is more noticeable in language than in almost any other element of a website—even more so than in design details that many visitors perceive only subconsciously.

Product descriptions and data sheets: AI can take on more of the work here

Not every piece of text on a website needs to reflect a company’s unique voice. With highly structured, fact-based content—such as product descriptions, technical data sheets, or variant overviews with many similar entries—there is little room for individuality anyway. What matters most here are completeness, consistency, and accuracy—and that’s exactly where AI excels.

An online store with several hundred products benefits enormously when AI helps create consistent descriptions, as long as someone ultimately checks the facts and smooths out obvious repetitive patterns. The effort required to craft every single product description individually from scratch is out of proportion to the benefit that such individuality would provide. Unlike on an “About Us” page, no one here reads ten product descriptions in a row and compares the writing style anyway.

"About Us" pages and landing pages: a human should have the final say here

The opposite is true for pages designed to influence a decision: “About Us” pages, service pages, and landing pages for specific offers. These texts carry the greatest responsibility because they are present at the very moment someone decides whether or not to submit an inquiry. This is where every investment in truly original content pays off most directly.

For these pages, we use AI at most to create the initial structure or a rough draft that a human then works on. Every statement is reviewed: Does it truly apply to this company, or is it a generic phrase that could just as easily appear on any other provider’s website? This review takes longer than the actual writing, but it’s the part of the work that really makes the difference.

A real-world example

For a project with a consulting firm, the team had already created their own service pages using AI to save time before partnering with us. The text was grammatically correct and the content was comprehensive. Nevertheless, they received no inquiries, even though traffic to these pages was steady.

During the analysis, it quickly became apparent what was missing: There wasn’t a single sentence on the page that couldn’t have appeared on a direct competitor’s website. No concrete method, no unique perspective on a well-known industry problem, not a single sentence expressing an original opinion. The text was correct, but it lacked the substance needed to set it apart from the competition—even though the consulting team itself possessed precisely that substance; it just wasn’t reflected on the page.

We redesigned the pages, using the same facts but incorporating specific details from real client projects, presenting a clearly identifiable stance on a controversial method in the industry, and using language that actually came from a real conversation with management rather than from a general knowledge base. The inquiry rate on these pages improved noticeably within a few months, with virtually the same amount of traffic. The difference lay not in the grammar, but in the substance.

The reaction of the management team itself was also interesting. Upon their first reading of the new draft, several members commented that the text felt “just like us,” even though it contained less continuous prose than before. Less, but more concrete material apparently comes across as more convincing than a longer, smooth text lacking any distinct character. This aligns with an observation we’ve made time and again across many projects: Trust is rarely built through comprehensiveness, but rather through individual, distinctive details that a reader wouldn’t find anywhere else.

How We Use AI in Our Own Editorial Process

Here, too, we now create blog posts, service descriptions, and customer communications with AI support at certain stages of the process—for the initial structure, for research, and to turn a briefing into an outline more quickly. What never happens automatically is the final step: proofreading to ensure the text reflects our own tone, removing generic phrasing, and verifying whether a statement is actually true or just sounds plausible.

We draw this line deliberately—not out of principle, but out of experience. A text taken from a model without being reviewed almost always contains one of the typical clichés or a claim that, upon closer inspection, turns out to be too general. Finding and replacing these passages takes time, but significantly less than writing a completely new version. The result is a text that was produced faster than it would have been without AI, yet sounds just as unique as before, because the actual work—checking it against our own voice—was not skipped at any point.

Why a writing guide is more helpful than a better prompt

Many companies try to solve the problem of interchangeability with a clever prompt. However, prompts like “Write this in a more casual style” or “Make this more personal” rarely lead to a truly different result, because the model still lacks the essential foundation: specific knowledge about how this particular company actually sounds, what it has said in the past, and where it stands on controversial issues within its own industry.

It works much more reliably when you provide real reference texts, existing customer communications, or a short internal document that reflects your own tone as a basis. The model then has something concrete to emulate rather than just a vague instruction to interpret. That’s exactly how we approach this blog: Our own style guide specifies which phrasing to avoid, how paragraphs should be structured, and which language patterns are typical of generic AI text. Without this foundation, even a good model would drift toward average language style, simply because that’s the most likely starting point when no more specific information is available.

For companies that regularly produce text, a one-time investment in a brief but specific style guide is therefore more worthwhile than constantly tweaking individual prompts. Such a guide doesn’t have to be long. A few sample sentences, a list of phrases to avoid, and two or three real-world text examples are usually enough to noticeably narrow the gap between generic text and text that is clearly original—regardless of which model ultimately writes the first draft.

The exam question that makes all the difference

There’s a simple question you can use to test almost any website text, regardless of whether it was created with or without AI: Could this sentence just as easily appear on a direct competitor’s website? If the answer is yes, then the text is lacking something, regardless of how it was created.

This question can be applied to every single statement on a page, not just to the text as a whole. “We value personalized service” almost never passes this test, because practically every company claims that about itself. “We respond within 24 hours, personally and without a sales pitch” passes it, because it is specific and verifiable. The difference between the two statements lies not in the choice of words, but in whether a statement actually reveals something about this particular company or merely repeats a general expectation that everyone has anyway.

If you’re writing your own content using AI, you can incorporate this check as a final step before a page goes live. Go through each paragraph once and honestly ask yourself whether it passes this test. If not, the paragraph needs a specific detail, a number, or a real-life example that makes it stand out. This final step rarely takes more than half an hour, but it often makes the biggest difference in the entire text.

Conclusion: A tool for speed, not a substitute for substance

AI changes how quickly a text is created, not automatically how effective it is. It’s a real time-saver when it comes to structure, research, and repetitive, fact-based content. When it comes to texts intended to influence a decision, the substance still requires manual work, because that is precisely where the difference between an interchangeable text and a compelling one lies. Those who clearly distinguish between the two—rather than using AI indiscriminately for everything or not at all—ultimately gain both: greater speed and texts that truly speak on behalf of their own company.

Related: The EU AI Act Explained Simply and The 5 Processes Where AI Pays Off First in Almost Every Company · Services: AI Consulting and Implementation

Would you like us to proofread your texts?

We’ll take a close look at your most important pages and give you our honest opinion on whether they measure up to the competition or come across as rather generic. Send us a quick message using the contact form. We’ll respond within 24 hours—personally and without any sales pitch.