Readers don't notice the tool, they notice the symptoms: stock phrases, generalities, no figures, no examples from the business and no point of view. An AI text that's been briefed with real expertise and checked before publication is indistinguishable from an edited human text. Google itself says it judges the quality of content, not how it was produced.
This question has two audiences, and for both the answer differs from what most people expect. Readers don't spot AI text as such. They spot bad text, and bad AI text has typical traits. Google, for its part, says it doesn't look for AI, it looks for content that wasn't written for people. Whether Google penalises AI text is covered on the answer page Does Google penalise AI text?. This page is about how to recognise a text that gives away that nobody took it seriously.
What readers actually notice
Put twenty unbriefed AI blog posts side by side and you'll see the same five patterns every time:
- Stock phrases at the start. "In today's digital world", "Let's dive in", "An often underrated topic". The first sentence tells the reader nothing they didn't already know.
- Generalities instead of statements. "The costs depend on various factors." True, but useless. A specialist business would name three factors and give a rough figure for each.
- No figures, no names, no places. The text could have come from any business in any country. Nothing points to Thurgau, to the equipment the business actually installs, or to a job from last month.
- No point of view. Every comparison ends with "both have pros and cons". A specialist would have a recommendation, with reasons.
- Symmetrical structure. Every section the same length, every list with three points, and a "conclusion" at the end that just repeats the text.
None of these patterns is an AI trait in the strict sense. All five also turn up in texts written by an unmotivated intern or an agency with no briefing. They're symptoms of missing input, not of missing humanity. That's why swapping out individual words doesn't help either. The list of telltale phrases and their alternatives is on Spotting and removing AI stock phrases.
What Google says about it
In February 2023, Google stated on its blog that how content is produced isn't the deciding factor. Automation has been legitimate for years, for example for sports results or weather reports. What matters is whether the content is helpful, reliable and written for people. According to Google, AI use becomes a problem when it's used to manipulate rankings, for example by mass-producing content with no added value.
The helpful content guidelines list the questions Google uses to describe quality: does the content show first-hand experience? Would you expect to find it in a printed trade magazine? Was it written by a person or company with demonstrable knowledge of the subject? These questions are tool-neutral. An AI text can pass them if experience and expertise have gone into it. A handwritten text can fail them.
Where the AI traces really come from
A language model writes whatever statistically fits a question. Without specific input, it delivers the average of all texts on the topic, and the average is, by definition, interchangeable. So the traces that put readers off arise at three points:
| Point | What's missing | What readers notice |
|---|---|---|
| Briefing | Business, region, target audience, your own prices and processes | Text fits any provider |
| Specialist input | Experience, examples, recommendation | Generalities, no point of view |
| Review | Someone who reads it and edits it down | Stock phrases, repetition, incorrect details |
Cover all three points and you get texts readers read as specialist posts. Skip all three and you get texts readers click away from after the second paragraph, whether or not they're thinking about AI.
How to avoid it
Three measures are enough for most SME blogs:
- A briefing with substance. Every post gets the actual customer question, your answer in three sentences, a figure or example from the business, and the recommendation you'd give in a conversation.
- One review pass with a fixed checklist. Read the first paragraph: is the answer there? Check the figures: are they correct? Look for "various factors", "holistic", "in conclusion": cut them or make them specific. You'll find the full list under Checking AI blog posts.
- A responsible person. A name and a role attached to the post, and that person has read the text. That's what Google means by experience and trustworthiness, and it's what readers are looking for when they ask themselves whether they can believe the text.
A blog system that derives posts from real search queries in the industry and submits them for approval before publication covers the first and third points structurally. BlogPilot, for example, works this way. The specialist substance still has to come from the business: no system knows your prices and your opinion before you enter them.
What to do with this
Read your last blog post and cross out every sentence that could just as easily appear on a competitor's site. What's left is the part readers recognise as yours. If that's less than half, the problem isn't the tool, it's the input.
Frequently asked questions
Do AI detectors work reliably?
There's no public evidence that detection tools reliably tell AI text apart from human text. They regularly flag human texts as AI and vice versa, especially with factual, well-structured texts. They're not fit for use as a basis for decisions. Checking for the symptoms that actually put readers off is more reliable.
Does Google use such detectors?
Google hasn't confirmed that it uses AI detectors, and its published guidelines focus on content quality, not how it was produced. What Google penalises is content mass-produced to manipulate rankings, regardless of whether a human or a machine wrote it.
Should I state on the website that AI helped write the content?
There's no requirement to do so in Google's guidelines. Google recommends giving production details where readers would reasonably expect them. More important than a disclosure is having a named, responsible person behind the post who has checked the content. Whether you disclose AI use is a matter of your own approach, not of visibility.
Does all AI-written text sound the same?
Unbriefed text does. Without details on the business, region, customers and personal opinion, every model falls back on the same generic phrases, and the texts end up looking alike across providers. With concrete input on prices, processes and examples, the texts differ as much as the businesses themselves.