More people now read web content wondering whether a person wrote it. When a page reads as generated, readers start to doubt it within a few sentences, even if everything on it is correct.
What gets read as AI writing
Most of the signs are in the structure. Generated text tends to announce a point before making it and count the items before listing them. Sections end by repeating what the reader just read, and paragraphs come out the same length and shape, often with lists of three.
Readers also notice certain phrases, such as “it’s not just X, it’s Y”, “in today’s digital landscape”, or words like seamless and robust, along with constant reassurance that none of it needs to be complicated. The em dash has picked up the same reputation, although careful writers used it long before chatbots did.
What gives generated content away most often is that it stays general. Where someone with experience would mention a price or a mistake they have seen, generated text says something true that could apply to anyone.
What gets read as an AI image
With images, people look for hands with extra or missing fingers, lettering on signs or packaging that turns to nonsense up close, airbrushed skin, and light coming from two directions at once. Backgrounds that dissolve into patterns when you zoom in are another common sign, as are people who look like nobody in particular, in an office that could be anywhere.
Websites had that generic look long before image generators. Heavily retouched stock photos and template layouts produced it for years.
First Nations organizations and community groups have more at stake here. A generated image of people or cultural imagery can misrepresent a community, and real photos used with consent avoid that.
Why the tells are not proof
Corporate copy and school essays were full of these writing habits before language models existed, and the models learned them from people. A careful writer can produce a paragraph that reads as generated, and a generated paragraph can be edited until nothing about it stands out.
Detection tools have a poor record. OpenAI withdrew its own AI text classifier in July 2023 because of its low accuracy: in its launch testing it caught 26 percent of AI-written text and wrongly flagged 9 percent of human writing. A 2023 study of seven widely used detectors found they flagged, on average, 61 percent of essays by non-native English writers as AI-generated. When researchers revisited the problem in 2026, one commercial detector still flagged 23 percent of non-native English essays and none written by native speakers.
For images, provenance standards such as C2PA Content Credentials can record how a file was made and edited. That record is often stripped when an image is screenshotted or re-uploaded, though, and a missing record proves nothing.
Most of the time nobody can say for certain whether AI was used, and a guess based on an em dash is still a guess.
What search engines look at
Google has said its focus is “on the quality of content, rather than how content is produced”. It treats content as spam when automation, including AI, is used mainly to manipulate search rankings. Vague or inaccurate pages written for search engines hurt a business website however they were produced.
Using AI without losing the human part
ALPHA V3 uses AI, and we don’t hide it. It speeds up parts of design and development, such as early layout options, first drafts of code, checking a page against our standards, and turning rough notes into a structure worth editing. The time it saves goes into the work that needs a person.
That work is where a website gets its value. Someone has to decide what the business needs the site to do and which claims the client can stand behind. Someone has to pick a real photo of the team over a generated one and notice when a draft doesn’t sound like the owner. When something is wrong, someone has to answer for it. AI can produce a page, but it has never met your customers.
Making your content read as yours
- Put in the details only you have: how you price, real timelines, where you work, and the questions customers actually ask.
- Show real people and real work. A photo of your crew on a job does more for trust than a polished image of strangers.
- Cut sentences that could sit unchanged on a competitor’s site.
- Read the page out loud. If it sounds like nobody in particular, customers will hear that too.
- If AI writes the first draft, edit it as the owner: check every fact, add what only you know, and take out anything you wouldn’t say yourself.
As AI becomes ordinary, readers are likely to care less about whether it was involved and more about whether the page is right about your business.
If your website could belong to any business in your field, contact ALPHA V3 to talk about a refresh.
Sources
- Google Search Central: Google Search’s guidance about AI-generated content (February 2023)
- Decrypt: OpenAI quietly shutters its AI text classifier due to low accuracy (July 2023)
- Liang et al.: GPT detectors are biased against non-native English writers (2023)
- Different Time, Different Language: Revisiting the Bias Against Non-Native Speakers in GPT Detectors (2026)
- Fotoware: What is C2PA?
