Why does my content look AI-generated?
The specific patterns that make writing read as machine-produced, and what to do about them — whether or not a model was involved.
Usually a cluster of habits rather than any single tell: hedged openings that say nothing, tricolon sentences, transition words at the start of every paragraph, a conclusion restating the introduction, and an absence of specifics. Human writing shows these too — which is exactly why the fix is the same either way.
The recognisable patterns
| Pattern | Example | Fix |
|---|---|---|
| Throat-clearing opener | "In today's fast-paced digital landscape…" | Delete it. Start with the answer. |
| Rule of three, endlessly | "efficient, scalable and reliable" | Pick the one that is true and say why. |
| Signposting every paragraph | "Furthermore… Moreover… Additionally…" | Cut most of them. The order should carry the argument. |
| Restating conclusion | "In conclusion, as we have seen…" | End on the last real point, or on what to do next. |
| Empty intensifiers | "crucial", "vital", "essential", "robust" | Replace with the specific consequence. |
| Balanced non-answers | "There are pros and cons to consider" | Give a recommendation, or explain what the choice depends on. |
| No specifics | No numbers, names, versions or dates anywhere | Add them. This is the biggest single difference. |
Why it matters beyond taste
- Readers disengage. The patterns signal that a page will take 800 words to say what fits in 100.
- Search systems increasingly weight helpfulness and originality, and generic content is what those signals are designed to demote.
- Answer engines cannot extract a quotable passage from text with no concrete claims — there is nothing to quote.
- It reads as low effort, which transfers to judgements about the organisation.
The underlying problem is usually not the model
Most writing that reads as generated has a simpler cause: the writer had nothing specific to say. A model asked to write about a topic with no source material produces exactly this, and so does a person in the same position.
So the fix is upstream. Before editing sentences, ask what this page knows that a generic article does not — your data, your testing, your customers' actual questions, the thing you got wrong last year. If the answer is nothing, the page should not exist.
A concrete editing pass
- Delete the first paragraph. Nine times in ten the second one is the real opening.
- Search for "In today's", "In conclusion", "It is important to note", "Furthermore", "delve", "landscape", "tapestry", "leverage". Remove or replace each.
- Find every three-item list of adjectives and cut it to the one that carries information.
- Add at least three specifics per section — a number, a name, a version, a date, a real example.
- Read it aloud. Anywhere you would not say it that way to a colleague, rewrite it.
- Cut the length by a quarter. It will improve.
What our audit reports about this
Every item below is measured directly, not inferred. Run it against your own site and the result names the exact rule or header responsible.
- Frequency of formulaic openers, transitions and hedging phrases, reported with counts and examples.
- Density of concrete specifics — numbers, dates, proper nouns — relative to page length.
- Repetition of phrasing and sentence-shape uniformity.
- Whether the page states anything not available in a generic article on the subject.
For agents and scripts, the same measurement is at
/api/v1/slop?url=yoursite.com —
see the API documentation.
Related questions
Will Google penalise AI-generated content?
Google's stated position is that it penalises unhelpful content regardless of how it was produced. In practice the patterns above correlate strongly with unhelpfulness, which is why generated content so often underperforms — the method is not the issue, the emptiness is.
Can detection tools tell?
Not reliably, in either direction. They produce false positives on human writing — particularly from non-native speakers and technical writers — and miss edited machine output. Do not make decisions based on a detector score.
Is it fine to use AI for drafting?
Yes, if you supply the substance and edit properly. The failure mode is asking for content on a subject you have no material about, then publishing it unchanged.
Read next
How do I write content that AI assistants will quote?
Assistants extract passages, not pages. Everything follows from that one fact.
ReadHow many words should a page have?
There is no target. Word count is not a ranking factor, and chasing one reliably makes pages worse.
ReadHow do I make my website look trustworthy?
The signals people and machines actually use — most of which cost nothing and are missing from most small sites.
ReadDoes my about page actually matter?
It is one of the most visited pages on most sites, and one of the few things both people and machines use to judge whether you are real.
Read