From the article Signs of AI writing on Wikipedia:
LLM writing often puffs up the importance of the subject matter by adding statements about how arbitrary aspects of the topic represent or contribute to a broader topic. [...] LLMs may include these statements for even the most mundane of subjects like etymology or population data. Sometimes, they add hedging preambles acknowledging that the subject is relatively unimportant or low-profile, before talking about its importance anyway.
...marking a pivotal moment...
...represented a significant shift...
...was part of a broader movement...
...an important center...
...solidify its role as a regional hub...
...This etymology highlights the enduring legacy...
...Though it saw only limited application, it contributes to the broader history...
According to AI, every random factoid is weighty with historical and cultural significance, and needs to be filled with hot air before it can stand upright. No, actually, it's worse than that: not every subject gets the inflationary treatment; only some do, and the chosen are selected at random.
when asked to discuss an animal or plant species [...] LLMs also tend to belabor the species' conservation status and research and preservation efforts, even if the status is unknown and no serious efforts exist.
...It plays a role in the ecosystem and contributes to Hawaii's rich cultural heritage....
...Preserving this endemic species is vital...
"It plays a role in the ecosystem" is laughably vague, too.
AI chatbots tend to insert superficial analysis of information [...] Newer chatbots with retrieval-augmented generation (for example, an AI chatbot that can search the web) may attach these statements to named sources—e.g., "Roger Ebert highlighted the lasting influence"—regardless of whether those sources say anything close.
[...] AI chatbots also commonly exaggerate the quantity of sources that corroborate these opinions. They may present views from one or two sources as widely held (often combined with the vague attributions above), mention the existence or opinion of multiple "reviewers" or "scholars" while only citing one person, or imply that lists of examples are non-exhaustive when the sources give no indication that other examples exist.
Another way to spot AI writing is to look for telltale words that are overrepresented in LLM output. Those include:
additionally, (especially beginning a sentence), align with, crucial, delve (pre-2025), emphasizing, enduring, enhance, fostering, garner, highlight (as a verb), interplay, intricate/intricacies, key (as an adjective), landscape (as an abstract noun), pivotal, showcase, tapestry (as an abstract noun), testament, underscore (as a verb), valuable, vibrant.
Hopefully these examples demonstrate why AI is not to be trusted: its writing is optimized to sound authoritative when it is nothing of the kind. It lulls and persuades the reader with tautologies and empty generalizations reminiscent of promotional babble. And then, when you're not looking, it slips in an outright fabrication.
The whole article is worth at least a skim.