AI tell · grade B
Let's dive in and deep dive: one real tell, one refuted one
'Let's dive in' and 'deep dive' get treated as the same red flag on most checklists. They are not the same tell. 'Let's dive in' as a transition into content, right after a compliment on the question, is a genuine chatbot leftover, part of the same family of assistant preamble we measured in 12 of 12 model logs. 'Deep dive,' 'circle back,' and the rest of the business-jargon list are not: our social-sentiment research scored that whole family as the weakest tell in the corpus, refuted because a workplace blog was mocking the same words years before mainstream chatbot use. underslop's checklist still tracks 'deep dive.' It never counts it.
What it is
Two different phrases share a family resemblance and get flagged together. 'Let's dive in' is a spoken transition a chatbot uses to hand off from its own preamble into the actual answer, doing the same job as an opening line that praises the question or a closing line that offers to keep helping. 'Deep dive,' 'dive deep,' and neighbors like 'circle back' and 'synergy' are meeting-room vocabulary that existed long before language models did. underslop's checklist keeps a list of that business-jargon family and computes it on every draft, but marks it refuted, so it never adds to a score.
How often it shows up
We did not log the literal words 'let's dive in' in our own 12-model corpus. What we did log is the family it belongs to: a validation opener before the content in 12 of 12 models (research/reports/ai-tells-empirical.json, entry emp-001), and an engagement hook after it in 9 of 12 (emp-007). 'Let's dive in' sits in the handoff between those two. For 'deep dive' specifically, our social-sentiment review gave the whole business-jargon family a sentiment strength of 2 out of 5, the lowest of any pattern in that corpus, and traced the same complaints to a corporate-communications blog post from before ChatGPT existed.
Why readers flag it
A reader flags 'let's dive in' for the same reason they flag a chatbot's opening compliment on the question: it is addressed to whoever typed the prompt, not to whoever is reading the finished page. A reader who flags 'deep dive' is usually pattern-matching on a checklist, not on anything the sentence actually does. The word predates the chatbot by decades and shows up just as often in a Tuesday standup.
The fix
Cut the handoff. If a paragraph opens with praise for the question and then pivots with 'let's dive in' or 'let's take a closer look,' delete both and start with the first real sentence. Leave 'deep dive' alone unless it is doing no work in the sentence. It is ordinary usage, not a symptom.
Examples
Great question! Let's dive in and unpack what's really going on with your churn numbers.
Churn is telling you something. The question is whether you're willing to hear it in time.
Before we dive in, let's take a step back and look at the bigger picture. Happy to help however I can!
Let's look at the bigger picture first.
Questions
What are some common phrases ChatGPT uses?
The most consistent ones in our own logs are an opening line that praises or agrees with the question, a closing line that offers to revise or expand, and the handoff phrase that sits between them, 'let's dive in' or 'let's take a closer look.' All three showed up in most or all of the 12 models we sampled.
What does 'let's deep dive' mean?
It means the same thing 'let's look closely' does. 'Deep dive' is decades-old business shorthand for a detailed look at one topic, and using it does not make a sentence more or less likely to be machine-written.
Is it obvious when someone uses ChatGPT?
Sometimes. A leftover opening compliment or a closing offer to keep helping is close to certain proof, because a finished document has no one to compliment and no follow-up to volunteer. A single business idiom like 'deep dive' proves nothing on its own.
What words give ChatGPT away?
The handoff phrases addressed to a prompt rather than a reader are the reliable ones: an opening compliment, an offer to keep helping, an invitation to ask for more. Vocabulary alone, like 'deep dive,' is weak evidence, because people used it before chatbots existed.
underslop marks this tell and the rest of the consensus set in any draft you paste, using the same open lint that scores it here. The free tier caps at 500 words a run, and your text is held for 0 seconds.