Andreessen Horowitz is reframing the debate around AI-generated writing, arguing that obsessing over whether text is human-written or machine-generated misses the point entirely. Instead of hunting for telltale signs of artificial prose, the venture firm suggests founders and communicators focus on whether writing actually does its job: explaining ideas clearly, building reader trust, and demonstrating that someone is making deliberate decisions behind the words.

What Are the Real Problems With AI Writing?

The a16z crypto team identified a taxonomy of writing problems that have plagued human communication long before large language models (LLMs) existed. Rather than treating AI-generated text as inherently flawed, the firm's editors argue that machine prose simply makes these longstanding issues easier to spot. The goal isn't to shame AI writing, but to use it as a diagnostic tool for improving communication across the board.

One of the most insidious problems is what a16z calls "semantically vacant connective tissue." These are phrases that sound weighty but collapse under scrutiny. Examples include "We're excited to announce," "We're at an inflection point," and "This is the next chapter of our journey." Before these became associated with AI, they were simply called corporate jargon. The issue isn't that they're wrong, exactly; it's that they're first-draft language that should never survive editing.

How to Identify and Fix Common AI Writing Habits

• Naturally Essenced Profundity: Phrases like "Something real is happening" or "The stakes couldn't be higher" sound important but are actually filler. Test whether the sentence means anything by trying to paraphrase it. If it boils down to "stuff exists," delete it.
• Empty Contrasts: The pattern "It's not just about X, it's about Y" often works only when the contrast is crystal clear. Before keeping it, check whether you can delete the first half and reach your point faster without losing meaning.
• Excessive Hedging: Words like "arguably," "in many ways," and "at some level" exist to make sentences impossible to be wrong about. While some industries like finance or healthcare require hedges for legal compliance, most writing benefits from removing them entirely.
• Too Much Parallelism: When every bullet point has the same grammatical structure and length, or when sentence constructions mirror each other too tidily, the result is boring writing that feels robotic, whether human or machine-written.
• In-Summation Phrases: Expressions like "At the end of the day" or "When the dust settles" almost always can be deleted with zero impact on the paragraph's meaning or clarity.

The a16z team emphasizes that meaningless, generic language is "at the top of our stack-ranked list of writing pitfalls." It will never move ideas forward, no matter how polished it sounds. The solution is ruthless elimination.

One particularly revealing example comes from a16z's own testing. When asked about punctuation, an AI model responded: "Human punctuation has a body; it fidgets. My punctuation is uniformly deliberate, and uniform deliberateness is itself the rhythm." The editors describe this tone as "person at an otherwise nice party talking about jazz improvisation." The irony is that humans write sentences exactly like this all the time, without any AI involvement. The problem isn't the origin; it's the substance.

What Practical Strategies Help Improve Writing Quality?

A16z recommends several concrete approaches for founders and communicators looking to improve their writing, whether they're using AI tools or not. The first is creating a rough style guide that defines what "good" writing looks like with specific examples and counterexamples. This approach holds up better than simply memorizing a checklist of AI tells.

The second strategy is using AI itself as an editing tool. Rather than relying on models to generate final copy, use them to run what a16z calls the "paraphrase test." Ask the LLM to write a "boring" version of what you've written. If the boring version is clearer, you've identified where your original prose was obscuring rather than illuminating. Another blunt-force approach is prompting the model to "write it at a 6th grade level," which forces unnecessary jargon and complexity out of the text.

The third insight concerns vocabulary. Default AI writing operates with a surprisingly limited set of words, roughly 400 of the million available in English. Rather than trying to pinpoint which specific words are AI hallmarks, a16z suggests testing for "fungibility." Could a sentence be lifted word-for-word from your writing and dropped into someone else's essay on a completely different topic without anyone noticing? If yes, it's probably too generic.

Researchers have tracked this phenomenon empirically. After ChatGPT launched in 2022, the word "delve" spiked dramatically in academic abstracts, becoming a telltale sign of AI-assisted writing. But the real lesson isn't to avoid "delve"; it's to recognize that when everyone reaches for the same word, the writing becomes interchangeable and forgettable.

A16z's core argument challenges the entire framing of the AI writing debate. Rather than asking "Is this human or machine-written?" the firm suggests asking whether the writing is doing its job. Does it explain things clearly? Can readers trust it? Is someone making deliberate decisions? These questions apply equally to writing produced by humans, models, or the increasingly entangled combination of both that characterizes modern professional communication.