Discerning AI "Writing": A Practical Guide for Writers
We need to relearn a basic rule: don't believe everything you see online. AI systems are pumping out posts that feel human enough to spread fast, and most people share before they think.
This isn't about "fake" versus "true." AI can be accurate. The real test is simpler and harder: can you tell if a human wrote it?
Why AI-written posts spread
- They're easy to produce: endless content with zero fatigue.
- They read smooth: clear, concise, agreeable - the default tone of a machine trained to please.
- They avoid specifics: broad stories that flatter beliefs and trigger quick shares.
- Algorithms reward engagement, not provenance.
Quick tells that a machine wrote it
- Vague facts: no named sources, dates, locations, or verifiable details.
- Over-polished voice: clean grammar, safe phrasing, no real point of view.
- Repetitive cadence: similar sentence length and structure across the piece.
- Stock empathy: "heartwarming," "inspiring," and neat moral endings without real texture.
- Generic quotes: statements that read like placeholders rather than lived language.
- Low-resolution nouns: "a hospital," "a town," "a coach," instead of proper names.
- Contradictions between paragraphs that a human editor would catch.
None of these are definitive by themselves. Also, don't lean on AI detectors - their accuracy is unreliable and can flag human work as machine-made. See: OpenAI's own note on detector limits.
How to check before you share or cite
- Find the first source: who published it first, and do they provide evidence?
- Trace claims: verify names, dates, and locations with a quick search from multiple outlets.
- Look for bylines and beats: does the writer or publication cover this topic regularly?
- Check for media: if there's a photo or video, is there an original credit and context?
- Stress test quotes: copy a sentence into search and see if it appears across many sites with no origin.
- Ask one blunt question: could a human plausibly report this, or is it stitched from feels and clichés?
If you want a structured habit, study source-checking frameworks like Civic Online Reasoning (Stanford). Five minutes of verification saves your credibility.
Write like a human (and make it obvious)
- Lead with particulars: proper nouns, timestamps, places, prices, sensory detail.
- Show work: link or cite where key facts came from.
- Let your edges show: rhythm shifts, a sharp take, and sentences that breathe.
- Include friction: a doubt, a constraint, or a small failure that shaped the outcome.
- Quote real people: conversations, not canned lines. Capture cadence, not perfection.
- Prefer field notes over summaries: what you saw, heard, or touched.
Set a simple AI policy for your practice
- Permitted uses: brainstorming, outlining, summarizing public sources you will verify.
- Prohibited uses: auto-generated stories presented as human reporting, or anything unsourced.
- Disclosure: if AI contributed words or analysis, say so.
- Proof trail: keep notes, links, and transcripts for every factual claim.
- Editor step: one human pass focused on voice and verification.
Use the tool, keep your taste in charge
AI isn't the villain. It's a calculator for language. Useful, if you set the rules and keep your standards higher than convenience.
If you want to learn how machines structure text - so you can both use them wisely and spot their fingerprints - explore prompt fundamentals and critiques of AI output patterns here: Prompt Engineering resources.
Rebuild human inputs
Put the phone down. Read a book from your grandfather's shelf. Skim the local ag paper with coffee. Ask a neighbor how they got through the 80s. Keep the Bible by your bed and read it.
Those inputs give your writing weight. Machines predict; people remember.
The upside
If AI "writing" floods the feed, maybe it forces a healthy reflex: verify before you share. As writers, that reflex isn't optional - it's the job.
What we consume shapes how we write. Choose sources that sharpen your eyes, not numb them.
Your membership also unlocks:
