AI writing tools are everywhere, which makes it harder to tell what was written by a person and what was assembled by a model. Detection is never perfect, and none of the signs below prove anything on their own. Put a few together, though, and you start to see the pattern. Here is a practical guide you can share with your team, plus tips for keeping your own AI assisted content sounding genuinely human.

10 common signs of AI generated content

1. Overly polished sentences
   Natural writing has bumps, shortcuts, and variety. If every sentence follows the same structure, with similar length and rhythm, it can feel machine made.
2. Lack of personal content and anecdotes
   Humans sprinkle in lived detail: “here is what happened when I tried this,” names, dates, little surprises. AI often skirts specifics and sounds impersonal.
3. Similar sentence transitions
   Watch for repeated bridges like “moreover,” “furthermore,” or hedges repeated in sequence: “perhaps,” “it seems,” “it could be said that,” “maybe.” Patterns like this are a tell.
4. Light on humour and emotion
   Unless prompted carefully, AI tends to be efficient and neutral. It can read like a manual instead of a conversation.
5. Overuse of the em dash character
   Not unique to AI, and many humans love it too, but some models lean hard on it. If every other sentence leans on an em dash, raise an eyebrow. The em dash looks like this (—)
6. Overuse of the Oxford comma
   Again, not proof on its own. Consistency matters. If the voice, punctuation, and spelling wobble between styles, it can be a sign of templated text.
7. Vague citations
   Phrases like “experts say” or “a recent study found” with no source link. If you cannot verify it in two clicks, treat it as decoration, not evidence.
8. Cliché soup
   “Fast-paced world,” “unlock your potential,” “game-changer,” “revolutionise your workflow.” A couple is fine. A bowl full is not.
9. Awkward numbers
   Hyper-precise stats such as 37.41 percent with no source, or units that change mid-post. Credible data comes with context and consistency.
10. Recycled or generic examples
    Case studies with no names, dates, or outcomes. Vague scenarios that could apply to any industry. If it reads like a placeholder, it probably is.

Quick checks you can run in seconds

• Quote search: Drop a distinctive sentence, in quotes, into a search engine. If nothing credible appears, be cautious.
• Consistency scan: Look for UK vs US spelling, date formats, and punctuation styles switching mid-article.
• Source test: Follow every claim with a number or a quote. If the link does not support the point, or there is no link at all, mark it down.
• Read-aloud test: Read a paragraph out loud. If it sounds flat or robotic, your readers will feel it too.

How to keep AI assisted content sounding human

Using AI does not have to make your writing bland. It is about owning the process rather than letting the model steer.

1. Set clear instructions before you start
   In ChatGPT: click your name, choose Personalisation, then Custom Instructions. Tell it what to do and what to avoid. Specify voice, audience, tone, and any banned phrases or clichés. Ask for varied sentence lengths, fewer hedges, and minimal filler words.
2. Provide real context
   Feed it your brief, target reader, offer, and examples of your previous work. The more grounded the prompt, the more specific the output.
3. Inject lived detail
   Add your experience, data from your work, named clients (with permission), dates, and outcomes. Replace generic lines with anecdotes and lessons learned.
4. Cite properly
   If you include a stat, link the original source, not a repost. Ask the model to propose sources, then verify and swap in the real links yourself.
5. Vary the shape of the copy
   Mix short punchy lines with longer reflective ones. Use lists sparingly. Add questions. Remove repeated openers like “Moreover” or “In conclusion.”
6. Edit like a human
   Trim hedges you do not need. Replace generic verbs with concrete ones. Swap clichés for plain English. Read it aloud, then cut anything you stumble over.
7. Decide your style rules and stick to them
   Pick UK spelling, agree your date format, and make a call on the Oxford comma. Put it in a short style guide and keep it consistent across posts.

A short checklist for editors

• Can I point to at least two specifics that only a practitioner would know?
• Do claims with numbers link to credible sources?
• Does the voice match our brand style and regional spelling?
• Are sentence lengths and transitions varied?
• Did a human add or approve anecdotes and examples?

Final thought

No single signal proves a piece was written by a model. Humans can be formulaic. Models can be lively when guided well. The goal is not to play detective for its own sake, it is to produce content that readers trust: accurate, useful, and unmistakably yours. Use AI to draft and accelerate, then rely on your brain to shape, sharpen, and sign it off.

What do you see in content that makes you think “hmmm.. that’s AI”? Drop your tell in the comments.