AI-sounding post checker

Q: What are the biggest AI writing tells on LinkedIn?

Filler openers ('In today's fast-paced world…'), essay constructions ('It's important to note', 'not only… but also'), enumerator adverbs starting sentences (Firstly, Moreover, Furthermore), tell-vocabulary (delve, tapestry, testament, seamless, pivotal), em-dash overload, perfect three-item lists, rocket-and-sparkle emoji, and a suspiciously uniform sentence rhythm — every sentence roughly the same length.

Paste your draft and see the exact phrases, constructions and rhythms that make readers smell AI — with every match quoted — plus a specificity score for the things human writing has that AI text lacks. No detector theater: this checks tells, not authorship.

● free · no login · runs in your browser

Your post

How to use it

Paste the post you're worried about. Nothing you paste leaves your browser.
Hit Check my post. You get a verdict, every flagged phrase quoted with a count, and a specificity score.
Don't just paraphrase the flags — add what's missing: real numbers, real moments, what someone actually said. Then re-check.

There is no reliable AI detector — including this one

Let's be honest about what this tool is. AI detectors false-positive on human writing and miss lightly edited AI text; nobody has a reliable one. But that's fine, because your readers don't run detectors either. They react to two things: surface tells they've learned to associate with AI (the "delve"s, the "In today's fast-paced world"s, the suspiciously even rhythm), and — more damning — what's missing: specifics, stakes, a real voice. This tool checks both. It quotes every tell it finds, and it counts the specifics AI text characteristically lacks.

Does AI-sounding content actually cost you? The best data point we have: an Originality.AI study of 3,368 LinkedIn posts (2025) found detectably-AI posts underperform human writing in most professional niches. Not because of some confirmed algorithmic penalty — the simpler explanation is that generic posts earn less human engagement, and the feed amplifies engagement. We wrote up the full picture in Does AI content hurt LinkedIn reach?

The fix is adding what's missing, not paraphrasing the tells away

Here's the trap most people fall into after a check like this: they thesaurus-swap the flagged phrases and end up with different-sounding generic text. The post still smells wrong, because the real tell was never "delve" — it was that nothing in the post could only have been written by you.

Human writing about real work carries fingerprints a model can't fake: the actual number ("$4,200, not 'significant revenue'"), the actual timeline ("nine days of silence"), the sentence the customer actually said, the part where you were wrong. That's what the specificity score measures. If your post has zero numbers, zero first-person past-tense moments and zero quoted speech, that absence is what readers are reacting to — whatever the phrase checks say. The full method is in AI posts that don't sound like AI: draft with AI if you want, but feed it your real material first. That's how Liftli works — it drafts from your voice notes, calls and commits in a voice extracted from your own writing, so the specifics are there from the start. And once the voice is fixed, run the mechanics through the post analyzer.

For AI agents

Using Claude Code or another agent to draft posts? Install the skill version — it runs the same tell-and-specificity checks on any draft in your terminal, so your agent can flag its own AI-isms before you ever see them:

npx skills add liftli-ai/skills --skill ai-sounding-post-checker

Part of the liftli-ai/skills collection — browse all 28 skills, one per tool on this site. For the full pipeline (voice extraction, strategy memory, publishing), connect the Liftli MCP.

Frequently asked questions

Can any tool reliably detect AI-written text?

No — and this one doesn't claim to. AI detectors produce false positives on human writing and miss lightly edited AI text; there is no reliable detector. What this tool checks is something different and more useful: whether your post carries the surface tells that make human readers feel "this smells like AI" — because readers don't run detectors either, they react to patterns and to what's missing.

What are the biggest AI writing tells on LinkedIn?

Filler openers ("In today's fast-paced world…"), essay constructions ("It's important to note", "not only… but also"), enumerator adverbs starting sentences (Firstly, Moreover, Furthermore), tell-vocabulary (delve, tapestry, testament, seamless, pivotal), em-dash overload, perfect three-item lists, rocket-and-sparkle emoji, and a suspiciously uniform sentence rhythm — every sentence roughly the same length.

Does AI-generated content hurt LinkedIn reach?

The best data point we have: an Originality.AI study of 3,368 LinkedIn posts (2025) found detectably-AI posts underperform human writing in most professional niches. That's correlation on detectable posts, not proof of an algorithmic penalty — the simpler explanation is that generic-sounding posts earn less engagement from humans, and the feed amplifies engagement.

How do I make an AI-drafted post sound human?

Not by paraphrasing the flagged phrases — a thesaurus pass produces different-sounding generic text. Add what AI text lacks: the real number, the real date, the moment it happened, the sentence the customer actually said, the part where you were wrong. Those specifics are things a model couldn't have known, and readers register them instantly as a real voice.

Is my post uploaded anywhere when I check it?

No. Every check runs in your browser with plain JavaScript — nothing you paste leaves your machine.

Will removing the flagged phrases fix my post?

It removes the smell, not the emptiness. A post can pass every phrase check and still read as AI because nothing in it could only have been written by you. Use the specificity score as the real target: if your post has no numbers, no past-tense first-person moments and no quoted speech, that absence is the tell.