# How to Check if Text Is AI Generated: A 7-Step Checklist

> Want to check if text is AI generated? Use this 7-step checklist: manual signs, two or more detectors, sentence highlights, version history and a fair talk.

```yaml
url: "https://humanizego.com/blog/how-to-check-if-text-is-ai-generated"
publisher: "HumanizeGo"
category: "Guides"
published: 2026-11-12
updated: 2026-11-12
```

## Key takeaways

- To check if text is AI generated, combine a careful read, at least two AI detectors and context such as drafts or version history.
- Common manual signs include stock phrases, uniform sentence rhythm, generic examples and confident claims with no specific sources.
- Sentence-level highlights are more useful than a single score because they show which parts of the text look machine-written.
- Version history and a direct, fair conversation with the writer are often stronger evidence than any detector result.
- No single AI detector score is proof, because every detector produces false positives and false negatives.

To check if text is AI generated, read it for common AI patterns, run it through at least two different AI detectors, look at which sentences get flagged, and then confirm with context like drafts, version history or a conversation with the writer. No single step is proof on its own. Put together, they give you a fair, defensible answer instead of a guess.

Below is the checklist we'd use ourselves, whether you're a teacher, an editor, a client reviewing freelance work, or someone wondering if a viral post was written by ChatGPT.

## What should you decide before you check if text is AI generated?

It helps to be clear about the stakes. Checking a blog post you're about to publish is very different from deciding whether a student broke an academic integrity rule.

The higher the stakes, the more weight you should put on context and the less on software. Even Turnitin, which sells AI writing detection to schools, [advises educators](https://www.turnitin.co.uk/blog/understanding-the-false-positive-rate-for-sentences-of-our-ai-writing-detection-capability) to use its highlights to start a conversation, not to draw a conclusion.

## Step 1: Read it yourself for common AI patterns

Start with your own eyes. Modern AI writing is fluent, but it has habits. Look for:

- **Stock phrases.** "It's important to note," "plays a crucial role," "in today's fast-paced world," "a testament to," "delve into."
- **Even rhythm.** Sentence after sentence of similar length and shape, with few short punchy lines or long winding ones.
- **Template structure.** Every paragraph opens with a topic sentence, lists three things, and ends with a tidy summary.
- **Generic examples.** Claims like "many businesses have seen great results" with no names, numbers or dates.
- **Confident claims with no sources,** or citations that don't exist when you search for them.
- **Missing lived detail.** No personal experience, opinions, local knowledge or small imperfections.
- **A tone that doesn't fit.** Polished, neutral prose from someone who normally writes casually.

None of these prove anything. Plenty of people write formally, and plenty of AI text has been edited to remove these traces. Treat this step as forming a hypothesis.

## Step 2: Run the text through an AI detector

Next, paste the text into an AI generated text detector. Our free [AI detector](/ai-detector) works without an account and gives you an overall AI percentage, a verdict (likely human, mixed or likely AI), and a breakdown of the signals behind the score.

A few tips for a cleaner result:

- **Check enough text.** A few paragraphs gives the detector much more to work with than a single sentence.
- **Remove things that aren't prose.** Headings, reference lists, tables and code can confuse some tools.
- **Check the original.** If you've already edited the text, you're testing your edits as well.

If you want to understand what the score is based on, our guide on [how AI detectors work](/blog/how-do-ai-detectors-work) explains perplexity, burstiness and classifier models in plain terms.

## Step 3: Run a second (and third) detector

This is the step most people skip. Different detectors are trained on different data and use different thresholds, so they often disagree.

Use at least two tools built in different ways. Then compare:

| What you see | What it likely means |
|---|---|
| All tools say likely AI, with similar sentences flagged | Stronger signal worth following up |
| All tools say likely human | Weak reason to suspect AI from the text alone |
| Tools disagree widely | The text is in a gray zone; rely on context |
| Only a few sentences flagged | Possibly mixed or lightly edited text, or a false positive |

Our roundups of the [best free AI detectors](/blog/best-free-ai-detectors) and [best ChatGPT detectors](/blog/best-chatgpt-detectors) can help you pick a second tool. If you're checking student work, see our list of the [best AI detectors for essays](/blog/best-ai-detectors-for-essays).

## Step 4: Look at the sentence highlights, not just the score

An overall percentage hides a lot. A 50% score could mean every sentence is borderline, or that half the text is clearly AI and half is clearly human. Those are very different situations.

Sentence highlighting shows you where the signal is. In our detector, each sentence is shaded as "Likely AI" or "Possibly AI," and you can hover over a sentence to see its own likelihood. Ask yourself:

- Are the flagged sentences clustered in one section, like the intro or conclusion?
- Are they the generic, summary-style sentences, or the specific ones?
- Do the flagged parts match the patterns you spotted in Step 1?

Turnitin's own published data is a good reminder to read highlights carefully. The company has said its sentence-level false positive rate is around 4%, and that false positives are more common in mixed documents, especially near the transitions between human and AI writing ([Turnitin, 2023](https://www.turnitin.co.uk/blog/understanding-the-false-positive-rate-for-sentences-of-our-ai-writing-detection-capability)).

## Step 5: Check version history and drafts

If the text was written in Google Docs, Microsoft Word with OneDrive, or a similar tool, version history is often the most useful evidence you can get. It shows how the document grew over time.

What to look for:

- **Gradual writing** over several sessions, with typos fixed and sentences reworked, suggests a human process.
- **Large blocks pasted in at once** can point to text copied from somewhere else. That could be an AI tool, but it could also be the writer's own notes from another app.
- **Supporting material** like outlines, research notes, earlier drafts and source lists.

Keep in mind that version history can be missing for innocent reasons. Some people draft in a notes app or offline and paste the final version in.

## Step 6: Ask the writer, fairly

A short, non-accusatory conversation often settles the question faster than any tool. You might ask:

- "Can you walk me through how you wrote this?"
- "Where did this example or statistic come from?"
- "What would you change if you had another day on it?"

Someone who wrote the piece can usually talk about their choices, sources and the parts they struggled with. Keep the tone curious, not prosecutorial. Remember that the writer may have used AI in ways that are allowed, like brainstorming or grammar checks, depending on the rules that apply.

## Step 7: Decide based on all the evidence

Finally, put the pieces together. A reasonable rule is that you need more than one kind of evidence before acting on a suspicion: for example, consistent detector flags *plus* missing drafts *plus* an inability to explain the work.

Write down what you found and how. If the outcome affects someone's grade, job or reputation, a clear record protects everyone.

## Why isn't a single AI detector score proof?

Because every detector makes mistakes in both directions. It can flag human writing as AI (a false positive) and miss AI writing that has been edited or paraphrased (a false negative).

The evidence on this is well documented:

- OpenAI shut down its own AI text classifier in July 2023, saying it was "no longer available due to its low rate of accuracy" ([Inquirer Technology](https://technology.inquirer.net/126335/openai-shuts-down-ai-detector)).
- A Stanford study in *Patterns* found seven detectors flagged a majority of essays by non-native English writers as AI-generated, while handling native speakers' essays well ([Liang et al., 2023](https://pmc.ncbi.nlm.nih.gov/articles/PMC10382961/)).
- The [RAID benchmark](https://arxiv.org/abs/2405.07940) found detectors were easily fooled by adversarial attacks and by models they hadn't seen.

Detectors have improved, and some newer independent tests show low false positive rates for the best tools. We cover both sides in [Are AI detectors accurate?](/blog/are-ai-detectors-accurate). The fair conclusion is that a detector score is a useful signal for deciding where to look, not a verdict.

## What if your own writing gets flagged?

It happens, and it's frustrating. If a detector flags work you wrote yourself:

1. Gather your drafts, notes, sources and document version history.
2. Run the text through a couple of other detectors to see whether the result is consistent.
3. Look at which sentences were flagged. They're often the most generic or formulaic lines.
4. Explain your process calmly and point to the evidence.

If your writing tends to read stiff, revising for varied sentence length and more specific detail makes it better for human readers too. Our guide on [how to humanize AI text](/blog/how-to-humanize-ai-text) covers editing techniques, and our [humanizer](/) can help you rework the flow of a draft. No tool can guarantee a particular detector result, and students should follow their school's AI policy.

Ready to try the checklist? Paste a few paragraphs into our [free AI detector](/ai-detector), review the sentence highlights, then compare with a second tool. If you need more checks per day, see [pricing](/pricing).

## FAQ

### Is there a way to know for sure if AI wrote something?

Usually not from the text alone. Detectors give probabilities, and both people and tools make mistakes. Drafts, version history or the writer's own explanation are the closest you'll get to certainty.

### What is the best free way to check if AI wrote this?

Run the text through two or more free AI detectors, compare the sentence-level highlights, and read it yourself for common AI patterns. Our free AI detector shows sentence highlighting without an account.

### How many words do I need to check if content is AI generated?

More is better. Most detectors become less reliable on short passages, so try to check at least a few paragraphs rather than a single sentence.

### Can AI generated text detectors be fooled?

Yes. Research has shown that paraphrasing and editing can lower detector scores, and that detectors struggle with models they weren't trained on. That's another reason not to rely on a single score.

### What should I do if a detector says my own writing is AI?

Stay calm and gather your evidence: drafts, notes, sources and document version history. Explain your process, and point out that detectors are known to produce false positives.

---
HTML version: https://humanizego.com/blog/how-to-check-if-text-is-ai-generated