Help Centre
Everything you need to know about how VEYL detects AI-generated writing, what your results mean, and how to get the most out of the tool.
VEYL is an AI writing detection tool that analyses text for patterns commonly found in machine-generated content. It uses a multi-module pipeline — up to six independent detection methods running in parallel — to produce a probability score and detailed flagged findings. It is built for writers who want to prove their work is original, and educators who need to verify student submissions.
There are two ways to submit text for analysis:
Click Scan Text (or the upload triggers automatically) and results appear within seconds.
No — you can run one free guest scan without creating an account. After that, sign up for a free account to get 15,000 words per month with Standard Scan. Creating an account also unlocks scan history, so you can revisit and compare results over time.
VEYL accepts .docx, .doc, .pdf, .rtf, .txt, .odt, and .md files. For the best experience with annotated exports (where flagged sections are highlighted in your original document), we recommend uploading .docx files.
The score is a weighted average of all active detection modules. It represents the estimated probability that the text was generated by AI. A higher score means more modules found patterns typically associated with machine-generated writing.
| Score Range | Rating | What It Means |
|---|---|---|
| 0% – 25% | Low Risk | The text shows strong human writing characteristics. Very few or no AI patterns detected. |
| 26% – 59% | Review | Mixed signals. Some AI-like patterns found alongside human characteristics. Warrants closer inspection of flagged items. |
| 60% – 100% | High Risk | Strong AI patterns detected across multiple analysis dimensions. Text is likely AI-generated or heavily AI-assisted. |
Module Agreement shows the range of scores across all active detection modules. If every module returns a similar score (e.g., all between 45% and 55%), that is strong agreement and the result is more reliable. If one module says 10% and another says 80%, agreement is low, which means the text has mixed characteristics and the overall score should be interpreted with more caution.
In this example, the lowest module scored 30% and the highest scored 51%. The spread is moderate, indicating reasonable agreement between detection methods.
Not exactly. The score is a probability estimate, not a measurement of how much text is AI vs human. A 55% score means the overall pattern analysis suggests the text is somewhat more likely to contain AI-generated content. It does not mean half the sentences were written by AI. Look at the individual flagged items to understand which specific patterns triggered the score.
No AI detector is 100% accurate. Several factors can influence scores:
Always review the flagged items for context rather than relying on the score alone.
Text uses 7.4 vague/abstract phrases per 1000 words vs 0.0 specific/concrete details. AI text tends to favour abstract generalities over concrete specifics.
Replace vague phrases with specific details — exact numbers, names, sensory descriptions, and concrete examples.
Very few contractions found. AI-generated text tends to use formal, uncontracted language ("it is" instead of "it's").
Add natural contractions where appropriate to make the writing feel more human.
Sentence length variation is slightly below typical human range. AI text sometimes produces more uniform sentence lengths.
Vary your sentence lengths more — mix short punchy sentences with longer, more complex ones.
Flags are specific patterns found in your text that are associated with AI-generated writing. Each flag includes a severity level, a description of what was detected, and a practical suggestion for addressing it.
VEYL checks for a wide range of AI writing patterns, grouped by detection module. Common flag types include:
Only if you want to. The suggestions are guidance for writers who want to reduce their AI score — for example, authors submitting to publishers who run AI checks. They highlight patterns that look AI-like, even if they appear in your genuine human writing. If you are confident your text is original, the suggestions can be ignored.
Not necessarily. Some human writing naturally shares patterns with AI output — especially formal academic prose, technical documentation, or carefully structured essays. Individual flags are just data points. What matters is the overall score, the number and severity of flags, and whether multiple modules agree. A few LOW or MEDIUM flags on otherwise original text is completely normal.
VEYL uses up to six independent detection modules, each analysing a different dimension of the writing. The Standard Scan (free) runs four modules; Deep Scan (Pro plan) runs all six.
Standard Scan modules (included free):
Deep Scan modules (Pro plan):
Each module analyses a different aspect of the writing. A text might have very natural vocabulary (low heuristic score) but statistically uniform sentence structure (higher statistical score). Disagreement between modules is actually valuable information — it shows which specific dimensions of the writing triggered detection and which did not.
The overall AI Probability score is a weighted average of all active module scores. Modules with higher confidence in their analysis receive more weight. The exact weighting formula is adjusted to minimise false positives while maintaining detection sensitivity.
Standard Scan runs four local detection modules (heuristic, statistical, stylometric, and hidden characters). It's fast, private, and included on all plans including Free.
Deep Scan adds two additional cloud-powered modules (semantic AI analysis and psycholinguistic profiling) for a total of six modules. It provides deeper analysis and is more accurate on edge cases, but requires the Pro or Team plan.
Absolutely. Standard Scan catches the most common AI writing patterns and is sufficient for most use cases. The four modules cover vocabulary, statistics, style, and hidden characters — the areas where AI text is most consistently distinguishable from human writing. Deep Scan adds nuance for difficult cases, but Standard Scan alone provides strong detection.
VEYL offers two export options:
Exported documents use colour-coded highlighting to show flag severity directly in the text:
Each highlighted section includes a margin comment (in Word) or inline label (in PDF) explaining what was flagged and suggesting how to address it.
Yes. The exported PDF or Word document is a standalone file that you can share with anyone — an editor, a teacher, a publisher, or a client. It contains all the analysis results without requiring access to VEYL.
VEYL's multi-module approach significantly reduces both false positives (flagging human text as AI) and false negatives (missing AI text) compared to single-method detectors. When multiple modules agree, accuracy is very high. However, no AI detector is 100% accurate — the technology is probabilistic by nature.
Light surface editing (changing a few words, rearranging sentences) may alter some indicators, but VEYL's deeper analysis modules — statistical distributions, perplexity, stylometric fingerprinting — look at patterns that are very difficult to eliminate through simple editing. Heavy rewriting that removes these patterns would essentially mean writing new text.
VEYL detects patterns common across major language models including ChatGPT, Claude, Gemini, Copilot, and others. Because it analyses fundamental properties of AI-generated text (statistical uniformity, vocabulary patterns, structural regularity) rather than targeting a specific model's watermark, it generalises well across different AI systems.
VEYL's detection models are trained primarily on English text. Results on non-English documents may be less reliable. We recommend using it only for English-language content at this time.
For best results, submit at least 300 words. Shorter texts give the statistical and stylometric modules less data to analyse, which can reduce accuracy. The minimum is 50 characters, but results improve significantly with longer documents.
Yes — the Free plan includes 15,000 words per month with Standard Scan (four modules). No credit card required. It is a genuine free tier, not a time-limited trial.
See the Pricing page for full details and current offers.
Yes. All plans are month-to-month with no long-term commitment. Cancel from your account settings and you retain access until the end of your current billing period.
You will be prompted to upgrade to a higher plan. Scans are not interrupted mid-analysis — if a document puts you over the limit, it will still complete, but you will not be able to start new scans until the next billing cycle or until you upgrade.
Your documents are processed in memory and are never stored permanently on our servers. We take privacy seriously — your text is not used for training, is not shared with third parties, and is deleted after analysis completes.
No. Your text is never used for training, fine-tuning, or improving any AI model. It is processed solely for your analysis and then discarded.
Standard Scan modules (heuristic, statistical, stylometric, hidden characters) run on our servers but do not send your text to any third-party AI service. Deep Scan's semantic analysis module sends text to a cloud AI provider for processing — this is clearly indicated in the interface, and you can choose not to use it.
Check that your file is one of the supported formats (.docx, .doc, .pdf, .rtf, .txt, .odt, .md) and is not corrupted. Files with heavy formatting, embedded macros, or DRM protection may fail to process. Try converting to .docx or pasting the text directly.
Standard Scan typically completes in under 10 seconds. Deep Scan can take 15–30 seconds due to the additional cloud-based analysis. Very long documents (10,000+ words) may take slightly longer. If a scan exceeds 60 seconds, try refreshing the page and scanning again.
Standard Scan results are deterministic — scanning the same text twice should produce identical results. Deep Scan includes an AI model component that may produce very slight variations between runs (typically within 1–2 percentage points). If you see large differences, please contact support.
Email us at support@veylapp.com. We aim to respond within 24 hours on business days.