Does Turnitin Detect Perplexity AI? What the Documentation Says

Turnitin's FAQ publishes a list of AI writing models it can detect content from, including GPT, Gemini, Claude, LLaMA, Mistral, Deepseek, Nova, Grok, and o1-mini. The list ends with "and tools based on these LLMs as well." The FAQ also states: "We will continue to expand our detection capabilities to other models in the future." Perplexity AI is a search tool that can generate text using underlying language models. Whether Turnitin's detector flags content from Perplexity depends on which underlying model produced the text and whether that model or its family is on the detection list. The FAQ acknowledges that detection capabilities are not static: "As we iterate and develop our model further to better detect newer LLMs, it is likely that our detection capabilities will also change, affecting the AI percentage."

The FAQ provides a list of models it can detect content from. The introductory sentence uses the verb "can detect content from," followed by a comma-separated list. The list includes major LLM families: GPT, Gemini, Claude, LLaMA, Mistral, Deepseek, Nova, Grok, and o1-mini. The list concludes:

"and tools based on these LLMs as well."

The next sentence: "We will continue to expand our detection capabilities to other models in the future."

This means the list covers not only the named models but also tools built on top of them. If a tool like Perplexity AI routes its text generation through one of the listed LLM families, the output may fall under the "tools based on these LLMs" clause. Does Turnitin detect Claude, Copilot or Gemini goes through the named entries.

The detection model processes text by splitting it into overlapping segments and assigning probability scores:

"When a paper is submitted to Turnitin, sentences from the submission are extracted and segmented into overlapping sections for prediction analysis. Each segment is classified by the AI detection model and given a value between 0 and 1, denoting the probability of the text being likely human or AI-generated."

The model is not programmed to evaluate burstiness or perplexity as named metrics: "Our model is not explicitly programmed to evaluate specific signals such as 'burstiness,' 'perplexity,' or other individual metrics sometimes referenced in public discussions." The next sentence: "Instead, it learns statistical patterns from our training data." The same FAQ states: "Our classifiers are trained to detect these differences in word probability and are adept at the particular word probability sequences of human writers." The detector evaluates word probability patterns learned from training data, not surface-level metrics. The name collision is worth clearing up on its own: does Turnitin use perplexity and burstiness.

The FAQ explicitly states that detection is not fixed:

"As we iterate and develop our model further to better detect newer LLMs, it is likely that our detection capabilities will also change, affecting the AI percentage."

The next sentence: "However, for a submitted document, the AI percentage will change only if it's re-submitted again to be processed."

This means the detection model is being updated to target newer LLMs. If Perplexity AI or any other tool uses a new model not yet on the list, there may be a gap until the next detection model update. Already-graded documents are not automatically re-evaluated.

The FAQ acknowledges a trade-off:

"In order to maintain this low rate of 1% for false positives, there is a chance that we might miss some AI written text in a document."

The detector is tuned to avoid false positives even at the cost of missing some AI text. A tool that produces text with word probability patterns closer to human writing would be harder to detect, at least until the detection model is updated. The promise to "continue to expand our detection capabilities" is the counterweight, but there is an inherent lag between new tool releases and detection coverage. The same lag drives does Turnitin detect AI humanizers.

To summarize what we have covered:

• Turnitin publishes a list of models it can detect content from, including GPT, Gemini, Claude, and others.
• The list covers "tools based on these LLMs as well," so tools routing through listed models may be detectable.
• The FAQ promises to expand detection to other models in the future.
• Detection capabilities change over time as the model is updated.
• The detector trades some missed AI text for a low false positive rate.
• Already-graded documents are not automatically re-evaluated when the model changes.

If you receive a Turnitin AI report and want to address the flagged passages, import the report and work on them. Eligible passages can be re-run at no charge.