AI-assisted practical guide. Examples are hypothetical; these are proposed editorial methods, not reported research results.
When processing interview transcripts or historical documents, it is often difficult to distinguish between a subject's literal words and a writer's summary of those words. Using an AI to separate quoted speech from paraphrase allows you to maintain the integrity of the original source while highlighting areas where the meaning might have been altered. This process is intended to help check that your final record is transparent about what was actually said versus what was interpreted.
Prompting for Distinction
To achieve this, you should provide the AI with a clear instruction to categorize text based on punctuation and attribution markers. Suggest that the AI use specific labels such as Quoted for verbatim speech and Paraphrased for summaries. To handle the difficult case of uncertain attribution, instruct the AI to flag any sentence where the speaker is unclear or where the transition from a quote to a summary is ambiguous. You might suggest the AI use a label like Uncertain for these instances. This can help prevent the AI from guessing the speaker's intent and forces it to highlight gaps in the source material.
Hypothetical example
Imagine you have a rough note that reads: Sarah said she hated the new policy and then she mentioned that the budget was too low, although it sounded like she was quoting the manager. You could use the prompt: Separate the following text into Quoted, Paraphrased, or Uncertain categories.
The AI output would look like this: Quoted: she hated the new policy Paraphrased: she mentioned that the budget was too low Uncertain: although it sounded like she was quoting the manager
Verifying the Separation
Once the AI provides the categorized list, you must check the finished result against the original source text. The primary goal is to ensure that no words were added to the quoted sections and that no verbatim speech was accidentally moved into the paraphrase category. Look closely at the Uncertain flags to see if you can resolve the ambiguity using surrounding context from the full document. Your final check should confirm that every single word inside a Quoted label appears exactly as written in the original input, without any AI-generated corrections to grammar or spelling.