Two ways 'collecting buyer language' stops being buyer language

The symptom

A coded corpus of buyer language exists, built from reviews and conversations, and the resulting copy still doesn't sound like something a real buyer would say.

The two ways this breaks

• Paraphrasing while collecting. A researcher's own cadence creeps into the notes without anyone noticing, so the 'buyer's words' in the corpus are quietly the researcher's words wearing the buyer's meaning. Reading ten rows aloud and listening for your own rhythm sneaking back in is the check that catches this.
• Saturation theatre: stopping data collection once themes stop producing anything new, commonly around sixty quotes. That's a legitimate signal that the theme itself is covered, but it tells you nothing about the causal story behind any one instance of it. The fix isn't thirty more quotes on the same theme. It's three more actual conversations that can explain why.

The fix

Read ten rows of your corpus aloud as a standing check for paraphrase creep, done deliberately rather than once in passing. When a theme has clearly saturated, don't keep harvesting more of the same quote type; go have a small number of real conversations instead, aimed specifically at the causal question the quotes can't answer on their own.

How to spot it

Pick ten rows from your own buyer-language corpus and read them aloud. If they sound like you, not like the forum thread or review they came from, paraphrase has crept in. If your corpus is full of quotes repeating the same theme but you can't explain why any single buyer actually felt that way, you've stopped at saturation theatre instead of understanding.