AI-Powered Codes (Formally "Autocodes") aim to empower you to be able to summarize the ideas behind dataset responses as quickly as possible. Hand coding is time-consuming and cumbersome. Canvs MRX has the ability to code a considerable percentage of Open Ends with high accuracy – completely algorithmically and automatically. 

After Canvs processing, each Open End verbatim response is tagged by a set of {emotion} and {topic} tags. Verbatims that share the same combination of {emotion} and {topic} are clustered into the same AI-Powered Code.

As a part of our AI-Powered Coding feature, we've also developed "precodes". Precodes have two distinct functions:

  • First, pre-codes summarize ideas that are not tied to specific topic, like a General Emotion (Enjoyment, Dislike, Indifference, Boredom, etc), as well as high-level concepts that may be tied to multiple or varied topics/emotions, like Intent (Would Buy, Wouldn't Buy, etc), Content (Kid Friendly, Nostalgic), or Other Options (Better than Others, Prefer Others), Recommendations, and more. 

  • Second, Canvs MRX pre-codes that categorize open-ends that do not provide valuable information (No Answer, Unsure, Need More Information, Likely Spam) as well as those that may require attention from the user (Long-Winded, Dataset Issue). 

Currently, we have implemented the following automatic Nets and Codes:

No Answer

  • No Answer (n/a, no comment): Open Ends that contain text, but do not have an answer to the question being asked, for example: nothing, n/a, unsure, no comment, etc. (Note: Blank responses are not processed.)

  • Unsure (i don't know, not sure)

  • Need More Information (don't know enough about that, want to read reviews before I decide)


  • Likely Spam (a;dlskfj, ????, -----): Auto detects when respondents are trying to skip to the next question for example: asdf, x, !!!, etc.  These are not helpful responses and represent an opportunity for customers to save money by eliminating those people from future datasets and help get to insights faster by removing non answers.

  • Dataset Issue (I don't understand the question, etc)

  • Long-Winded (responses over 500 characters) : Open Ends with 500 Characters or more. Researchers will likely want to read these open ended responses manually due of how much detail they contain.

General Feelings

  • General Enjoyment (it was good, no issues)

  • General Dislike (not my type of movie, not for me, didn't like it)

  • General Indifference (meh, I don't care, could go either way) 

  • General Boredom (it was boring, nothing unique)

  • General Confusion (it was confusing, I didn't understand it)

  • General Sadness (made me cry, it looks sad)

  • General Mixed Emotion (I loved the suspense, but I hated how sad it made me)

  • Nothing (responses when asking questions like "what would you change" or "what 

  • Funny (it made me laugh)

  • Not Funny (thought it would be funny, but it was not)

  • Not Scary (it wasn't as scary as I expected it to be)


  • Would Recommend

  • Wouldn't Recommend


  • Would Watch (i will definitely watch this, i am going to go watch this, i always turn this show on)

  • Might Watch (depends on what else is on at that time, might watch it if i'm free)

  • Wait to Watch (wait to stream it, I'll watch it on DVD instead)

  • Don't Watch in Theaters (I don't see movies in theaters, I won't go to the movies anymore)

  • Would Buy (I would buy this, I always buy)

  • Might Buy (I would consider buying it if, I'd maybe buy)

  • Wouldn't Buy (I would never buy, I don't buy) 

Other Options

  • Other Options - General (there are other movies out, depends on what other shows are on)

  • Prefer Others (Would rather see other movies, I prefer to use products from another company)

  • Better Than Others (This bank is the best bank in New York, better than all other brands of toothpaste)


  • Price Conscious (when respondent is considering price, but does not explicitly state how they feel about it)

  • Expensive (too expensive, the rates are too high) 

  • Affordable (great prices, price is right, not too much money) 


  • Kid Friendly (great for kids, family friendly movie)

  • Not Kid Friendly (too violent for kids, can't take my kids)

  • Nostalgic (always watched it growing up, remember it from my childhood)

Note: When you are looking at the Codes Ranker, the numbers on the list will only add up to the Open Ends that have been Coded - either through AI-Powered Coding or Manual Code updates. Underneath the Tree Map, you will find text that tells you how many Open Ends have a code on them versus the amount that are not yet coded.

AI-Powered Coding will likely not code 100% of the Open Ends in a Dataset. The percentage of the verbatim responses that will have auto-codes will fluctuate. AI-Powered Coding aims to group together similar answers to questions. 

NOTE: If you have your own codes frames that you'd like to make sure we always look for, please reach out to We'll work with our Data Science team to make sure your unique code frames are available automatically. 

Did this answer your question?