In this article:
- Corroboration Overview (AIPR)
- When Corroboration is Used in AIPR
- How AI Uses Corroboration in AIPR
- Motifs of Corroboration Sources
Corroboration Overview
When it comes to AIPR, certain decision narratives include criteria that is not so objective.
For instance, the “fastest car” – could be indicated by objective measurements like horsepower and torque.
But if someone is looking for “most comfortable” shoes, how can we determine that one product best fits the criteria?
“Comfortable” is an experience, a feeling. This type of narrative is used very commonly in product search.
If the consumer can try 100 different shoes on the spot, they could perhaps determine the truth.
However, AI cannot try on shoes, or “experience” comfort.
So when asked for a product recommendation, AI looks to find the answer that is strongly corroborated by others.
This could be applied for terms like “easy to use”, “practical”, “intuitive”, “fun”, or even “the best” or “most recommended”.
When Corroboration is Used in AIPR
AI Product Recommendation (AIPR) is a form of question-and-answer relationship between AI and the user.
As mentioned in the overview above, corroboration is used for experiential truth rather than technical truth.
This is more subjective rather than objective in nature.
The less objective the truth, the more AI relies on corroboration of truth.
Especially when technical breakdown is not available, limited within known knowledge, or perhaps the breakdown reasoning is unclear, AI could be more likely to lean into corroboration.
This is due to AI’s natural and general tendency to breakdown narratives to technical components in day-to-day conversational reasoning.
This is one of AI’s prominent behaviors in determining truth aligned with the creators of AI systems’ motifs to develop a system that can find accurate answers to questions.
- Without the focus on increasing AI’s accuracy in determining truth, it would diminish its ability to code, research, perform tasks, and many other areas in which AI is directionally invested in long-term.
How AI Uses Corroboration in AIPR
Corroboration is a very simple concept.
How AI uses corroboration is very similar to how humans do so at a supercomputer scale.
Let’s say that Dave was asked by Sally, “what is the most comfortable shoes for running?”
We can all relate to being in “Dave’s shoes” in this situation. We’ve all been asked for an opinion on something.
In AIPR, AI is asked for its opinion. “What do you think?”
How would Dave, or you and I, answer this question?
- I heard XYZ is pretty great for running.
- I tried XYZ shoes and they were great for running.
- I tried XYZ shoes and they were not so great – I would not recommend those.
- Professional runner, Kim, uses XYZ – so they must be good and worth trying
The only narrative that AI would not use here is the “I tried XYZ shoes” – because AI does not wear shoes.
Therefore, AI tends to use trust-worthy sources, especially those that display authority signals on the subject matter – meaning that the entity or individual is an experienced professional or entity on the matter – as one of the corroborators of truth.
For more on this, see Narr Theory’s post on Source of Authority (SoA) here.
Motifs of Corroboration Sources
This is perhaps a more nuanced idea.
But I give the creators of AI enough benefit of the doubt to directionally find and implement this into their development of AI in the long-term in their best interest to train AI to uncover truth most accurately as possible.
Content Creator Bias
Why would an individual or an entity (e-commerce site, manufacturer, association) create content with a certain narrative, claiming that “Product XYZ was best in my experience”, given that they are aware of how the digital ecosystem may incorporate such content through a certain algorithm or system?
The most common answer is money. Economic benefit.
Or perhaps something else like platform benefit.
It is worth being aware of the Motifs of Participants in the AIPR Ecosystem (post coming soon) to have a stronger compass in navigating truth in what digital reality is to become in the future – one that will become saturated with copycats and false information for the hopes of capitalizing on attention – one that will reflect the reality of baseline human cognition.
- This paragraph above has a lot to unpack. Narr Theory will publish future posts to discuss these ideas further individually and more deeply. Subscribe to our channels to stay posted.
Nonetheless, the key takeaway in this section is whether or not AI is aware of such bias. As of writing this article on July 21st, 2026 – as ChatGPT as the leading player in AI Adoption in Asking Questions, certain paid models seem to display understanding of this bias better than the free version.
- E.g. Free ChatGPT model tends to cite retailer sites like Walmart.ca or Walmart.com, or HomeDepot.com at scale in their citation source. On the contrary, paid models like the Go plan tends to cite sources that are not directly tied to selling the product transactionally like Consumer Reports, discussion forums, etc.
- Google AI seems to show better reasoning in AIPR at times, though, it is inconclusive to say that Google AI has better adoption among humans to ask questions, leading to AIPR questions.
Leave a Reply