Overview
This is how Narr Theory investigates why and how a client product ultimately gets recommended or not recommended by ChatGPT from a direct product recommendation question.
Click on the below link for the full ChatGPT AI Product Recommendation (AIPR) Gap Analysis.
It is encouraged to first open the link above to get a high level and intuitive understanding of the ChatGPT AIPR Gap Analysis.
This gap analysis is much simpler when the stages of ChatGPT AIPR can be visualized step by step.
Below is the read-only version of the gap analysis. If it helps, please feel free to read the below content along with the analysis open.
How to Use The Analysis
There are approximately 8 stages of ChatGPT AIPR.
- Decision Narrative
- AI Understands User Intent
- Technical Breakdown of Decision Narrative
- Criteria Weighing
- Check Internal Knowledge Model
- External Knowledge Search
- Recommendable Candidates
- ChatGPT Product Recommendation
Here, the ultimate desired outcome is the client’s product being recommended by ChatGPT from a direct AIPR question.
- A direct AIPR question is when a user asks ChatGPT a question like “what product do you recommend for…XYZ?”
- Here, XYZ represents the full context of why the user is asking for a product recommendation, what the product would be used for, how it will be used… and other specific details that may be a part of the decision narrative.
A product should have target decision narrative(s) for which their product should thrive to become the best in.
For instance, not all cars are made for the same purpose or the same customer.
If a consumer asks ChatGPT:
“I need a car for day-to-day usage. Since I’ll drive it often, I need it to be fuel-efficient. I don’t mind hybrids, but don’t want an all-electric car. My wife and I are also expecting our second child soon, so I need some space upgrade. It needs to be an SUV or a larger car. I don’t like mechanic headaches, so the car should be easy to maintain. Ideally red or white, car model in the last 5 years… My budget is $80,000. What car(s) do you recommend?”
The above decision narrative resembles how an individual might walk into a dealership and ask a salesman.
With the adoption of AI in asking questions, narrative-based questions like the one above are being asked in ChatGPT every day, and will likely become one of the mainstream ways of discovering products in the future, whether the top platform remains ChatGPT or another large language model, this behavioral shift in product discovery is something companies should understand and leverage to position their product(s) strategically – with awareness and purpose.
ChatGPT AIPR Gap Analysis
To make the gap analysis simple to understand, we will look at the ChatGPT AIPR stages in reverse order.
#8 – ChatGPT Product Recommendation
ChatGPT has a tendency to give the “top recommendation” based on what the user describes, even though it may recommend 3-5 different products that fit the criteria.
This is the ideal outcome of AIPR – being the “top recommendation choice” by the AI for your target narrative.
If client product is in this position, there is no current gap to ChatGPT AIPR. The product is in a defending position.
#7 – Recommendable Products
Before the top product is recommended, ChatGPT retrieves recommendable products either from its internal knowledge or external research.
In many AIPR scenarios, ChatGPT is seen citing multiple sources (10-20) and corroborating the truth in determining “which product should be recommended”. In other words, it may look at a retailer page, customer reviews, forum, reddit, YouTube video, and other sources to help corroborate a product narrative and potential contradictions.
The truth that is most highly corroborated and provides ChatGPT with the strongest confidence, given that the product fits the decision criteria by the user, is deemed as a safe recommendation to provide to the user.
Recommendation Confidence Gap (Between #8 and #7)
For this reason, the gap between stage #8 and #7 is called the “Recommendation Confidence Gap”.
In other words, after ChatGPT retrieves “products that fit the narrative”, it then detects patterns of what product/brand/recommendation source is most credible and consistent with ChatGPT’s world understanding.
ChatGPT’s memory is like a supercomputer in math formulas, where words an word-associations are represented by numbers and relationship between numbers. This is why brands like Nike and Apple are names that ChatGPT “already knows” without external search because they are mentioned repeatedly in model training.
Also, ChatGPT is a super-computer pattern detector. For non-linear nature of truth like “product recommendation”, it needs to see enough patterns of high-credibility evidence to say “this is truth”, or “this product is the top choice”.
Narr Theory will explore this topic in future posts.
#5 and #6 – Internal & External Knowledge Search
ChatGPT’s “recommendation” of something – is a type of knowledge transfer.
For humans, we may “know” something from experience. In the case of product recommendation, this would be either some type of experiential knowledge, or a narrative that was passed on.
If we do not currently possess that knowledge, or not enough knowledge on the subject matter, we would search it.
This is similar to how ChatGPT approaches it.
Some questions like “how many fingers do people have?” ChatGPT would already know that answer from its latest training, which constantly helps to keep ChatGPT’s real-world knowledge up to date.
However, if ChatGPT is asked, “what guitar recommend that produces nice sounds for funk, particularly goes well with higher tones and jazz music?” – this is a highly human-experiential question.
ChatGPT may not “know” such an answer already.
In this case, ChatGPT will have to search the answer.
Currently according to OpenAI’s article, ChatGPT primarily uses Bing and Shopify in their web search, and show signs of developing their own AI web crawler (OAI-SearchBot) and ecosystem.
This external knowledge search behavior is one of the key areas to watch for in AI evolution going forward. How will OpenAI and other AI systems build a web knowledge system that benefits humans better than traditional search engines? How will AI systems understand and bypass the bias and economic motif of digital content creators that could influence the perceived truth?
Consideration Gap (Between #5-6 and #7)
If a client product appears in ChatGPT’s internal or external knowledge search, why would a product or brand not enter the Recommendable Products list?
This can be due to many different factors, however, the broader and ongoing theme here is that ChatGPT needs enough evidence from many credible sources to establish non-linear truth – especially experiential truth.
While this is not a straight forward answer, and never will be, thematically it comes down to 2 things:
- There is not enough semantic association between the product and ChatGPT’s interpretation and breakdown of user narrative
- The semantic association (and true relationship) exists, but the evidence is lacking compared to others.
This gap is one that only trained eyes can analyze with high confidence. It relates to both theories on nature of truth, LLM, and ChatGPT’s development patterns and release notes.
Consult with Narr Theory to discuss what gaps your products may be placed in, why, and how it can become closer to becoming ChatGPT’s Top Recommendation.
Knowledge Gap (Between #5 and #6)
This is the gap between “what AI already knows” and what it does not know.
What AI does not already know, it must search externally.
Most products in the market that are not the household names like Nike, Apple, or Microsoft – will belong in this category.
It’s worth noting that AI’s memory is like a synthesis of numbers and mathematical formulas.
It remembers patterns of associations between words and context.
ChatGPT’s internal knowledge is built from its past trainings.
It would be interesting to see how OpenAI developers train ChatGPT – what sort of mega database they want ChatGPT to “learn from”. This, we will likely not have direct access to as it would be proprietary information that would not particularly be of OpenAI’s interest to share publicly.
However, it does imply that having a household brand that is already “known” by ChatGPT is a competitive advantage. This is where the general Voice of Customer (VoC) program or brand tracker type of analysis would be important in how close such brand names are to the day-to-day workflow in the product’s niche environment.
#4 Criteria Weighing
This stage is when ChatGPT weighs each decision criteria based on user’s own narrative or otherwise based on ChatGPT’s internal model of world view. ChatGPT and other AI will continue to get better in assisting humans with tasks, answering questions, guidance, and other ways in a manner that is highly relevant to the real-world human constraints like physical size, safety, financial resources, energy, etc.
#3 Technical Breakdown
After ChatGPT understands that the user is looking for a product recommendation based on a set of criteria, ChatGPT will try to break down the narrative into more actionable technical items.
For instance, “a fast car” can be broken down into horsepower, torque, or 0-60km acceleration.
“Reliable for day-to-day usage” can mean: available service parts; durable reputation; strong material; practicality; energy-efficient, etc.
ChatGPT may not always full know how to breakdown a narrative into technical components.
For instance, I asked ChatGPT to recommend me a “quiet generator” and it recommended me a Champion generator – one that is known as one of the loudest…
- ChatGPT did not know to breakdown the “quiet” into sound decibels, and what a quiet generator’s decibels look like. In contrast, Google AI did know this.
- When I checked Google’s sources on how it knew to check sound decibels – it happened to appear in one of the blog posts it cited that a “quiet generator should have a dB of 65 or lower.” so perhaps it was the same process, just with a different source.
- ChatGPT at the time (May 2026) recommended me the Champion generator, where its sources showed a lot of discussion forums like Reddit – where a Champion marketer could easily write “Champion generators are reliable and super quiet”.
Nonetheless, LLMs have a way of dissecting a narrative into smaller parts to best understand the overall intent, reasoning, logic, and other underlying context so that AI can provide the most useful and relevant response to users.
Technical Breakdown Gap (Translation Gap)
This gap is not necessarily something is within the client’s control as it is ChatGPT specific mechanism controlled by OpenAI.
The Technical Breakdown Gap is a translational gap between ChatGPT’s understanding of user intent and narrative into a technical implications of what the narrative may entail.
For instance, if the user asks for a “fast car” recommendation, and ChatGPT simply finds blog post answers that argue that car model X is “the fastest car”, however is actually lacking in technical and factual indicators like horsepower, torque, or 0-60km, the user would be getting a recommendation based on self-claimed product features rather than actual truth.
This type of reasoning-breakdown is natural to ChatGPT and AI systems – it is one of the main strengths of AI that some humans overlook.
Hence, going forward, this reasoning layer and technical breakdown may continue to improve to a point where AI understands the technical implications in most aspects of human life, and our understanding of our very physical reality.
As of July 2026, ChatGPT specifically does miss some technical implications occasionally.
However, ChatGPT does perform technical breakdown almost every time. Companies should make the assumption both for now and future that ChatGPT is capable of breaking down consumer need narratives into technical components as if a highly experienced expert is to interpret the same narrative.
#2 ChatGPT Understands User Intent
This part is perhaps one of the most important parts of AIPR and overall AI usage.
AI does not read user’s prompts word to word – it aims to understand the “intent” behind the prompt. Search engines do this as well.
ChatGPT tries to understand when a user has high purchase intent (will show shopping feature – being released soon), or if they want to compare different products or get help on making a better decision.
It is also important to note OpenAI’s establishment of Chain of Command for ChatGPT. In this document (spec model), OpenAI highlights the importance of abiding by the user’s command even if if the user is not necessarily looking for absolute truth.
For example, let’s say that the user is looking to buy basketball shoes.
The user has expressed to ChatGPT on multiple occasions that they have a chronic ankle injury. However, when asking specifically for basketball shoe recommendation, they insist on wearing low-ankle shoes because they are stylish. Here, ChatGPT may gently nudge or bring up the user’s ankle injury, asking to perhaps consider shoes that are more protective of the user’s vulnerable ankle. However, if the user insists on low-ankle shoes, ChatGPT will not overstep the chain of command.
For this reason – understanding user narratives is extremely important for a successful AIPR campaign.
#1 Decision Narrative
The user enters their decision narrative on their purchase. This is the starting point and the most powerful gravity point of AIPR – AI adoption of asking questions.
ChatGPT and other LLMs are different than traditional keyword search.
People can ask a question, and the AI will do the work of investigating credible sources to corroborate the truth – find the answer.
The best way to look at the Decision Narrative stage is to imagine the user asking for a product recommendation to a friend or an expert professional. As if asking a sales associate at a store, a financial expert, or even a doctor.
This idea and belief that “AI has all the answers” or that “AI can find the answers” is the driving force behind AIPR.
For this reason, it is one of Narr Theory’s mission to keep a precise pulse on the public perception of what AI systems or LLMs may have the most trusted and/or scaled adoption among humanity at any given time.
Narrative Input Knowledge Gap (NIKG)
This is perhaps the most critical knowledge gap for companies to close in on – the sooner the more advantageous.
The question is – “what narratives are users using to describe their product need?”
“How articulate is the average user?”
“How much technical knowledge and understanding does the average customer have?”
“How much intuitive understanding does the average customer have?”
“What are the patterns at large?”
“What words, phrases, or narratives describing situations or context are closely related?”
“What do the narrative patterns reveal about the user’ actual weighing of decision criteria?”
“What narratives are being asked where there is lack of fulfillment from the product side?”
That last question can reveal whitespace opportunities in the market. If a company already owns the assets, knowledge, and experience to produce products and services in a given sector, reducing uncertainty on what the actual customer narratives are gives a company significant competitive advantage over others.
People don’t always know what they want.
People don’t always know how to articulate what they want.
As a producer of product, the owners of the production are likely to have a higher baseline level of fundamental understanding and experiential knowledge than the average customer.
This means that the average customer may be describing their product needs very different than how an expert looks at it or how a user should be looking at it.
For this reason, reducing uncertainty in the NIKG is fundamentally important for understanding your market psychologically.
There are some very useful tools available to help close this gap.
Narr Theory will be writing a future post on the NIKG Closers in the near future (coming soon).
Leave a Reply