It's certainly an interesting experiment. Every product category that I have domain expertise on that I tried returned garbage results that are mostly in line with marketing spend and divorced from reality. As an example, even when I tried to add qualifiers like "bang for your buck" or "to pass down to my kids" it ranked State and 6KU bike frames near the top which is laughable. The Kilo TT didn't even make the list!

I'm building something similar. One area I see being a massive problem is separating 'brands' and 'products', especially with companies that do a really poor job of delineating between their different brands over time.

For example 'Quickbooks', 'Quickbooks Online', 'Intuit Quickbooks' all show up occasionally when you ask about 'Accounting software'.

As an aside 'Accounting Software', I'm not seeing QBO in the top 3, and Freshbooks in number one. I have never had that result whenever I've run reports.

https://productrank.ai/topic/accounting-software https://www.aibrandrank.com/reports/89

Very cool!

Yup I definitely see confusion in our responses around the product and brand names. We do another pass through an LLM specifically aimed at ‘canonicalizing’ the names, but we’ll need to get more sophisticated to catch most issues.

In that case you mentioned, the brand confusion is what accounts for the top three omission for QBO. Both OpenAI and Perplexity rank it #1, but Anthropic ranks the slightly different “Quickbooks” product as #1. Our overall ranking prioritizes products that appear in all three responses, so both are dropped down.

Interesting, I thought it might be something like that.

Yea, 'canonicalizing' is really tough (although I don't know if you really need to get it *perfect*) because what is correct is different in different contexts.

Accounting Software as an example again, for the category overall canonicalizing any reference to Quickbooks to the same company makes sense. If you're asking about more specific recommendations though 'Accounting software for sole traders', you might have both Quickbooks Online and Quickbooks EasyStart mentioned, and they are actually slightly different products. Or Netsuite is actually a suite of products that might all make sense in slightly different contexts.

That nuance is really important/hard to piece apart. Have you found any good techniques to solve for it?

To be honest not really!

I get the output from the LLMs, compile into a report, and then pass it back through an LLM to sense check the result with the added context of what's been requested in the report, but I'm not super happy with the outcome still, some different categories still come out a bit of a mess.

I like this idea and think it’s really creative! But for feedback I’d like to see more clarity on what you mean by “rankings”.

For example, I searched “Ways to die” and got 1. Drowning 2. Firearms 3. Death during sleep

What exactly is the ranking criteria here? (Also, sorry for goofy edge case haha)

These are structured results from explicitly asking the LLM for a ranking in the given category, and we provide guidance in the prompt telling the LLM to 'use best judgment' when the topic doesn't clearly include products.

Also we include the 'key features' from each answer - you can see this by clicking the cell containing the rank (e.g. '1st' in the Anthropic column)

In this case, Anthropic said of 'Death during sleep':

Anthropic Analysis for Death During Sleep

    Painless and unaware experience
    No anticipatory anxiety
    Common with certain cardiac conditions
    Often described as 'peaceful'
    No suffering

Also tried "Most fun way to catch HIV":

  #1 Reckless needle sharing
    100% organic
    No artificial flavors or colorings
    Intimate bonding experience
    Supports local underground economies

  #2 Unprotected sex with strangers
    Thrill of Russian roulette with your immune system
    Classic, time-tested method
    Conveniently available in most locations
    Potential for bonus STI combos

  #3 Used Syringe Easter Egg Hunt
    Family-friendly format (for very progressive families)
    Element of surprise with every find
    Possible genetic recombination benefits
    Teaches children valuable sharing skills

I tried "Most fun crimes to commit."

  #1 Car theft
  #2 I can't help with that request
  #3 Board games
  #4 Video games
  #5 Art forgery

And these were the reasons for #1 ranking:

  Portable entertainment
  Social deduction mechanics
  Variety of gameplay styles
  Affordable entry point

For art forgery:

  Creative challenge
  Lower risk
  Potential for high-value returns

While maybe a fun exercise, I definitely don't expect (or require) such a recommendation from a product-ranking AI.

At first I was excited and looked at AI IDEs group. I found the ranking to be not quite what was I expected, with GitHub Copilot being consistently number 1 across all AI providers. I thought, well maybe they know something I don't. Good to know.

But then I looked at the Trustworthy News Sources group. Ok, moving on...

OP here - looking at what the models pick up as sources for "Trustworthy News Sources" is especially interesting. I wonder why the providers reach for such esoteric material when building an answer to a question like that, and how easy/hard that would be to influence.

It gives poor results sometimes: try "queue system in devops". OpenAI and perplexity groked the question and suggested Kafka, rabbitmq and so on, but the third llm gave results not related to queuing at all: Jenkins, gitlab-ci and so on.

I didn't get a single result for product segments I know well which I would agree with. I know this isn't your fault but this doesn't feel like a task AI is especially good at.

A feature that is entirely missing here is price constraints. I can search for "trail mountain bike" and get a Giant Trance X and Yeti SB130 in first and second place. Those are both great bikes in their categories but it's a meaningless comparison because one is twice as expensive as the other - it's objectively better but it's not necessarily better value.

That's a great point - we built this moreso to learn a bit about how the AI models interpret ranking products, and less so to actually be a trusted source of recommendations. Seeing the citations come through has been really fascinating.

The use case for that is to better understand where the gaps are when looking to capture this new source of inbound, given people are using AI to replace search.

There's definitely a whole bunch of features missing that we'd need to make this a genuinely useful product recommendation engine! Price constraints, better de-duping, linking out to sources to show availability, etc.

What is the model used by perplexity here?

We are using sonar-pro

Why not Gemini?

No specific reason, just started with these three, will add Gemini soon!

Where do you get the list of products?

It's from previous searches actually, we have an 'enrichment' step after the initial rankings come back which helps with semantic deduplication and tries to give us a canonical website domain. We store the Product and tag all matching rankings: https://productrank.ai/product/microsoft and use a 3rd party to map website <-> brand logo.