Thoughts on Kayak's Ai so far?

Oh wow. Not a terrible idea to prove out the demand and conversion from it before overhauling and destroying your main site. Maybe it flops (deprecate it), or is incremental (run it in parallel) or it totally blows things out of the water (replace the main).

their robots.txt and sitemap are also worth contemplating over

Haha I wouldn't be surprised someone in 10 Canal St is looking at this but... we're talking about you!

TLDR; It appears they want "share" pages that are created (UGC) to get indexed and served to bots... !!!

I was on mobile when I first looked - I see what you mean!

Sadly no LLMs.txt though :whitefrowningface:

If they're successful in getting people to use their version... they'd get a lot of "ChatGPT-like behavior" in their app analytics

IMO that's the 3D chess move there.

Share-gating is probably a combination of oauth + you could also have a more elaborate system prompt to deter spam

Wheels are spinning with this

 

Haven't figured out the criteria for sitemap inclusion yet, but :joy:

Looks like if the chats are shared, they’re aggregated into topics and are then available as a share route. My main very basic assumption for criteria would be:

• is it shared
• Are there N number of prompt answers available for any topic
• Is there demand for said topic (they’d have that data from kayak.com)

We're so back to doorway pages :upsidedownface:

It’s wild out there

The fact that it shares by language parameters bewilders me though :joy:

Back to the pinterest playbook...

Honestly idk if/how/why users would use this over GPT etc but I do love the idea as a test. And yea the ability to get all that prompt data from the tool :drooling_face:

https://kayak.ai/share/sfo58yDTFR_K7bhMGC-HEXA

This is basically the Reddit Answers playbook.

I'm all for this. Planning trips and itineraries can be complex and nuanced. I used Comet to book a full trip with an itinerary and car rental and saved $257. But let's see how well it works and the adoption rate...it's good prototyping

upside is big to test
in larger orgs, at times the bureaucracy demands such moves too

With a few days to think about this:

1. Kyle's right about using kayak.com data to validate - or even kayak.ai data to validate - whether a "share" is worth indexing. Some sort of "review of quality" is needed before spammers are creating "backlinks"
a. Right now their use of parameters and indexing different countries concerns me on their XML sitemap.
b. Their XML sitemap hasn't really budged since their launch, which is a good thing.
i. That being said, I haven't been messing with browser user-agents to test it (they could very well be doing this to throw off some competitors, but IYKYK)
c. We'll know this is a serious play when kayak.ai is promoted.
d. Depending on how tight they'd want to define their clusters via embeddings, you could see a lot of shares actually all redirect to a "canonical" share that's both most popular but similar in intent.
2. Doing tests like this (Throw your website as an AI chat wrapper) on a separate domain will become a new norm if LLM's to unlock data for providers of content and experiences.
3. If the app is helpful and people are sharing really helpful trip plans... then the content is actually not crap or considered doorway pages. How judicious you are with various signals to shared trip plans on indexing is what makes or breaks this UGC programmatic SEO play.
a. Spammers will need prompt engineering skills to get the right kind of response from an LLM which benefits their blackhat plays - which may also serve as a deterrent - but every hack that gets through can be collected as an edge case to train against.
b. Travel planning isn't a DAU kind of activity, let alone WAU. To make habits stick you need usage. Without usage, everything falls apart.