Session 4
Personal
Musing:
I’m also glad I didn’t go to Config. LinkedIn is making it seem like a repetative groupthink and it makes me feel ahead that we’re on a completely different tack exploring the off the common waypoint. Too much same same.
Did a Korean style photoshoot with the family. Some selects!
Canopy
Cleaned up and reorganized our pitch. This architecture feels cleaner -- basically pitched this “story” to my two friends at the bar and they immediately got it. Highlights I’ve shared already:
What to build, when
These are my current thoughts about what to build, in what order. Loosely held at this point and looking for alignment / feedback! I’m 100% sure this will change but just a strawman for now.
Framework here is basically 1. OUTPUTS 2. INPUTS 3. USABILITY and FLOW 4. COMMUNITY 5. BRANDS 6. IDENTITY
Step 1. Product Reviews and Guides
. We’ll have to first be able to use AI to create a useable, aggregate review.
[I have a strong suspicion that this is actually all we’ll need to raise a pre-seed because it demonstrates the output]
-
A : Identify a starting category EX: consumer electronics, software, consumer hard goods, health and wellness product. OR maybe it works regardless of category as long as they can use a similar template.
- B: Build a great product review page that works with the intended template EX: Leica Q3 camera.
-
C : Build a great, aggregated review from many sources with the right format. EX: Best Travel Camera
-
I see B and C here as very related, just using two tamples. B needs to scrape to see what the internet thinks about one camera, and C needs to do the same thing but for a category of cameras.
-
We should be able to generate these similarly to “reports.”
-
D : Enhanced review details (extracting parameters from search and displaying them visually)
Step 2. Build a working chat that can identify the right query from a user to the point where it’s useful enough for Canopy to generate the correct review or guide.
(Building the beginning of the real “input” machine)
-
We’ll need to handle both deterministics an indeterministic questions, like “Hey, what can I search for?” and “I’m looking for a camera, but not sure what the options are” as well as “show me a guide for the Lecia Q3” and “what are the best travel cameras under $4,000?”
- Interaction Model for the app (IA)
Step 3. Build memory and saved chats, reviews, guides
- Ability for users to create collections
- Recall guides, objects they’ve saved
- Begin taste graph creation for users based on what they save
- Still in the utility phase
Step 4. Build Home feed (recommendations based on past search, taste graph)
- Moving into the retention and discovery phase
- Get people coming back to peruse and really show the taste graph working
- Ability to save individual items you find
- Collecting first party data
Step 5. Build Community around objects and reviews
(I’m not 100% sure if we want to consider this sooner or later, but strikes me that we need a solid foundation first). We could have a community that we spin up SUPER early alongside step 1 to critique our output work that ends up growing into this) - A. : Tribes you can join
- B : Critique and provide feedback for existing guides
- C : “KOM” lists for categories (latte, cocktails, neighborhoods, etc.)
- D : Answer AI questions about products and guides (if Canopy needs to fill in information about a product, we can ask the relevant people). Builds on our first party data set established by “saved” and other interaction level details like what people click on and converse about.
Step 5 Parallel. Brand Integreation
(Another where I don’t know yet WHEN this will be important for us). I think we will know when we need to do this as we start to test / build / get feedback and it will be very apparent.- A : Sign up brands / companies and integrate their data
- B : Offer customization tools for Brands to modify their content and create an AICO™ engine (AI Chat Optimization Engine)
- This COULD be much more important earlier as a key value prop
Step 6. Identity and UGC
A : Identity / Profile
B : Allow creation of first-party guides on Canopy (someone creates their own, like an influencer, a brand, a celeb, or a reg user).
C : Expand categories to things like places, destinations, regions, neighborhoods, experiences, food and beverage (another thing that I don’t know WHEN is the right time for -- we may learn early that actually people really want guides for locations, so we end up doing that becuase we get feedback)
Things I read / links
“a major implication is the need for curation and "taste development" tools. As artists spend less time doing mundane repetitive stuff (thanks to Al), they'll have more time to explore ideas and create better stories. This is something my teams and I are thinking a lot about these days.”
A few key things that I think when reading this:
-
They don’t know what we know -- rather than playing the same old SEO model where they suggest “optimizing for AI Search starts with optimizing for search using traditional SEO strategies + is an evolution of new optimization strategies” isn’t something I agree with. They are taking the long path by suggesting companies adjust their SEO strategies when really they need to skip that and use Canopy’s better RAG tool (the brand facing ACO tool I mentioned above
- They say “There are no tools to find what questions people ask - new strategies will need to be made to source how people ask questions” -- this is something WE WILL DO
- They say “There are no tools that allow companies to track their presence in AI Answers” -- WE WILL DO
From their post:
Key Points
* AI Search is increasingly used for search and discovery
* Graphite is building strategies to help track and optimize for AI Search
* Optimization for AI Search will likely be similar and an evolution of traditional SEO vs. something entirely different
LLM vs. LLM+RAG
* Early versions of LLMs were based on next-word prediction, which is different from search
* More and more, AI chat and search use RAG+LLM
* RAG (retrieval augmented generation) starts with a search (retrieval), then summarizes/reforms the search as an answer (generation)
* Perplexity has built its own search engine, CoPilot frequently uses Bing, and OpenAI is reportedly building a search engine - so AI Search is moving towards RAG+LLM more and more
* Therefore, optimizing for AI Search starts with optimizing for search using traditional SEO strategies + is an evolution of new optimization strategies
Keyword Research > Question Research
* "Question research" is the new keyword research
* Rather than finding keywords, instead we need to find all the ways people ask questions for our product
* We may target thousands of keywords, but questions vary more than keywords, thus there are millions of variations of questions
* Keyword research can use known data from Google and other keyword tools
* There are no tools to find what questions people ask - new strategies will need to be made to source how people ask questions
* Based on the questions people ask, we create landing pages and content to target questions similar to SEO
Good Content for Questions
* "Good content" in Google SEO is comprehensive and answers the questions users have for that page via the presence of TF-IDF terms, sub-topics, and embeddings
* A landing page that targets a topic in AI Search needs to first understand the thousands of questions users have for that topic, then answer as many of them as possible
* Example - ResortPass's West Hollywood Edition page should answer questions about reserving a cabana, how much does a day pass cost, and what are opening hours
Citation Optimization
* RAG+LLM performs a search, looks at multiple pages, summarizes them, and cites its sources
* Companies can optimize for being cited using SEO strategies similar
* Forbes and NerdWallet can try to optimize for being a citation for "best credit cards" across many variations of credit cards
SERP Tracking > AI Answer Tracking
* SEO tracks single positions for keywords (e.g. I rank #5 for "best credit card")
* AI Answer Tracking is a distribution or frequency across surfaces, question variants, and question runs
* There are no tools that allow companies to track their presence in AI Answers
* Graphite is launching AI Answer Tracking - https://lnkd.in/d9w8XQjh