Feeding Google’s Knowledge Graph with accurate data is crucial for brands, as it helps Google understand the brand’s identity, offerings, and target audience. Historically, brands have dedicated considerable efforts to optimizing for the Knowledge Graph. With the emergence of Generative AI, this focus becomes even more crucial, especially for E-Commerce companies. As people increasingly start to use GenAI technology when making buying decisions, the risk of hallucinations generated by an “ill-informed” AI becomes much greater. To mitigate this risk, the key lies in supplying precise information through the Knowledge Graph. The icing on the cake is that you also get your knowledge panel with all relevant information. In this session, Sara will show you how to mitigate the Gen AI risk by paying more attention to the Knowledge Graph.
2. Let me introduce myself
– SEO Specialist at Liip, a web & mobile
development agency
– Kalicube contributor
– #SEOnerdSwitzerland Meetup co-
founder
3. What you need to
know about the KG?
01
Why is the KG
important for GenAI?
02
Is the KG used in
SGE?
03
Is the KG used in
Gemini?
04
Agenda
05
How can e-commerce
obtain a product KP?
05
How can e-commerce
obtain a company KP?
5. What is the knowledge graph?
A knowledge graph is a type
of database that stores
information and is used to
represent knowledge in a
way that is easy for
machines to understand
and process. The KG
answers factual
information.
6. Google has many MREID for the
same brand. There is likely more
than one knowledge graph or the
knowledge graphs have several
verticals that are inter-connected
11. Why is the Knowledge graph important
from a business perspective?
Pubcon Fabrice Canel
No brand wants users to be given inaccurate answers.
Knowledge panel + GenAI
GenAI
12. Why is the Knowledge
Graph important for
GenAI?
13. GenAI
GenAI is a conversational artificial
intelligence (AI) that can create
content at the click of a button…
but it also has a pitfall –
hallucinations.
The risk is that GenAI
gives the wrong
information about
your product and
brand. Neither you
nor Google wants
that.
14. Knowledge Graph
A knowledge graph is a type of database that
stores information and is used to represent
knowledge in a way that is easy for machines
to understand and process. The KG answers
factual information.
This is the kind of info
that you want Google
to pick up when
talking about your
brand and product.
16. Let me tell you the story
of Anaconda
March 2023
17. Solving the hallucination problem with RAG
Retrieval-Augmented Generation for Large Language Models: A Survey
External
knowledge bases
=
Knowledge graph
Web indexing,
etc.
18. Is the Google Knowledge
Graph used in the search
generative experience
(SGE)?
24. Is the knowledge graph
used by Gemini?
24
Double-check response
Factualinformation
25. Let’s ask Gemini a factual question
The knowledge graph contains factual information. Which means
that a factual question should obtain a response from the
knowledge graph.
28. Unclear if Gemini uses knowledge
graphs yet, research ongoing, but…there
are things we can observe
– Gemini has the info but do not always display it
– Gemini data has been updated recently
– Gemini has layers of protection (LLM Guardrails)
30. Gemini data has been updated recently
Credits goes to Gus Pelogia
Credits goes to Jason Barnard
31. Gemini has layers of protection (LLM
Guardrails)
For GDPR reasons, Google has limited requests for
person information via a prompt system, so if the
LLM detects a person, you can't get information
about that person. However, when you make
mistakes such as misspelling the first and/or last
name, it no longer detects a person entity but
tokens that fall outside the rule and responds as
best it can
Vincent Terrasi
34. We are working on understanding better
the protections…to be continued
Public figure Not Public figure
Not Public figure
Not Public figure
Ulrika Viberg Sara Moccand-Sayegh Gus Pelogia Nicolas Piquero
35. Why you care about the layers of
protection ( Guardrails in LLM)
LLMs outputs could be inaccurate, offensive, or biased.
Guardrails help filter out such content, preventing brand
reputation damage and potential customer outrage.
They may be too cautions and filter out product that are safe.
36. Bonus: understanding Gemini results
If you click “understand the results” in Google Gemini
No code = no schema
(schema are structured data).
38. What info is shown in the product
knowledge panel?
- Product
information
- Product price
range
- Where to buy the
product online
- Product definition
- Product review
39. What e-commerce should do with a
new product
Description Schema Merchants Manufacturer
Product schema
Offer schema
Product variant schema
40. Where does the product Knowledge Panel
take its information from?
– Merchant center
– Manufacturer center
– Structured data
– Website product review (structured and unstructured data)
There a wide range of sources! The web is vast, and it’s difficult for
search engines to inter-connect all the information. Which is why
you need to help.
41. Results of connecting the information
Andrea Volpini of Wordlift shared this graph showing the results
after the information on the web was inter-connected for a product
of a well-known brand
42. If there is one thing to choose to
connect the elements, it’s the GTIN
(Global Trade Item Number)
Andrea Volpini
45. What if you don’t have a GTIN…
Find another Unique Product Identifiers to connect everything…
46. Bonus: some Wordlift resources on the
product knowledge panel
1. The GS1 Digital Link explained for SEO Jedis
https://wordlift.io/blog/en/gs1-digital-link-seo/
1. From GS1 Global Forum to SEO Innovation: Insights for a
Connected E-Commerce Ecosystem
https://wordlift.io/blog/en/from-gs1-global-forum-to-seo-innovation/
1. An SEO Passport for your Products: the Unique Product
Identifier
https://wordlift.io/blog/en/unique-product-identifiers/
1. The Power of Product Knowledge Graph for E-commerce
https://wordlift.io/blog/en/product-knowledge-graph/
49. Kalicube Process
The Kalicube process allows
companies to feed the correct
information to Google/Bing.
This will ultimately result in a
rich SERP and in driving
GenAI answer/related
questions.
https://kalicube.com/guides/
49
Become an entity
50. What understanding means
In this phase, companies want to build an understanding of the brand, including
its identity, its offerings, and its target audience. In practice, this means
becoming an Entity and feeding the machine with the right information about
the company and the ecosystem.
To be in the knowledge graph and ultimately have the knowledge panel Google
needs:
1. To understand who you are
2. Google needs be confident in understanding. So you need consistency in
the information.
51. — Google understands who I am
— Google needs to trust what it understands
— So you need consistency in the information
Let’s check an example using
myself
52. Step-by-step for understanding
1. Have a description of your company
2. Make sure that there is consistency on the web in the
company description
3. Use the Entity Home (concept created by Jason Barnard) to
feed the machine information about your company → This is
normally your/about-us page
4. Use your Entity Home to connect the dots with Schema.org
53. The Entity Home is a term created by Jason Barnard to define the web
page recognized by Google as the authoritative source for information
about a given entity (brand, company, person, product, podcast, music
group, etc.).
What is an entity home?
Your goal is that the entity home is your website
and not some social media page, because by
having your website as an entity home, you can
better control the information Google is getting.
54. Bonus: knowledge panel checklist
Jason Barnard has created a checklist for the knowledge panel
which you can obtain for free:
https://solutions.kalicube.com/knowledge-panel-checklist
To be in the Knowledge Graph, a brand needs to be an entity.
If for example someone goes to a chatbot and ask what’s the best diving watch. Than you want to have your brand there.
How many people used Chrome dev tools?
…and you can use the devtool for that.
So you get close to devs and you can better work with them.
constrained by its pre training data, lacks knowledge of recent events. RAG addresses this gap by retrieving up-to-date document excerpts from external knowledge bases
Communication!!!
If you do not have an understanding of dev needs & you do not communicate properly Then, there is no harmony.