E-E-A-T Was the Beginning. COSMO Shows Where Search Is Going Next.
We learned how to prove authority. Now machines want to understand intent.
TL;DR
Google's E-E-A-T framework taught brands to create content that demonstrates experience, expertise, authoritativeness, and trust. You still need that. But AI-powered discovery is now trying to understand something beyond attributes and keywords: intent and context.
Amazon's COSMO research is the clearest window into that shift. Amazon built it because traditional ecommerce knowledge graphs understood products and attributes but, in their words, "fail to discover user intentions" (Amazon Science). COSMO infers who a product is for, what it's used for, what event makes it relevant, and what it's capable of.
Keep building authority. Start building understanding.
We Learned to Prove Authority. Now Machines Want Intent.
For years, brands trying to improve organic visibility heard some version of the same advice: demonstrate expertise, show experience, build authority, be trustworthy, create genuinely useful content.
Google gave us a framework for that thinking, E-A-T, which became E-E-A-T when Experience was added. It still matters. Google continues to encourage original, helpful, people-first content, while making an important clarification: "While E-E-A-T itself isn't a specific ranking factor, using a mix of factors that can identify content with good E-E-A-T is useful" (Google Search Central).
But something else is happening at the same time. AI is getting much better at understanding intent. Not just what is this product? but who is it for, why would someone need it, when would they use it, what problem does it solve, and what other needs connect to it.
Amazon's COSMO project gives us a fascinating window into that shift. The next evolution of content isn't abandoning what you learned from SEO. It's adding a layer: answering the questions machines need answered to understand your product in context.
First, What E-E-A-T Actually Says
Google is explicit that trust leads: "Of these aspects, trust is most important. The others contribute to trust, but content doesn't necessarily have to demonstrate all of them" (Google Search Central).
Don't just create content. Create content worth trusting, with named authors, firsthand experience, original photography, published testing, and reliable sources.
None of that is going away.
But Ecommerce Search Has Another Problem
Imagine you're Amazon. You have hundreds of millions of products and enormous amounts of structured data: name, brand, category, size, material, color, features, price, specifications.
That's useful. But an enormous piece of information is missing.
Why does someone actually want the product?
This is exactly the gap Amazon's researchers named. Existing ecommerce knowledge graphs "integrate a large volume of concepts or product attributes," but "they fail to discover user intentions, leaving the gap with how people think, behave, and interact with surrounding world" (Amazon Science).
The system may know what the product is without understanding why a human would want it. That's a big difference.
COSMO: The Commonsense Behind a Purchase
COSMO is a large-scale ecommerce commonsense knowledge generation and serving system built at Amazon. It's "a framework that uses large language models (LLMs) to discern the commonsense relationships implicit in customer interaction data from the Amazon Store" (Amazon Science).
The goal is a knowledge graph that "encodes relationships between products in the Amazon Store and the human contexts in which they play a role: their functions, their audiences, the locations in which they're used, and the like" (Amazon Science).
Figure 1
What a commonsense graph knows that a spec sheet doesn't
usedFor, capableOf, isA, and cause, then refines into forms including used_for_function, used_for_event, and used_for_audience (Amazon Science).The output is entity-relation-entity triples. Amazon's own example: <co-purchase of camera case and screen protector, capableOf, protecting camera>.
The example that makes it click
Someone searches for "shoes for pregnant women." The shopper never types "I need slip-resistant shoes because pregnancy affects balance and I want to reduce my risk of falling." But Amazon states the engine "should be able to deduce that pregnant women might want slip-resistant shoes," linking the two through the used_for_audience relationship (Amazon Science).
That's a fundamentally different way of understanding a query. It moves from matching attributes toward reasoning about human intent.
And this isn't a lab experiment.
All three figures come from Amazon's published description of the system (Amazon Science).
The Questions Are Changing
For years we got very good at answering: What is it? What's it made from? What features does it have? What size is it? How does it compare?
Those questions aren't going away, machines and customers both need factual product information. But AI creates an opportunity to answer another class of questions entirely.
| Ask this | Because | A real answer looks like |
|---|---|---|
| Who is this for? | Situations and needs, not just demographics | Owners of large, high-shedding dogs who travel weekly |
| What is it used for? | Category doesn't explain real function | Contain shedding, protect leather from nails, ease post-hike cleanup |
| When would they use it? | Events create the need | Road trips, after swimming, winter mud, camping season |
| Where is it used? | Location changes relevance | Rear seats of cars, trucks, and SUVs |
| What problem does it solve? | Features aren't problems | "Protects leather when your dog jumps in wet after swimming" |
| What is it capable of? | Outcomes, not specs | Travel with the dog without damaging the interior |
| What goes with it? | Co-purchase reveals a bigger job | Cargo liner plus travel bowl means a road trip is being planned |
| When is it the wrong choice? | Tradeoffs help humans and machines | Not ideal if you need a seat free for a child seat |
That last one matters more than brands expect. Saying "our product is the best" gives a recommendation engine nothing. Explaining when it's a great fit, when another solution is better, and what tradeoffs exist gives it something to reason with.
From Keywords to Context
Here's the simplest way to see the shift.
Figure 2
The same product, described two ways
Keywords
Dog seat cover · waterproof dog seat cover · back seat dog cover · car seat protector for dogs
What a machine learns
Context
Who it's for, what problem it solves, when it's useful, where it's used, why one design over another, and what outcome the customer wants.
What a machine learns
Old product thinking. Dog seat cover. Keywords: dog seat cover, waterproof dog seat cover, back seat dog cover, car seat protector for dogs. Those terms still matter, but now add the context.
AI-era product thinking. Who is it for? Dog owners who travel frequently with their pets. What problem does it solve? Mud, water, hair, scratches, and general wear on upholstery. When is it especially useful? Road trips, hiking, camping, swimming, dog parks, everyday transportation. Where is it used? Rear seats of cars, trucks, and SUVs. Why choose one design over another? Seat configuration, dog size, passenger needs, seat belt access, whether seats fold independently. What outcome does the customer want? Travel with the dog without worrying about destroying the interior.
E-E-A-T Plus Intent Is the Real Answer
This isn't "E-E-A-T is dead, COSMO replaces it." They solve different problems.
1E-E-A-T answers
- Why should this information be trusted?
- Who created it and what do they know
- Whether the source is credible beyond its own site
2Commonsense context answers
- Why does this make sense for this person?
- What need, situation, or event it fits
- When it's the right choice and when it isn't
Put them together. Your company has demonstrated experience. Your authors have expertise. Your brand has authority. Your information is trustworthy. And you've clearly explained who your products are for, what problems they solve, when they're useful, and why someone might choose them. That's a far richer information environment than either half alone.
Your website has to answer more than product questions
This reinforces something we believe strongly at VEWO: your website should become the knowledge base for your brand. Your product detail page can't answer every possible question, and it shouldn't try. That's where the broader content ecosystem matters. Imagine supporting one product with articles like:
- What's the best seat cover for a large dog?
- How do you protect leather seats from dog nails?
- Hammock vs. non-hammock, which should you choose?
- Can you use a dog seat cover with child car seats?
- What if you still carry human passengers?
- How do you keep dog hair out of upholstery?
Now you're doing more than targeting keywords. You're documenting the real-world situations surrounding the product.
Where AI Agents Make This Urgent
Today a customer searches "waterproof dog seat cover." Increasingly, the interaction looks like this:
"I have a 90-pound Labrador, a new SUV with leather seats, and two kids. We hike every weekend, so the dog is constantly muddy. I still need one side of the back seat available for my daughter. What's the best way to protect the seats?"
That's not a keyword search. It's a problem. An agent has to reason through seven separate constraints at once:
- Large dog
- Leather seats
- Constant mud
- Kids in the vehicle
- Split-seat requirement
- Frequent outdoor use
- Vehicle compatibility
If your website has clearly answered each of those, the agent has what it needs to conclude that your product fits this specific situation. If it hasn't, the agent can't infer it, and it will recommend a brand whose information is clearer.
Don't Throw Away Your SEO Playbook
There's an important warning here. Don't read this and decide SEO doesn't matter. It does.
Keep building technically sound websites, creating helpful content, demonstrating expertise, earning reputable mentions and links, structuring information clearly, improving product data, and answering search intent.
But expand the thinking. The question is no longer only "what keywords should this page rank for?"
What would a customer, or an AI agent helping them, need to understand to know when this product is the right answer?
A practical framework
You don't need to rebuild your website tomorrow. Start with your most important products or services and answer ten questions for each.
| Dimension | The question to answer |
|---|---|
| Who | Who is this really for? |
| What | What does it actually help them accomplish? |
| Why | Why would someone need it? |
| When | What situations or events create the need? |
| Where | Where is it typically used? |
| Problem | What problem is the customer solving? |
| Outcome | What does success look like? |
| Compatibility | What does it work with, or not work with? |
| Comparison | When should someone choose this versus an alternative? |
| Connection | What other products, needs, or activities connect to it? |
Then look at your website and ask whether someone can actually find those answers. If not, you've just found your content roadmap.
Frequently Asked Questions
Is E-E-A-T a Google ranking factor?
No. Google states plainly that "while E-E-A-T itself isn't a specific ranking factor, using a mix of factors that can identify content with good E-E-A-T is useful." It's a way of thinking about the qualities Google's systems try to identify in helpful content, and trust is the most important of the four.
What is Amazon COSMO?
A large-scale commonsense knowledge generation and serving system that uses large language models, customer behavior data, and human review to infer relationships product listings never state, such as who a product is for, what event it suits, and what it's capable of. It has been deployed in Amazon search applications including search navigation.
Does COSMO affect Google rankings?
No. COSMO is Amazon's system. It matters as a signal of direction: it shows how sophisticated AI-driven discovery has become at reasoning about intent rather than matching attributes, which is a shift happening across search and AI assistants generally.
Where should we add this context, product pages or blog content?
Both, with a division of labor. Product pages carry the structured facts, compatibility, and fit. Supporting articles carry the situations, comparisons, and use cases a product page can't hold without becoming unusable.
Isn't this just writing for machines?
No, the same information helps buyers decide. Explaining when a product is the wrong choice, what it works with, and what problem it actually solves is what a good salesperson does. Machines simply reward it now too.
Keep the Authority. Add the Context.
Google helped push the web toward better information: experience, expertise, authoritativeness, trust. Those things still matter, arguably more than ever.
But Amazon's COSMO research gives brands another clue about where AI-powered discovery is heading. Machines aren't only getting better at identifying what something is. They're getting better at reasoning about why someone might need it.
So keep proving you know what you're talking about. Keep creating original, trustworthy information. Keep demonstrating experience. And start answering another layer of questions: who is this for, what is it really used for, when would someone need it, what problem does it solve, what is it capable of, what else naturally goes with it, and why is it the right choice in one situation but not another.
Because the future of discovery isn't matching a product to a keyword. It's understanding whether that product makes sense in context. The brands that provide that context give both customers and machines much more to work with.
Don't stop building authority. Start building understanding.
Want AI to recommend your brand?
See exactly where you stand with a free visibility audit.
Get Your Free Audit