Why we created LLM Optimizer

[CedricHuesler]

Many of you have been building on AEM for years - crafting sites that don’t just look great but drive measurable business outcomes. You’ve set the bar high.

Now, the landscape is shifting. People are finding answers and getting work done through new, powerful tools - ChatGPT, Google AI Mode, and others.

The AEM team is committed to making sure this community stays ahead of that curve. That’s why we’ve built LLM Optimizer - a new capability that gives you deep visibility into how generative AI systems (like chatbots and agents) consume and interpret your site’s content. It connects those insights to business outcomes and provides step-by-step guidance to help you capture emerging opportunities.

This week, we’re starting to roll out LLM Optimizer for AEM customers on AMS and Cloud Service. You’ll start seeing real data - tracking agentic traffic and surfacing your first set of optimization opportunities.

We started this journey just over 100 business days ago, and it’s only the beginning. As we continue to learn, we want this community alongside us - shaping how LLM Optimizer evolves and helping every one of you become experts in building agentic websites.

Questions - let's talk below

[aanchal-sikka]

Thanks @CedricHuesler for initiating the thread. 

What are the potential applications of LLM Optimizer for companies like Mondelez, Unilever, and Nestle, which rely heavily on sales through shelf placement in retail stores and online marketplaces like Amazon, Walmart, or Zomato/Swiggy (In India)?

Also, genAI expects a more structured contextual content. What should be the changes in terms of AEM Sites & Assets to assure we meet the requirements?

[CedricHuesler]

While CPG companies don't typically do sell directly, they are the authority for the product information - therefore ChatGPT and others will look on their domains for the depth of the content. CPG companies need to at details talk about the use cases the product can be used - and how exactly - so they can control the narrative of the product.

Yes, structured content - but not the way you expect as in JSON or so.. - but semantic content in sense of HTML - h1, h2, h3, lists.. avoid tables unless they are comparisons. 

I'm planning to write a longer post here on recommendations we have put together so far to discuss.

[Himanshu_Jain]

Hi @CedricHuesler ,

LLM optimizer is the need now , as everyone focusing on utilizing genAI tools but how LLM will read our data efficiently and make it available in LLM search .

Thanks

[kautuk_sahni]

@CedricHuesler Really exciting to see this update, Cedric! With AI reshaping how users find and engage with content, a few questions came to mind:

• How does LLM Optimizer quantify the impact of agentic traffic on key business outcomes, such as conversions, form submissions, or content engagement? Are there specific dashboards or visualizations that help identify which AI-driven interactions are driving measurable results?

• For sites with complex content structures or dynamic sections, how does the tool help identify which content is most likely to be consumed by AI agents?

[CedricHuesler]

Go try it out on https://play.llmo.now - and look at 3 data sources - and combine them into brand specific opportunties. 

If you have an example of a complex content structure and dynamic section in mind - share it here - it's easier to talk about a concrete case.

[somen-sarkar]

Thanks @CedricHuesler . 

LLM Optimizer is truly the need of the hour, especially as we move into an era of zero/one-click search and the rise of AI-driven discovery. I have a few questions to better understand its capabilities:

1. Can LLM Optimizer quantify the measurable business impact — in terms of engagement, revenue, and conversions of optimizing brand presence for AI-driven search compared to traditional SEO methods?
2. How does the tool identify, measure, and optimize for 'agentic traffic' originating from LLMs, AI browsers, or other generative AI interfaces?
3. What specific best practices should content teams follow so that LLMs assign higher GEO scores during model training?
4. Are there recommended workflows or governance models for integrating LLM optimization into existing AEM content production cycles?

Thanks,

Somen

[A_H_M_Imrul]

Thanks @CedricHuesler

I have some questions regarding LLM Optimizer: 

- How should our content strategy change to better serve AI agents, while still maintaining a great experience for human visitors?

- Will it be possible to give a concrete example of an "optimization opportunity" that LLM Optimizer might surface?

- Are there plans to integrate LLM Optimizer more deeply with other Adobe Experience Cloud products, like Analytics or Target?

Thanks

A. H. M. Imrul Hasan

[manjunathdj]

Few questions on my thoughts.....

• How does LLM Optimizer integrate with AEM Cloud Service and Edge Delivery Service?

• What are the prerequisites for onboarding LLM Optimizer with my current AEM setup?

• Does LLM Optimizer support multilingual content optimization?

• How does it handle structured vs. unstructured content.....

[blissmishra]

Hi @CedricHuesler,

Thank you for the uodate and great to see how Adobe is thinking of the cutting edge and use LLMs.

My questions are different:

1. As a business I will get the insight but will my data be used for training the model?

2. If you will not be using my data how will you plan to get the AL MOdel better?

3. What guardrails will be there to make sure the AI Bias and AI hallucinations are not impacting the results?

4. How will you be keeping human in loop?

Thanks

Prahlad

[BrianKasingli]

Hey @CedricHuesler ,

This is awesome and very exciting. As the team is rolling out the LLM Optimizer this week, I am sure that there would be some Adobe best practices, case studies, and documentation that would be rolled out as well.

Thank you for the update,
Brian

[SantoshSai]

Hi @CedricHuesler,

Amazing to see this rolling out so quickly.

For AEM on-prem or hybrid customers, is there a roadmap for extending LLM Optimizer capabilities beyond Cloud Service and AMS, or guidance on partial adoption through APIs or connectors?

Also, as we start thinking about “agentic website design,” will Adobe provide best-practice frameworks or tooling to ensure AEM pages are structured for LLM readability and citation accuracy?

[gregd59500919]

This is very exciting! Thank you @CedricHuesler and team and thank you to @kautuk_sahni for nudging us in the AEM champion channel.

• For clients, and by extension partners, OOO reporting will be key to show business impact. Are we ready to share further details on implementation/integrations (e.g. how would analytics, ML suggestions, etc. look like)?
  
  
• Similarly, as it pertains to the MCP integration into any existing client enterprise MCP server solutions, what kind of integration and framework approaches are there with the LLM Optimizer?

Thank you, again!

[Shashi_Mulugu]

@CedricHuesler Great work. 

My perspective on LLM Optimizer:-

As LLMs increasingly mediate how end-users discover and interact with website content, AEM authors and brands aren’t just optimizing for algorithms—they’re shaping the very digital reputation and trust signals that will represent their orgs in conversational interfaces like ChatGPT, Gemini , others.
That means we’ll need to think differently about:

• Content accountability: Authors need to ensure that what LLMs “learn” and present about their brands is accurate, unbiased, and in line with company values.
• Governance frameworks: As agentic traffic grows, Every organizations should have a dedicated framework, process and people in place to own auditing, remediation, and continual improvement for content discoverability in AI.

My questions:-

1. How can LLM Optimizer recognize and apply our organization’s brand rules—such as tone, terminology, compliance, and visual standards—when analyzing and recommending optimization for AI consumption?
2. Could future releases enable us to upload or configure brand guidelines directly, ensuring every recommendation supports brand consistency and compliance?
3. What actionable steps—content structuring, metadata, markup, or semantic cues—can LLM Optimizer surface so teams can systematically improve their brand’s GEO score, ensuring we consistently appear at the top of LLM-generated results?
4. Is there a roadmap for making these scoring mechanics and recommendations more transparent and directly tied to business objectives?

Looking forward to your thoughts, Cedric, and hearing how others are tackling these priorities!

[Wilson-Faure]

Would the recommendations be available or consolidated on Adobe Site Optimizer or it will be displayed on a separate dashboard?

[DPrakashRaj]

I do have couple of questions on this.

• is this feature will be available to all costumers on ams and AEMcaS wothout any additional license?
• sometimes its not possible to do much change on content strategy in large enterprise applications due to limitations from content team requirements and to meet accessibility needs. so how adobe LLM optimizer can help in this case?
• Is it possible to use some keywords like meta keywords for seo and aria-label for accessibility to leverage the adobe LLM optimizer without changing the content strategy?
• Is it possible to provide some ranking for certain content to appears top of ai search for your pages?

[VeenaVi1]

Hi @CedricHuesler 

I just ran the plugin to check my org website and it gave me a score of 79%.

Few Questions

1. What does this score tell us ? 

2. How can we improve these scores ? 

Thanks

Veena

[MayurSatav]

Thanks @CedricHuesler  for sharing this update! I’m excited to see how LLM Optimizer can help us better understand and optimize agentic traffic. As someone working with clients who have highly sensitive data, I’m particularly interested in learning more about the privacy and security measures in place. How does LLM Optimizer handle sensitive content and ensure compliance with data protection requirements?

[EstebanBustamante]

Great work, @CedricHuesler — awesome to see this is now GA! Is there any extra cost for AEMaaCS clients? And are there any limits, like quotas or usage thresholds, that we should keep in mind?

[narendragandhi]

This is looking awesome @CedricHuesler . I ran the chrome extension for some of the sites and got different score and it was great to be able to visualize what the LLMs can see for those sites.

Also, currently the AI searches seems to be pulling content from any available sources so there is a high chance that some of these sources may not be credible? Is there anything specific consideration being made around this scenario in the product? Like an allowlist/denylist of sources?

I am looking at this article which goes in great detail on this topic - https://www.gsqi.com/marketing-blog/ai-search-core-systems-anti-spam/ 

Thanks

Narendra