How to optimize for LLMs (Playbook)
Here's how AI notetaker Jamie dominated LLM visibility
đ Hey, Iâm George Chasiotis. Welcome to GrowthWaves, your weekly dose of B2B growth insightsâfeaturing powerful case studies, emerging trends, and unconventional strategies you wonât find anywhere else.
This note is brought to you by Attio.
I launched a new company a month ago.
The CRM we use is Attio.
And I wouldnât recommend it here if we werenât customers ourselves.
Itâs a modern CRM that doesnât feel like it was built in 2008.
Clean UX, fast, and powerful where it matters.
Some of the features we rely on:
Enrichment and AI fields that actually save time
Call intelligence to analyze conversations and spot signals
Powerful workflows to automate the boring stuff (e.g., company research)
Sequences you can combine with workflows for outreach, follow-ups, and pipeline building
They also have a solid freemium plan that is perfect for small teams that want to build momentum without the usual CRM friction.
In our second guest post on GrowthWaves, my friend and Co-Founder John Ozuysal shares a playbook based on what his team did at AI Notetaker Jamie to increase visibility in AI search engines.
A few weeks ago, I came across a LinkedIn post where Jamieâs team broke down the results they saw.
âThat could be useful to GrowthWavesâ audience,â I thought.
Plus, AI search optimization keeps coming up in conversations with clients, partners, and friends.
So, I asked John if we could feature the playbook as a guest post on GrowthWaves.
He said yes, so here we are.
John takes it from here.
Enjoy!
Introduction
ChatGPT Search was released on the 25th of July, 2024, with the goal of fundamentally changing how people search for information and solutions.
The concept was simple: why should users have to read through thousands of words to find what they are looking for and spend hours doing due diligence when they could simply get a recommendation?
Instead of searching for ââbest email marketing software for small businessesââ and scrolling through articles, landing pages, and comparison guides, users can now get a direct recommendation of the few best tools, and users can provide as much context about their business, niche, ideal customers, and use case as needed.
In this new golden land of opportunity, where advertising has not come yet and itâs at its purest form, SEOs around the world started trying to ââcrack the codeââ and ââGenerative Engine Optimizationââ (GEO) was coined.
In the beginning, it was all about making assumptions, but as time went on, SEOs like me have been experimenting with various ways of getting ChatGPT Search and other AI LLMs like Perplexity to recommend the brands we work with.
Itâs similar to how we all started with SEO in the beginning â trial and error, combined with reverse-engineering of the algorithm until you start finding patterns.
Note: After months of trial and error, we managed to get Jamie AI, to be recommended as the best AI note-taking app on the market â ahead of established competitors like Fireflies and Otter AI.
I wanted to create this GEO playbook so I can share my findings and approach to getting recommended in AI search engines based on the work Iâve done for Jamie AI, as well as other best practices that Iâve seen from other brands.
But first, letâs get started with the basics.
What do we know about AI LLMsâ recommendation algorithm so far?
Even though AI LLMs do not explicitly disclose their recommendation algorithm, here are a few things the industry has noticed when optimizing for them:
AI LLMs like ChatGPT Search are looking for accurate, updated, and relevant information.
Similar to Google, AI LLMs have a preference for high-quality, well-written and relevant content.
AI LLMs are logical â they want to read your arguments on what makes you stand out from the competition and the #1 choice for customers.
AI LLMs focus on content closely aligned with the userâs question and organized to facilitate the extraction of useful details.
AI LLMs look at the whole picture â your content, website, social media channels, brand mentions (especially from leading publishers), and review websites like G2.
AI LLMs source their information from educational content and listicles,and focus less on home pages.
What is the difference between how Google and AI LLMs interpret and recommend solutions?
Fortunately for all of us, AI LLMs like ChatGPT and Perplexity provide sources from which they are basing their claims.
That means a big part of the reverse engineering of the algorithm has come +from analyzing the ââsourcesââ section of AI LLMs.
Here are the key differences between how Google and AI LLMs interpret and recommend solutions:
If you were to search for ââbest email marketing softwareââ on ChatGPT Search, youâd receive a synthesized list, such as âThe best email marketing software includes Mailchimp, Constant Contact, and Sendinblue, based on ease of use, integrations, and pricing,â with a brief explanation for each choice.
And if you were to search for the same thing on Google, youâll see a list of links-review sites, blog posts, vendor pages-ranked by relevance, authority, and most likely ads.
What is the fundamental difference between how users search on AI LLMs and on Google?
AI LLMs like ChatGPT Search and Perplexity provide users with what Google has not been able to do: the ability to provide as much context about your situation as needed, and to get a concise answer.
Users are looking for quick and reliable information that is backed up by credible sources, and are not looking to read a 3,000-word article on why your brand is the best.
They simply need a recommendation with a 100-word summary of why your brand is a good fit for them.
Letâs take a look at this example of me prompting ChatGPT Search to find me accounting software with plenty of context:
Note: The reason why ChatGPT Search has been gaining so much traction and share of search recently is because people know that they will get an accurate or good enough result if they provide a complex prompt, providing their industry or needs.
Another reason for search behavior is that users are likely to follow up and ask more questions in the AI LLM chat to better understand the proposed solution or learn more about the topic.
In Google, this user behavior gets you punished. If the user did not end their search journey with your content, then thatâs a negative sign to Google that your content was not useful enough.
Thereâs even a word for this (pogo-sticking) for when users click your page but quickly return to the results page, which is being used as a quality indicator.
Compared to AI LLMâs "exploration is good" model, Google's is more about "one-and-doneââ satisfaction.
Setting the fundamentals right: How AI LLMs can access and interpret your content
#1: Making sure LLMs can access and read your content
First things first, before we proceed to the ââtips and tricksââ section, we need to make sure that LLMs can access and read our content in the first place.
AI LLMs, unlike Google, cannot render JavaScript. They can't render or interact with dynamic web pages the way a browser does.
That means our goal is to make sure that our websiteâs content is predominantly in the output HTML without JavaScript (i.e., the content of the page should be visible in the raw HTML output with JavaScript disabled).
You can do a quick check on your content to see if it depends on JavaScript:
Go to a page that youâd like to appear in LLMs for whatever reason. Make sure to remember what it looks like and what content it has.
Press F12.
Press CTRL + Shift + P.
Type in ââDisable JavaScript.ââ
See if the whole page has remained the same or if there is content that is not there.
Thatâs not where it ends.
You should also open your Robots.txt file and see if you have blocked crawlers from crawling certain aspects of your pages (e.g., developers sometimes make the mistake of blocking code packs).
#2: Add structured data to help AI LLMs better understand your content
Structured data, also known as schema, helps search engines like Google understand website content.
Context includes (but is not limited to):
Comprehensive company information.
Compliance certifications.
Product ratings and reviews.
Author name.
Price of the product.
As a result, it has been a prominent recommendation in Technical SEO for years.
Benjamin Tannenbaum, a hardcore SEO, found through a test that AI LLMs like ChatGPT and Perplexity also crawl and take into consideration your structured data.
The test was simple: Create 2 identical websites â one with schema and one without.
The results were obvious: The website that featured schema provided more context about the company, with the on-page content being exactly the same.
Hereâs a breakdown of the complete results:
Hereâs the schema that works best according to the research:
Organization Schema.
Certifications Schema (e.g., ââISO 27001).
Product Schema with Ratings, Reviews, and Price.
10 best practices to get recommended on AI LLMs
Here are the best practices to get recommended on AI LLMs based on my experience so far and the initial best practices in the industry:
#1: Brand mentions
The #1 factor that seems to work best when optimizing for AI LLMs is brand mentions. Brand mentions are when other websites mention your brand.
Brand mentions (both linked and unlinked) function as signals of entity prominence, semantic authority, and contextual relevance.
But whatâs the logic behind this?
LLMs build âentity embeddingsâ by processing vast corpora. When a brand is repeatedly mentioned, especially within topic-relevant contexts, its embedding becomes stronger and more tightly connected to related concepts (e.g., HubSpot as the best CRM on the market).
As a result, the cosine similarity between that brandâs embedding and relevant query vectors increases, making it more likely the LLM will surface that brand.
It can be in citations or inbound links â the goal is to get them from visible sources and trusted brands within your industry.
Test to try: One thing Iâm keen on trying is community-driven brand mentions. That includes encouraging your clients or customers to share their stories on personal blogs or Q&A forums (e.g., Stack Overflow, Quora). These user-generated touchpoints can feed into the broad training data LLMs ingest.
#2: Move away from marketing talk when describing features
The second thing you should do if you are serious about getting recommended in AI LLMs is to move away from marketing talk.
When you are describing your brand in comparison guides or even on your home page, you should logically explain to both your users and search engines what makes you different and why your solution is a smart choice.
Example of marketing talk: ââWeâre an award-winning returns management solution that works with some of the biggest enterprises in the world. We are returns.ââ
Example of logical argument: ââOur returns management solution is built for retailers looking to optimise and automate their returns process across the UK and 170+ other countries. Our returns app lets you build a custom returns portal and offers your shoppers different return options that help you keep revenue in the business and lower the costs of returns.ââ
Be it in your homepage or your article content, you need to clearly articulate and explain your companyâs strengths and what makes you different.
That also includes writing clear headings, bullet points, and paragraphs so it can be easier for LLMs to parse your content.
A few pro tips on writing:
You should be writing in a conversational tone(like they do) to make your content more relatable for users while aligning with how AI LLMs analyze and summarize information.
Write like you're explaining to a smart friend, not an algorithm.
Write in short facts with information compression. That means conveying the same meaning with fewer words.
Consider information priority. Always start with the most important information (ditch the storytelling at the beginning of a section) and prioritize the order in which you explain your reasoning and solutions. This is because AI LLMS want to cite specific pieces of important information that are best for the user concerning their query.
#3: Large-scale BOFU content creation that clearly explains how your brand is better than the rest
I think Iâve noticed in the ââsourcesââ section of ChatGPT and Perplexity Search that they take into consideration listicle content (e.g., 10 best accounting software for small businesses) and do not manually go through each website.
This is why, if you want to be recommended for the best accounting tools for small businesses:
You need to create a listicle with the best accounting software providers for small businesses.
Position your product as #1.
Explain in good enough detail what makes you different from your competitors and how your solution can help small businesses.
The same logic does not only apply to software businesses, but also to service businesses, such as marketing agencies.
Your goal is to exhaust feature, industry, and competitor comparison opportunities, as they are most referenced in AI LLMs.
Feature pages: best gantt chart software
Industry pages: best project management software
Competitor comparisons: best Asana alternatives
The key is to make the content digestible for both the users and AI search engines so they can get the most crucial information about your solution as quickly as possible.
This is why you should focus on explaining your best use case(s), standout features, pricing, and highlights for this category (e.g., what do you offer small businesses if you are writing ââbest email marketing software for small businesses).
#4: Work to get mentioned on ââbest software for Xââ listicles to build consensus
Like I mentioned above, I have noticed that AI LLMs are looking for consensus and existing popularity.
ChatGPT, a bit similar to Google, likes to play it safe and recommend already popular brands.
For example, you can rank #1 on Google for ââbest Asana alternativesââ with your brand, but not be recommended on ChatGPT Search.
Why? Because youâre not mentioned in the other listicles. Thereâs only 1 place that says youâre the best or one of the best in the category.
Even though ProofHub ranks highly for ââAsana alternativesââ, theyâre not in the top 5 recommended tools for that prompt in ChatGPT Search.
The winner, ClickUp, is mentioned in a lot of these listicles.
Personal prediction: We might see an era in content creation, where we are being extra selective on what brands we are mentioning as the 2nd or 3rd best in the category, so as not to give them a boost in AI LLMs.
If you are working on a brand that is not already massively popular, what you can do is to secure mentions in ââbest software for Xââ or any ââbest solutionââ industry listicles.
#5: Align with a PR team to get you mentioned in media outlets that have agreements with LLM providers
AI LLMs need to get their data from somewhere to train their models. This is why they have agreements with large media outlets so they can study their content.
You can leverage the fact that the media outlets with a partnership with LLM providers like ChatGPT Search are public, so you can try and get mentioned there.
Media companies include:
Associated Press.
Axel Springer.
FT Group.
Dotdash Meredith.
News Corp.
The Atlantic.
Vox Media.
Guardian Media Group.
You can check out each media website of these publications here.
Hereâs an example of a brand getting recommended for ââbest returns systemââ after being ââmentionedââ as the best returns system by The Guardian.
#6: Strengthen your profiles on review websites, such as G2, Capterra, Reddit, and TrustPilot
Another observable factor in the ââsourcesââ section is the fact that AI LLMs want to provide users with unbiased information by tapping into reviews from platforms like:
G2.
Capterra.
Reddit.
TrustPilot.
This is why I recommend the brands I work with to invest in strengthening their profiles on review channels like G2.
By ââstrengtheningââ I do not mean putting fake reviews, as some companies have started doing, but rather asking your happy customers to leave reviews.
#7: Verify your domain in Bing Search Console & track indexed pages
The most popular AI LLM, ChatGPT Search, uses Bing's search results as a primary source of information.
That means if your website isn't indexed in Bing, it won't be present in ChatGPT Search results.
This is why itâs worth verifying your domain in Bing Search Console and keeping track of crucial pages that are not indexed on Bing.
Rankings on Bing also matter, as we can see from this research done by Seer Interactive, which found that thereâs a strong correlation between LLM mentions by the keywords you rank on Bing.
#8: Build up your social media presence
Hereâs when things start getting interesting: you can often see social media posts appearing in the sources.
AI LLMs are also browsing through platforms like TikTok to find you the best possible solution for your needs.
This is why further building up your social media presence, including publishing relevant content on YouTube and TikTok, can mean AI LLMs considering that context.
#9: Keep your content fresh with up-to-date information
AI LLMs want to provide their users with up-to-date, fresh information that is correct.
This is why you should regularly update your older content to ensure it stays relevant and useful to readers and AI search engines.
For the clients I work with, I like to revisit their article content even if performance has not dropped â content hygiene is important for not only optimizing for search engines, but also for users.
#10: Produce content that the AI LLMs are not pre-trained on
There was a recent Claude system leak that showed how the tool would avoid linking out to creators on the internet when it already has that data, pre-trained (e.g., how much water to drink per day).
Now, what does that mean for us?
We should be creating original content that LLMs have not been trained on (e.g., the latest information).
So that means, in theory, if we want to rank in LLMs, our content should be:
Short, clear, and well-argued, so itâs quotable.
Unpredictable, original, and ideally not pre-trained.
How can you track and evaluate success from AI LLMs?
Tracking traffic generated from LLMs
You can track the traffic generated from LLMs by looking at your Google Analytics platform.
Itâs easy to track because every click has a UTM_source tracking link.
Tracking leads generated from LLMs
The most effective way to track leads generated from LLMs is to look at your CRM, such as HubSpot, to see the first point of contact of how the contact was created.
Tracking recommendation positions from LLMs
You can use a tool like Peec AI (thatâs the tool I use) to track brand mentions in LLMs and monitor prompt performance to get an idea of your brandâs overall visibility.
The way it works is that you can:
Track and follow conversations where your brand is mentioned.
Measure your performance against your competitors on who is getting mentioned more in AI LLMs.
Track your brand mentions across custom prompts and AI platforms.
Get instant alerts when your visibility changes and see exactly where you need to optimize.
Tracking how AI perceives your brand
Last but not least, I wanted to show you a tool that can show you what AI LLMs know about your brand (with their current and pre-trained data) and how they perceive you.
The software is called Waikay and shows you the facts that different LLM models know about your brand.
You can then flag wrong facts, and the tool will give you recommendations on how to fix these mistakes.
It also shows you how different AI models perceive your brand, such as ChatGPT, Gemini and Perplexity.
Iâll let George wrap this up and close with some final thoughts.
Final thoughts
This is the second post weâre doing on LLM optimization.
The first one was once again a case study on how to rank on AI search.
It seems that AI search is quickly emerging as a new marketing channel.
Thatâs a good thing for marketing.
Google and AI search can co-exist and expand the channels available to drive business impact.
Iâd like to stress one thing, though:
There is a lot of snake oil salesmanship currently in LLMO, as there is (and was) in SEO and virtually every other marketing channel.
We all fall for shiny objects now and then, and unfortunately, trends like LLMO or GEO only amplify that.
Thatâs not to say that you shouldnât look into it.
Just be careful, and keep the lessons of other marketing channels at the top of your mind:
Always diversify, and never keep all your eggs in one basket.
Good luck!