Here's everything I learned on how to rank on LLMs & do AI Optimization

Episode 98

May 06, 2025

Friends,

The way we search has changed and so has the way we show up. As large language models like ChatGPT, Claude, and Perplexity become our go-to interfaces for discovery, the old SEO rules aren’t enough.

Welcome to the age of LLM Visibility Optimization (AIO).

In this essay, I’ll break down how LLMs actually work, how they surface your content, and what you need to do to show up. (based on my real experiences)

Brought to you by:

Market Curve is a premium tech storytelling studio run by me Shounak that turns tech companies into media companies. I’ve had the fortune of working with some of the top tech companies backed by some of the top investors like YCombinator, 20VC who have gone on to be acquired by companies like Roblox & Amplitude.

Here are a few things I can help you with —

Design a custom content system that helps your brand show up on LLMs.
Optimize your entire buyer journey across your landing page, pricing page and onboarding flow so you can get more customers without spending more on customer acquisition costs.
Build custom AI agents for your marketing/GTM/content teams that are embedded deep into your tech stack and can function autonomously.
Take over your social media profile via my Service with a Software system which will help you get more followers & help grow your personal brand without breaking the bank.

Interested?

Let's talk!

Okay so how on earth do LLMs work?

First things first.

In order to show up on LLMs, we first need to understand how LLMs work. If you ask ChatGPT to give you pricing data about a SaaS product (like Slack), it doesn’t have training data on that.

So if it shows you an answer on Slack’s pricing, it will fetch relevant information from the web. That’s the concept of citations/sources, which is where you want your product to show up.

For example, you might think that Perplexity simply uses Google Search to show results. So you think to yourself “if I rank on Google, then I’ll show up on Perplexity by default”.

But that’s not the case. Instead, Perplexity uses its own web crawler PerplexityBot which gathers information from the internet to index for its search engine.

While it does use multiple inputs and does indeed reference Google’s & Bing’s web rankings to inform its search results, it does not solely depend on Google or Bing’s data.

Aravind Srinivas has said that while Perplexity does use Google’s ranking results, but only if the AI deems them the best indicator of link quality. But they do not copy Google’s search results themselves. So even though you might think Perplexity is a fancy Google wrapper, it’s not. It has its own search functionality powered by its native AI models.

Let’s compare:

If I type into Google “best onboarding tools for B2B SaaS”, I get an AI overview of tools:

If I scroll further down, then I can see Reddit discussions pop up:

So here we see that Reddit is a great source for companies to rank on Google’s AI overview.

Two things matter here: (1) The domain authority of blogs like Userlist, Pendo etc. (2) Forums and communities discussing the best tools. If a tool lies at the intersection of both (1) & (2), it gets picked up by LLMs.

Case on point: My Reddit post on the top 7 user onboarding tools features both Pendo and Appcues. Safe to say both tools are featured in the AI overviews. The reason is that Google & Reddit have entered into a partnership. In February 2024, Reddit and Google announced a data licensing partnership worth $60 million per year. Which is why you can see Reddit threads pop up in Google’s first page and referencing the Reddit post as a source in its AI overviews.

These are the remainder of Google's SERP data.

Now let’s look at Perplexity -

Right off the bat, you can see that Perplexity’s sources are different from Google’s above.

And though there is overlap across the top 4, the remaining 3 entries of ProductFruits, Intercom, Moxo are native to only Perplexity and not Google. Product Fruits & Moxo are nowhere to be mentioned on Google.

Which shows that Perplexity isn’t just a frontend wrapper for Google. It sources data from Google yes but isn’t solely dependent on Google for its results.

What about ChatGPT?

ChatGPT has the usual suspects but there are a few interesting points of difference here. For one, much like Google, ChatGPT picks up Reddit threads.

My Reddit post is once again picked up by ChatGPT much like how it was picked up by Google. This again confirms that posting on Reddit in addition to shipping on your blog is useful if you want to get picked up by LLMs.

The second point here is that companies like Whatfix and Coursebox weren’t picked up by Perplexity or Google but it is picked up by ChatGPT.

When a query is made to an LLM, the AI model uses algorithms to score content based on relevance. And the thing about large language models is that they go beyond simple keyword matching, when retrieving answers.

When a user asks an AI tool like Perplexity: "Which is the best product onboarding software?" The AI tool processes this query something like this:

Step 1: It identifies keywords first - words like “best”, “product onboarding”, “software”.

Step 2: It does a semantic analysis - “best” equals top-rated, most effective, highest quality.

Step 3: Topical relevance - “app development”, “user engagement tools”, “customer communication systems”, “recent awards, reviews”

Step 4 - And then the AI scores the available content based on relevance.

This is how most LLMs source data to show to your users based on their search query.

Next step: Make sure your content is structured in such a way that makes it easy for LLMs to pick them up

Now that you know how LLMs work, you want to make it easy for them to find your content.

This is similar to the technical optimization we are so used to doing for SEO, like writing meta-titles, meta-descriptions, adding alt-tags to Google etc. Except now, we gotta do the whole thing but for LLMs.

1: Add clear metadata:

LLMs rely on structured data to improve comprehension, entity recognition, and categorization. Without clear metadata, AI models will misinterpret your content or fail to associate it with relevant queries. How do you make sure LLMs don’t do that? Use Schema.org markup - this way, you provide explicit signals about your content’s meaning, increasing the chances that LLMs will reference your information correctly.

I recommend the following schema:

Article (schema:Article): Helps AI understand blog posts, news articles, and general web content.
FAQPage (schema:FAQPage): Ideal for structuring frequently asked questions so AI can directly pull Q&A pairs.
HowTo (schema:HowTo): Guides AI in categorizing step-by-step instructions for tutorial-style content.
Product (schema:Product): Ensures LLMs correctly categorize eCommerce and SaaS-related content.
Organization (schema:Organization): Establishes credibility and links your brand with authoritative mentions in AI-generated outputs.
Person (schema:Person): Useful for personal branding and expert author recognition in AI-generated summaries.

To do this, use Google's Structured Data Markup Helper. Select the appropriate schema type and enter the URL of your blog post. Tag the relevant elements on your page & generate the HTML code which will be in JSON-LD format. Copy the generated JSON-LD code & paste it into the code of the individual pages.

This is an example of Json LD code:

{

"@context": "https://schema.org",

"@type": "Person",

"name": "John Doe",

"jobTitle": "Software Engineer",

"worksFor": {

"@type": "Organization",

"name": "Example Tech"

}

Fun fact: Json LD is how you fine-tune LLM models in ChatGPT too.

2: Optimize for LLMs.txt:

If you’re an AI startup or have an AI-use case, you probably have a documentation page or knowledge base page.

In that case, you should know about the llms.txt file. It’s a web standard designed to help websites communicate effectively with LLMs like ChatGPT, Google Gemini, and Claude.

It’s like robots.txt but for LLMs. While robots.txt controls how search engines crawl and index pages, llms.txt focuses on making content understandable for AI systems.

Next up - How do you get your content or brand to be mentioned & referenced by LLMs?

Now that you’ve set up the technical optimization bit, here’s how you can get sourced/mentioned in LLMs.

I will be drawing on my own experience here with examples and reverse-engineering them but also observing patterns from other successful product placements.

1. Write something worth reading.

This is an example of an article that’s ranking on ChatGPT written by yours truly. See image below:

Here’s why this works:

It answers a direct question that people (in this case founders) ask LLMs. In this case, how do b2b companies get their first 100 users?

If you go through the article, you’ll find that it's written in an engaging way, and gives super actionable tactics, which is important because it’s tied to the user intent of the query.

The point here is that you have to make sure the answer is relevant to the corresponding user query - and you can only do that by writing something people will find worth reading.

2. Share it across the internet.

Now that you have written a great piece, it’s time to build digital PR backlinks (much like Google). Here we will go to our friend Reddit and share this story. Here’s what I did for this piece above:

A digital PR from Reddit and fellow subredditors signals that this content is useful, unique and (most importantly isn’t AI generated). It’s cited from authentic sources and draws upon a personal experience, a personal story that people can buy into.

To take it a step further, you can contribute to other guests' posts via other influencers/podcasts , get on their newsletter/link roundups so you can build even more digital PR backlinks. Repost this on Medium, on Substack, a Linkedin Article and write a thread on it or a few LinkedIn posts on it (or better yet get some influencer in that niche to repost it)

3: Write pillar pages & subpages

When creating content, you want to make sure you’re as comprehensive as anything out there. Unbundle a topic into its constituent elements and go all in or amalgamate different fragmented sources of information and bring them together in one place with your unique POV.

This is an example of an essay I wrote on Hubspot’s growth strategy - its comprehensive with stories, factual analysis, growth strategies. And this is one of the sources that ChatGPT references via its user query.

You can follow a clear template here (which I’ll share more about in the sections below). As for this bullet point, make sure to be as comprehensive as possible and add your unique POV to help stand out.

4: Be consistent with your messaging.

When trying to rank on LLMs, you have to be consistent with your messaging across all webpages and all pieces of content. Coursera ranks for “best online courses” every time on ChatGPT while Masterclass doesn’t.

Why? Because Coursera uses “free online courses” as its headline across almost every page. You shouldn’t say you are the best software for founders and on a different page say you’re the best software for designers.

AI will be super confused with these discrepancies & will think you aren’t the best software for anything, and won’t recommend your product.

Okay so what type of content do you create so LLMs reference you or your brand?

Now that you know how to get your content picked up by LLMs, the next crucial step is to figure out what you exactly write about - how do you do research on what to actually create.

The first thing to know before doing research is to understand it’s not like keyword research here. It’s different because we are optimizing for user intent here where the tone is conversational.

There are four types of user intent:

Informational Intent – This query corresponds to the user trying to find definitions and learn more about a topic. For example, “What is an AI agent?”
Navigational Intent – A user is trying to find a website/specific webpage. For example, “Slack Pricing.”
Transactional Intent – The user is purchasing a product or trying to complete an action. For example, “I want to buy a gaming laptop”
Commercial Intent – A user is conducting research about a product or service before making a purchase decision. For example, “What are the best product onboarding companies 2025.”

So the first thing to do here is:

1: Create a question related to your brand.

Decide which of the four types of users you are targeting (informational, navigational, transactional, commercial). Ask Perplexity these questions and it will give you a summary with links. At the bottom of each page, you’ll find that there is a list of additional questions that users ask related to your question. This will give you insights wrt what people are interested in searching for.

2: Choose prompts you wanna rank for:

Start by choosing prompts you wanna rank for. For example, you could say that you wanna rank for “best linkedin scheduling software for creators”. If you break this down, you’ll see that this prompt can be broken down into further components:

(a) it has a question identifier (what, why, how)

(b) a qualitative element (the best, the top)

(d) niche/industry/use case identifier (for creators).

At this stage, you can then ask ChatGPT/Perplexity these questions to reverse-engineer search intent. For example you can ask it questions like:

What are the most common questions {YOUR ICP} is asking you nowadays about {topic/main keyword]. Provide a list of 10 well structured questions. Use these questions to come up with content ideas.
Ask AI “what are the best blog articles to learn about {main keyword/question} - see what they’re missing, what you can add and then write your blog in that way.
There are secondary prompts you can optimize for too. You can check “people also ask” results to rank for those prompts. In a lot of cases, follow up prompts follow a similar syntax like “why is XYZ better than ABC”, “does ABC do XYZ better”? “How do I do X task with ABC tool?”
Optimize for follow ups. If a user follows up with "What's the pricing?" You need to have a conversion-friendly pricing page ready so the AI can use it and respond. This applies to all of the objections a prospect could have. Have those pages ready so you can handle all the objections and don't leave it to the AI to make something up.

3: Scour Reddit to find content ideas:

Go to relevant subreddits to find recurring questions that people ask in your niche and then answer those (and share those back into Reddit)

This goes into the heart of product marketing here - you wanna make sure who your customer is and what your JTBD framework is - the more clarity you have on this, the more the LLMs are gonna favor you.

Before finalizing which idea to write on, classify questions whether they signify high buying intent and have adequate search volume. If buying intent is high & show decent search volume then go for it.

Proven content formats that you can create to get referenced by LLMs.

Research is always good, but if you don’t have the time, there are some proven content formats that LLMs seem to love - so creating content around these themes will have a pretty high probability of being picked up by LLMs.

1. Share exclusive research & data to show deep expertise:

For b2b companies, this means talking to subject matter experts, researching strategies and creating premium content playbooks around them. Create editorial style newsletters. Hubspot did it, Clay is doing it, Amplitufde does it, Typeform does it. Command AI does it. The list goes on. It’s the future of B2B Content.

2: Repurpose them for socials:

It’s not enough to just create content either - you have to repurpose it. You need to turn those newsletters into social media posts so LLMs can pick up on it. LinkedIn / X posts / LinkedIn articles are all crawled by LLMs.

Write LinkedIn articles. Ship them on Substack. The more authority you have on social media & on websites with high DA, the higher the chances of LLMs picking up key information and providing more context.

3: Have dedicated landing pages for each feature/use-case:

Optimize your home page/landing pages to get ranked on LLMs. Make sure you use words that resonate with user intent. Type in a prompt and see if your competitors show up.

If they do, then optimize for that prompt for LLMs to pick up. Find long tail user intent queries too - for example, what is the best ai ad UGC platform for B2C apps. Here the variable is {B2C apps} because that’s who one of your customers is. So having a dedicated landing page to show this off will help LLMs scrape your website.

4: Create comparison pages/BOFU articles & landing pages with high buying intent:

Don’t just write normal stuff - offer your unique POV to show who it is for, how you’re better than the alternatives across what factors, and why people in your target market chose you. Think about it - if you do this, there’s a person who’s choosing “what is better for my X use case? A or B product?” so if you answer that in your page, then AI will suggest your product for X use-case.

5: Ship case studies:

If you’re in an industry where case studies matter (almost any B2B industry), add a lot of case studies on site! There’s definitely some correlation between number of case studies and ranking in the LLMs.

6: Do Programmatic SEO:

Programmatic SEO is gonna be a great way to rank for GEO - create dedicated landing pages & blog posts for ultra niche specific topics to give the LLMs context on exactly what your offering is and how it is that you can help them.

7: Partner with influencers:

If your brand shows up in one of those “best of” listicles, you’ll rank higher. So you may want to partner with these bloggers, newsletters etc who are writing about your space and asking them to include you in their listicles.

Here’s how to write your content - Writing guidelines to help you get picked up by LLMs.

Now that you know what type of content to create, here’s how you actually write the content. In this section, I’ll share best practices with you based on my own experiences & those learned from others:

Danny Sullivan (Google’s Search Liaison) confirmed at Search Central Live NYC that AI Overviews are rewriting the rules of search.

The era of churning out basic blog posts for SEO is over (to some degree).

To win, you need to create content that AI can’t summarize in a single snippet. You need original research, expert insights, unique frameworks.

How to write/what to include:

Write like an answer. Use question based H2s (What is X + how does Y work) and answer those in the first 1-2 lines.
Use bullets / numbered lists & include definitions and examples.
Offer unique POV. Make it comprehensive.
Make sure to add quotes and cite your sources.
No matter your business, use FAQs wherever possible
Include references or statements that tie your content to trending topics or current events in your industry. This contextual relevance signals to LLMs that your content is timely and authoritative.
Answer questions fast - Deliver the answer in the first few sentences of a piece of content.
Eliminate fluff
Make sure the first sentence after a heading answers the heading
Avoid modal verbs (will, should, etc).
Structure sentences to place key attributes early. For example, for "What is PLG?", lead with "PLG is a growth motion where the product.... "
Structure your content like a database and less like a novel
Periodically refresh your content even if it's evergreen content
Answer complex questions with multiple parts in clearly structured content that breaks down the question into its pieces with clear subheaders.
Write in explicit problem-solution syntaxes showing clear thought processes so LLMs can pick it up (here’s the problem - here’s the solution)
Always add additional context, quotes and examples to your posts.
The more straightforward your content is, the better. If your relevant content (paragraph or heading section) fits within just one chunk, it's more likely to be used in the answer.

What tools to use to get your brand ranked on LLMs.

Use Profound to store what sources you are ranking for on LLMs. Profound stores those citations so you know what pages are being scraped so you can double down on those and make it relevant. Data flows in real time so it’s automatically added to profound. Do this at scale and this can make a massive difference.
Use Semrush AI toolkit to Understand

how your brand is being mentioned compared to competitors on ChatGPT
Quickly see how your brand is perceived in the market & how many mentions it's getting compared to competutors.
See what type of intent searchers have when asking about your brand in ChatGPT.
Review what people are searching on ChatGPT relevant to your brand.
See what ChatGPT is saying "good" and "bad" about your brand
get AI insights into how you can improve your business to rank better on ChatGPT.

Use knowatoa.com/ to see whether LLMs are referencing your brand or your competitors in its results.
Omnia - You can use Omnia to identify the most searched topics in AI engines to boost your brand visibility, see how your brand ranks against competitors across key topics and learn the source from where AI engines pull their information.
Rankscale- With Rankscale you can analyze and optimize your AI Search Visibility with results tracking, get deep insights and Actionable Recommendations. You can Track your Presence in AI Search Results over Time, Analyze AI Result Citations & Brand Sentiments, Uncover, Survey, and Outpace Competitors, Audit your website to see how AI understands it, Identify Gaps and Content Optimization Potential.
You can use Graphite to drop in a query and see which brands LLMs are mentioning.
Use Unify’s AIO checker to plug in your company, Pick the keywords that you want to see how you rank on, See how often you are mentioned or cited, and who are the top references.

In Aldous Huxley’s words, it’s a brave new world we are living in. It feels like we are on the cusp of something here. There will be losers and there will be winners. It’s in our hands to make sure which side of history we end up being on.