How to reverse-engineer LLM brand visibility and why it matters

How do you get your brand recommended by AI search platforms? This has been the hottest topic in search for a while now — and will likely continue trending throughout 2026. The SEO playbook has been rewritten for AI discovery, and while traditional SEO fundamentals can be used as a strong starting point, the entire process is not the same. 

Optimizing for AI discovery means dealing with fragmentation (taking multiple platforms into consideration). Each LLM has its own models with its own cut-off dates and is built better for certain tasks, making it a challenge to determine what truly works for AI search optimization.

Your customers search everywhere. Make sure your brand shows up.

The SEO toolkit you know, plus the AI visibility data you need.

Start Free Trial

Get started with

In this article, we’ll provide guidelines for how to reverse-engineer your brand visibility on AI search platforms. Whether you call it SEO, GEO, AI search, or something else, you’ll be equipped to do your own brand analysis by the end of this piece. 

Understanding how LLMs surface brands

There are primarily two ways to get your company to surface inside LLM recommendations, and the first is through the training data. If your brand appears on the pages or domains the LLM used in their training model, your brand will be known to the model and could appear in its outputs. The answers produced from training data are very hard to influence because you can’t know how the data was processed or even which pages on your website were used, if any.

Note: “Model” refers to the version of an LLM, such as ChatGPT 5.3 (from OpenAI) or Opus 4.6 (from Anthropic’s Claude). Models are only periodically updated, usually once every few months.

The second route to LLM recommendation is via retrieval-augmented generation (RAG), which involves supplying LLMs with external context to produce more accurate, relevant, and targeted outputs. RAG will use webpages to validate answers. This is the more likely way to appear in recommendations, because new pages could become eligible for RAG in a matter of hours.

Some other highly relevant factors for getting used as an LLM source include:

• Consensus: Multiple independent sources all citing your brand increases confidence in its inclusion. If the sentiment they express is consistent, it makes your brand appear more trustworthy.
  
• Contextual relevance: The answers provided by LLMs will cite and recommend brands associated with the topic of the query. In other words, the brand must be topically relevant to the prompt to get mentioned.
• Entity understanding: Clear and consistent information about a brand makes it easily recognizable as a distinct entity. Mentioning your brand and associated properties across social media channels, articles, and website will build a stronger entity.
  
• Structured data: While there’s a big debate about whether structured data is used directly by AI search platforms, we do know that it’s used by Google to build its knowledge graph, which is utilized by Google’s AI Overviews.

Reverse-engineer results through prompt and output analysis

While investing in tracking tools is important, it’s critical to understand both what to track and what to do with it. Otherwise, you’re just tracking prompts. 

Start with a list of prompts. For this example, we’ll pretend we’re looking for a restaurant suitable for the whole family:

• “I’m looking for a family-friendly restaurant in San Francisco”
• “What are the top-rated family-friendly restaurants in San Francisco?”
• “Recommend five interesting restaurants to visit with a five-year-old in SF”
• “List some restaurants to visit while in San Francisco with my family”

The list above includes four prompts with small differences: Some are more broad (e.g., “family-friendly”) while others are more specific (e.g., “with a five-year-old”, “top-rated”), but all have similar intent.

Take note of what restaurants are mentioned, what the source URLs are (in practical terms, that’s what “RAG” means), and the dates these pages were last updated (if available). Do this daily for a week, taking note of brands, sources, and dates. LLMs prefer recently updated pages, which is why the publishing/updated dates are so important.

Test different prompts strategically

Your list of prompts to test and track will continue to grow. Think about what your business is well-known for and create more prompt variations based on this, such as:

• “What restaurants in San Francisco are fun for kids?”
• “Are there any themed restaurants in San Francisco for children?”
• “Where should I eat in San Francisco with a young child?”
• “I’m visiting San Francisco with a five-year-old — where should we eat?”

LLMs will respond differently depending on phrasing, even when the intent is the same. There’s no official data source for the exact prompts people use or the demand behind them — something SEOs got used to when reviewing keyword search volume. 

However, there are ways to make educated guesses when exploring which prompts to track.

You should also create prompts where your brand is directly mentioned. This will provide a sense of how LLMs perceive your brand:

• “Is [brand] return policy reliable?”
• “Are [brand] products durable?”
• “Does [brand] ship internationally? Are there any extra fees?”
• “Compare [brand] and [competitor] in terms of [feature]”

Compare outputs across multiple AI tools

Before AI disrupted search, you’d probably just check your keywords on Google. However, the increased use of multiple LLMs is making the space more and more fragmented. ChatGPT, AI Mode, Gemini, Copilot, Claude, and Perplexity are some of the most common. Each is trained differently, with its own sources, biases, filters, and rules. Naturally, the results can be wildly different. For example, you might be popular on AI Mode but invisible on Claude.

Run your prompts through each platform and take note of the brands mentioned, sources, dates, and so on. You might start noticing patterns, such as the same sources powering multiple answers.

Compare the results across multiple LLMs. For example, the table below includes the different results for three AIs that were prompted with “What are the top-rated family-friendly restaurants in San Francisco?”

Over time, you’ll learn which sources are the most relevant for your business. Focusing your optimization efforts on these sources will help you influence the LLM results. 

Note: Since LLMs are generative AI platforms, you’re unlikely to get the same results every time.

Results are probabilistic, so you should regularly review your prompts and use them as a guide. Remember that you’re not pursuing a specific ranking, which is a common SEO metric that isn’t really relevant for AI search.

Pro tip: When analyzing your own website’s performance as a source, there are two additional reasons why it might not appear in LLM outputs:

• Cloudflare CDN automatically blocks AI bots for new domains. If your website runs through Cloudflare, you should specifically request to allow AI-related bots.
• Bots have been blocked by your site’s robots.txt file.

In most cases, unless you specifically took action to block LLMs from accessing your content, your website should be available. Your website could technically block certain AI bots, but an LLM may still know about your company because it learned about it from third-party sources.

Since July 2025, Cloudflare has been blocking new domains from AI scrapers automatically. Existing clients who signed up before this date are not affected and must select which bots to block, if they require any.

Track patterns in cited sources and mentions

Do your prompt tracking while not logged in to your normal account on each LLM. These platforms learn enough about you to know what type of answers you want and could favor your brand through this personalization. Obviously, this isn’t ideal when measuring how a company is promoted by the LLM to others.

If a specific page or domain is consistently cited as a source in your prompts, it’s a good idea to find a way to get mentioned on this page or domain so that you can also become a source in the future.

Omniscient ran a study about what types of sources vary by user intent. This illustrates how LLM understanding of your brand is defined by a large pool of source types: your own content, editorial sources, social media, review sites, and directories.   

In a way, brands must still rely on known methods to get their names out there: social media, blogs, and media coverage. While AIs can give you one answer, they still often rely on external sources. For each prompt you want to track, tag the types of websites used to create an understanding of your brand. This will help you direct where efforts should be made.

Your prompt-tracking table or spreadsheet should include:

• Prompt
• Brands mentioned in LLM outputs
• Sources cited (specific URLs)
• Breakdown of source types (media, own website, social, etc.)
• Prompt intent (comparison, reviews, features, etc.)

AI platforms often show results similar to Google’s, to the point that Google is suing SerpApi for scraping and selling its data. Google hasn’t named any names, but OpenAI was listed as a client of SerpApi, and multiple independent tests show evidence of OpenAI using Google SERPs. If you’re very short on time and resources, continue to focus on Google — including AI Mode and Gemini — as your best bet.

Infer actionable insights

During your your observations, you have the opportunity to identify real insights — that if properly acted upon — can lead to improved performance in LLM mentions and outputs:

• Missing sources: Watch for specific URLs from external websites cited in AI answers that mention competitors but not your brand. These are opportunities for outreach. Getting mentioned on these pages or domains will increase your chances of earning AI mentions.
• Topic opportunities: Take note of prompts where only your competitors are visible. Creating content specifically for these is a decision that sometimes involves the business more than marketing. Only create content for these prompts if you want your brand to be known for these topics (e.g., Apple doesn’t want to be known as a budget phone and generally doesn’t publish content on that topic). 
• Cited sources: The URLs mentioned by AI systems when generating answers regarding your business. These sources influence how your brand is represented in AI and traditional search. You may ask for updates or changes, or in some cases (like Yelp or Google Maps) influence with the help of your customers.
• Key sentiment drivers: What factors contribute toward positive or negative perceptions of your brand? By carefully monitoring these drivers, you can identify areas for improvement that could change the discussion around your brand.

Taking steps to optimize for these areas should lead to a higher probability and frequency of mentions when someone is looking for your product or service. Keep in mind that a metric like AI visibility is a proxy for performance rather than a guarantee. Prompt output inconsistencies are expected, as indicated by this SparkToro study. AI brand visibility tracking is still evolving and isn’t yet 100% reliable.

Identify visibility gaps and competitive patterns

By now, you should be familiar with how LLMs perceive your brand. You might notice they mention your brand and products using language similar to what appears on your website, social channels, customer reviews, online profiles, and press releases.

This shouldn’t be surprising — if customers have certain pain points, complaints, or compliments that get frequently mentioned across the web, they’re likely to be reflected here. Likewise, you might also notice that you have some visibility gaps: topics where you should be getting mentions but are largely absent.

While the data you need to see these patterns can be manually collected, doing so is time consuming. And of course, analyzing so many prompt answers once you’ve finished the data collection then requires even more time.

Using a prompt-tracking tool will help you pull actionable insights from these AI outputs. Here’s an example from Semrush’s AI Visibility Toolkit: Within Semrush, go to  “AI Visibility” > “Perception” > enter your domain > “Analyze.” 

This will generate a report similar to the example below, which is an analysis of a large collection of prompts with answers in a digestible format. The report breaks the information down into two areas: “Brand Strength Factors” and “Areas for Improvement.”

Some of the insights provided might be obvious, and you should take them with a pinch of salt (e.g., luxury brands defined as “expensive”), but many others can be truly exciting and interesting. For example, imagine that a brand doesn’t offer a warranty on its products but then decides to start offering them at a later stage. It will take some time before LLMs pick up on this information, and it’s your job as a marketer to “teach” the bots about this news.

For Warby Parker, we can see in the screenshot above that a strength is “Customer-friendly policies and service (30‑day returns, 6‑month scratch guarantee, good support)” This is definitely something to put customers at ease before buying a premium product.

In the areas for improvement, it suggests, “Product and RX limitations (no traditional bifocal glasses, constraints for very strong/complex prescriptions, no reglazing non‑WP frames).” Changing this is a business decision with many implications (e.g., costs, warehousing, and staff training). It may not be something the company wants to invest in — your job as the human in the loop is to decide which actions to take.

Another interesting report is the “Business Drivers” report. This report can provide a clear view of where your brand has strong sentiment and visibility, as well as where your brand is weak or absent compared to your competitors.

The AI Visibility Toolkit and similar tools allow you to compare your brand to competitors over multiple areas. Reviewing this data will give you both a solid idea of where your brand has visibility gaps and what your competitors are doing to appear in these answers. After identifying these patterns, you can then apply these strategies to your own content.

Build the signals that influence brand inclusion

Similar to SEO, the narrative around your brand in AI search is not directly controlled by you. In traditional search, humans click on multiple results to view reviews, find discounts, and read multiple pages until they reach a decision that hopefully leads to a conversion.

LLMs do something similar: They look at multiple sources, expand the search through query fan-out, and then summarize multiple answers into one output, often using RAG to validate their answers from multiple sources.

While LLMs use webpages to create and validate their answers, they actually use many different factors and sources. This means there are multiple areas you should invest in to influence LLM outputs:

Measure your brand presence and continuously iterate

Unlike traditional search engines, which return a list of webpages, LLMs generate answers in a probabilistic way. Each response is generated on demand, and results are highly personalized. However, you can still measure how your brand performance improves over time.

Measuring your brand presence will help you identify top competitors, key sentiment drivers, and which content types are most used as sources. Track how your brand performance in AI answers progress over time by monitoring key metrics:

• Number of mentions
• Number of citations
• Cited pages
• AI share of voice (AI SOV)

As you iterate and make improvements based on the insights you’ve identified, you should see all of these metrics increase over time.

You should also track how sentiment toward your brand changes over time. Obviously, the more favorable the sentiment, the better. Again, if you’ve identified the correct insights and taken appropriate action, you’ll hopefully see positive trends in the sentiment AIs display toward your brand in their answers.

Take control of how LLMs see your brand

If the rules of SEO have always been difficult to ascertain, optimizing for AI discovery is even more obscure. Professionals are still discovering the best methods, and some strategies, such as self-promotional listicles and unverified/automated AI content generation, might be short-lived.

See the complete picture of your search visibility.

Track, optimize, and win in Google and AI search from one platform.

Start Free Trial

Get started with

Even the AI platforms you should focus on are likely to change over time. ChatGPT has been a consistent leader in the space, while Gemini and Claude have been gaining in popularity. They reach different audiences for different purposes, and in a fragmented market, optimizing for just one platform won’t make the cut. Try Semrush One to see your AI visibility and continue your journey to reverse-engineer LLM outputs and increase your brand’s visibility.