LLM SEO: How to Get Your Brand Cited by ChatGPT, Perplexity and Google AI

LLM SEO is the practice of structuring content and brand signals so large language models retrieve, trust and cite your pages inside the answers they generate. It replaces the ranked list as the unit of competition. Instead of asking “what position do we hold,” you ask “how often are we named, and how often are we linked.”

That distinction matters more than most guides admit. Between 27 April and 26 July 2026 we monitored 5,550 AI responses across three client brands using Searchable. The brands collected 27,929 mentions and 6,709 citations. Those two numbers are not the same thing, and the gap between them is where most LLM SEO advice quietly falls apart.

Here is what the data showed, and what we changed because of it.

What is LLM SEO, in one paragraph

LLM SEO, sometimes written as LLM optimization or SEO for LLMs, is the work of making your content retrievable and quotable by systems like ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews. It combines three older disciplines: technical accessibility so crawlers and retrieval agents can read you, content structure so a model can lift a self-contained passage, and entity building so the model associates your brand with a topic even when it is not reading your page at all.

If you want the terminology map: GEO (generative engine optimization) and AEO (answer engine optimization) describe the same work from different angles. LLM SEO is the engineering-flavoured name for it. We cover the differences in our GEO vs SEO vs AEO breakdown.

The finding that changed how we work: mentions are not citations

The single most useful number in our dataset is the ratio between times a brand is named in an AI answer and times that brand’s own site is linked as a source.

Brand	Market	AI responses	Brand mentions	Citations	Citation rate
uncap.com	B2B ecommerce, US	3,750	22,885	4,133	18.1%
tetto94.it	Roofing, Italy	900	2,384	1,216	51.0%
rama-abdichtungstechnik.de	Waterproofing, Germany	900	2,660	1,360	51.1%

Source: Searchable, 27 April to 26 July 2026. Citation rate is total citations divided by total mentions.

The B2B ecommerce brand was talked about four times more often than it was linked to. The two smaller European service brands converted mentions into citations at roughly half the rate they were mentioned, nearly three times better.

Why? The two European brands compete in query spaces with thin published content, so the model has few alternatives and reaches for the brand’s own pages. The US brand competes in a space saturated with Shopify’s documentation and dozens of agency blogs, so the model happily discusses it while citing somebody else.

What this means for you

• Being mentioned is a brand-awareness outcome. Being cited is a traffic outcome. Track them separately or you will misread your progress.
• In crowded categories, expect a mention-heavy, citation-light profile early on. That is not failure, it is the normal first stage.
• The lever that closes the gap is publishing the specific asset the model currently has to borrow from a competitor.

ChatGPT, Perplexity and AI Overviews behave differently

Every guide in the current top 10 for “llm seo” treats AI search as one surface. It is not. Same brands, same prompts, same week, three very different outcomes.

Platform	uncap visibility	tetto94 visibility	Citations issued (tetto94)
ChatGPT	20.0%	33.3%	1,061
Perplexity	14.1%	19.7%	45
Google AI Overviews	16.8%	14.7%	110

Two things jump out.

ChatGPT is the highest-yield surface for both brands. It mentioned them most often and issued by far the most citations. If you have budget for one surface, this is it.

Google AI Overviews mentions brands without crediting them. For uncap, AI Overviews produced 416 brand mentions but only 2 brand citations out of 59 total citations issued, a 3.4% share. Google will describe your company and link to a directory. Perplexity, which issued only 45 citations in total for tetto94, gave 9 of them to the brand itself, a 20% share. Perplexity is stingy with links but generous about who gets them.

We break down this surface specifically in how to rank in Google AI Overviews.

What this means for you

• Report on AI visibility per platform. A single blended “AI visibility score” hides the fact that you may be winning on one surface and invisible on another.
• If your goal is referral traffic, weight effort toward ChatGPT and Perplexity. If your goal is brand presence in the buying conversation, AI Overviews still counts even without the link.
• Do not assume a tactic that lifted ChatGPT citations will move AI Overviews. In our data they moved independently.

Branded prompts flatter you. Unbranded prompts tell the truth.

This is the number I show clients first, because it reframes the entire project.

For the B2B ecommerce brand, we tracked 28 topics. Broken down by how the brand performed:

Topic type	Example prompt	Mention rate
Branded	”Uncap Connect vs other B2B ecommerce integrations?“	93%
Category, mid-intent	”Shopify Plus agencies specializing in distributor solutions?“	36%
Category, high-intent	”Best unified commerce platforms for product distributors?“	0%

Of the 28 topics tracked, 21 returned a 0% mention rate. One was strong. The strong one was the branded topic.

A model that already knows your name will happily discuss you. The commercial question is whether it names you when the buyer has not heard of you yet, and for most brands the honest answer today is no.

What this means for you

• Audit unbranded prompts only. Branded performance is a vanity metric in AEO exactly the way branded search traffic is in classic SEO.
• Build your prompt set from how buyers actually phrase problems, not how you phrase your product.
• Zero-percent topics are your content roadmap. Each one is a page you have not written.

The content formats models actually cite

We pulled the top 25 most-cited domains across the B2B ecommerce prompt set and looked at what kind of content each was cited for. The pattern was blunt.

Content type	Appears in top 25 cited domains
Ranked list (“9 best X for Y”)	19
Topic guide	17
Comparison	3
Case study	2
How-to guide	1

Ranked lists and topic guides accounted for almost everything. Case studies, the format agencies love most, were cited twice.

This is not a claim that case studies are worthless. They convert readers. But if the goal is being retrieved by a model answering “best X for Y,” the model wants a page that already contains a comparable set of options with attributes it can extract. A case study about one client does not give it that.

What this means for you

• Your first LLM SEO asset should be an honest ranked list or comparison in your category, including competitors.
• Structure every entry the same way, with the same attributes in the same order. Consistency is what makes a table extractable.
• Keep case studies. Just do not expect them to earn citations.

Fewer, denser pages beat more pages

The Italian roofing brand is the cleanest natural experiment in the dataset.

Domain	Type	Pages cited	Total citations	Citations per page
tetto94.it	Brand	9	243	27.0
prontopro.it	Directory	74	325	4.4
pro-tetto.it	Competitor	30	194	6.5
shopify.com (US set)	Platform	1,147	5,692	5.0

Nine pages generated 243 citations and made tetto94.it the single most cited domain in its own category, ahead of a national directory with eight times the page count.

Shopify, with 1,147 cited pages, earned five citations per page. The small roofing site earned twenty-seven. Volume is not the variable. Fit is.

What this means for you

• A 9-page site can out-cite a 1,000-page site inside a defined topic. Depth in a narrow space beats breadth.
• Before commissioning fifty articles, check whether five very specific ones would cover the prompt space.
• Publishing cadence is a weak lever here. Prompt coverage is the strong one.

What actually moved the needle

Between 22 and 26 July we ran a focused intervention on the roofing brand and measured every 24 hours. Six reports, 150 responses each.

Report	Date	Visibility score	Mention rate
1	22 Jul	16.7	25.3%
2	22 Jul	14.7	44.0%
3	23 Jul	22.0	63.3%
4	24 Jul	24.0	60.0%
5	25 Jul	26.0	54.7%
6	26 Jul	32.0	71.3%

Visibility score went from 16.7 to 32.0, a 92% relative lift. Mention rate moved from 25.3% to 71.3%.

Three caveats, because I would want them if I were reading this. The window is short. The sample is 150 responses per report, so day-to-day noise is real, and you can see it in the dip at report 5. And AI answers are non-deterministic, meaning some movement happens without you doing anything. Treat the direction as the signal and the exact numbers as indicative.

The changes that preceded the lift, in the order we made them:

1. Rewrote the opening 60 words of every service page as a self-contained answer. No “welcome to our site.” A direct statement of what the company does, where, and for whom, phrased so it can be lifted verbatim.
2. Added a comparison table to the two highest-intent pages, listing approaches rather than only the company’s own service.
3. Standardised entity naming. One company name spelling, one address format, consistent service labels across the site, the Google Business Profile and the main directories.
4. Claimed and corrected the directory listings the models were already citing instead of the brand. In this category, four of the top seven cited domains were directories. We stopped fighting them and made sure they carried correct information pointing back to the brand.
5. Added FAQ blocks using the exact question phrasings we saw in the tracked prompt set.

Point four is the one most teams skip. If a directory outranks you as a source, getting your data right inside that directory is faster than trying to displace it.

A note on schema, and the evidence against it

The standard advice is to add structured data everywhere. Our own step list above includes FAQ blocks, so I am not going to pretend schema is useless.

But it is worth knowing that an Ahrefs analysis of 1,885 pages found little relationship between schema markup presence and AI citation frequency. That study is being actively discussed by practitioners, and not one of the current top-ranking pages for “llm seo” engages with it.

My read, based on what we saw: schema helps a machine parse a page it has already decided to read. It does not persuade a model to read you. The retrieval decision is driven by topical fit, entity clarity and whether a passage answers the prompt cleanly. Schema is hygiene, not strategy. Do it, then spend your remaining time on the passage itself.

The LLM SEO framework

A five-part loop you can actually run. Each part maps to a measurable output.

1. Define the prompt set. Twenty to fifty prompts a real buyer would type, phrased their way. Split branded and unbranded. This is your keyword list equivalent, and it is the input everything else depends on.

2. Establish the baseline. Measure mention rate and citation rate per platform, not blended. Record which domains are being cited instead of you. If you need a starting point on which platform to measure with, we compare the options in best AEO and AI visibility tools.

3. Fix retrieval access. Confirm your pages are crawlable by the AI agents, that critical content is in the HTML rather than rendered client-side only, and that your robots rules are not blocking the assistants you want to be read by.

4. Publish the missing asset. For each zero-percent topic, write the ranked list, comparison or topic guide that the model currently has to source elsewhere. Answer-first, self-contained sections, consistent attributes.

5. Correct the entity graph. Same name, same description, same categories across your site, your Google Business Profile, LinkedIn, industry directories and any review platform in your category. Models triangulate identity across sources, and a mismatch reads as two different companies.

Then re-measure and repeat. The loop is the product, not any single tactic.

Your LLM SEO checklist

• Prompt set built from buyer language, split branded and unbranded
• Baseline recorded per platform: ChatGPT, Perplexity, Google AI Overviews, Gemini
• Mention rate and citation rate tracked as separate metrics
• List of domains currently cited in your place
• AI crawler access verified in robots.txt and server logs
• Critical content server-rendered, not JavaScript-only
• First 60 words of every key page rewritten as a liftable answer
• One honest ranked list or comparison published per core topic
• Consistent attributes and ordering inside every comparison table
• FAQ blocks using verbatim prompt phrasings
• Entity name, description and categories consistent across all off-site profiles
• Directory and review listings claimed and corrected
• Re-measurement scheduled at a fixed cadence

Run it with our Agent Readiness Scanner if you want the technical half automated, or talk to us about answer engine optimization if you’d rather it was done for you.

Frequently asked questions

What is LLM in SEO? LLM stands for large language model. In an SEO context it refers to the systems behind ChatGPT, Claude, Gemini and Perplexity that generate a written answer directly rather than returning a list of links.

What is the difference between traditional SEO and LLM SEO? Traditional SEO competes for a ranked position on a results page. LLM SEO competes to be retrieved and quoted inside a generated answer. Ranking earns clicks, LLM SEO earns citations, and in our data the two do not reliably correlate. A page can rank on page one and never be cited, and a directory listing you do not control can be cited constantly.

Is SEO going away with AI? No. Every major assistant retrieves from the live web, and the pages it retrieves are overwhelmingly pages that already rank. Classic SEO has become the qualifying round rather than the finish line. What is going away is the assumption that a ranking automatically converts into a visit.

Which LLM is best for SEO? For visibility, ChatGPT. Across our 5,550 monitored responses it mentioned tracked brands in 20% to 33% of answers, against 14% to 20% for Perplexity and Google AI Overviews, and it issued the most citations by a wide margin. If you are asking which model is best for doing SEO work, that is a different question and depends on the task.

Is ChatGPT an LLM or generative AI? Both. ChatGPT is a generative AI product built on top of a large language model. The LLM is the underlying model, ChatGPT is the assistant wrapped around it, plus retrieval, memory and tools.

How is LLM SEO different from GEO? In practice, very little. GEO, generative engine optimization, is the academic term. AEO, answer engine optimization, emphasises the answer format. LLM SEO is the engineering-flavoured name. All three describe getting cited inside AI-generated answers, and anyone claiming they are fundamentally different disciplines is usually selling something.

What I would do first

If you are starting from nothing, do not begin with content. Begin with measurement.

Write down twenty prompts a real buyer would type, run them through ChatGPT and Perplexity yourself, and write down which brands and which domains come back. That exercise takes an afternoon and it will tell you more than any framework, including mine. Most teams discover two uncomfortable things: they are absent from the questions that matter, and the sources being cited instead of them are often directories and forum threads rather than polished competitor content.

Then pick the single topic where you are most absent and most commercially exposed, and write the one page a model would have to reach for. Not ten pages. One. Measure again in two weeks.

The reason I keep pushing people toward that sequence is that AI visibility work goes wrong in a specific, predictable way. Teams publish volume, feel productive, and cannot tell whether anything moved because they never captured a baseline. The roofing brand in this article gained 92% on its visibility score in five days, and the only reason we can say that with any confidence is that somebody wrote down the number on day one.

Start with the number.