LLM Optimization: How to Get Your Brand Into AI Answers (Practical Guide)

A customer asks ChatGPT which agency they should trust, or which product solves their problem, and a name comes back. Maybe it's yours. Maybe it's a competitor you've never even heard of. Either way, that recommendation happened without a single click to anybody's website.

That's the bit that should keep you up at night. The decision was made inside the answer. LLM optimization is the work of getting your brand to be the name that comes back, and after running AI visibility campaigns across dozens of client sites, we've got opinions about what actually moves the needle and what's a waste of a Tuesday.

This isn't a future problem we're flagging to look clever. The behaviour has already shifted. Let's get into how to play it.

What is LLM optimization?

LLM optimization is the practice of making your brand and content more likely to be surfaced and cited when large language models answer a user's question. People also call it GEO (generative engine optimisation) or AEO (answer engine optimisation). The labels are multiplying faster than the tactics, frankly.

It covers the AI surfaces your customers actually use: ChatGPT, Google's AI Overviews and AI Mode, Perplexity, Gemini, and Claude. The goal is simple to say and harder to do. When one of those tools forms an answer in your category, your brand should be part of it.

Here's the most useful mental model we give clients. Traditional SEO is about ranking a page. LLM optimization is about teaching the wider web what your brand is, what it's good at, and why it deserves a mention. Those are related jobs. They are not the same job, and the data backs that up.

Why this matters now, not in three years

The clearest evidence comes from Pew Research Center, which tracked the real browsing behaviour of 900 US adults across nearly 69,000 Google searches in March 2025. When an AI summary appeared, people clicked a traditional search result only 8% of the time, compared with 15% when there was no summary. That's nearly half the clicks, gone.

It gets worse for the "they'll click the citation" hope. Pew found users clicked a source link inside the AI summary in just 1% of visits. One percent. And people were more likely to end their session entirely after a page with an AI summary, 26% of the time versus 16% on a standard results page.

Around 18% of the Google searches in that study produced an AI summary. Search Engine Land's coverage of the same study noted summaries showed up on 53% of long ten-word-plus searches, the exact detailed questions where buyers are closest to a decision.

So the clicks are shrinking, and the answer itself is doing the persuading. If you're not in the answer, you're not in the conversation. We dig into the brutal click maths in our breakdown of how AI Overviews are changing click-through rates, and it's not a cheerful read.

How do LLMs decide what to cite?

Nobody outside the labs has the full recipe, and anyone who tells you they do is selling something. But we can read the data, and a few patterns are consistent enough to plan around.

Ranking number one does not automatically get you cited

This is the myth we spend the most time killing. The old assumption was that AI just summarises the top organic results, so win SEO and you've won AI. The evidence says otherwise.

Across the tracking period in one widely cited analysis covered by Search Engine Land, only around 17% of sources cited in AI Overviews also ranked in the organic top 10. Roughly five out of six citations came from pages that weren't even on page one. Ranking and getting cited are connected, but they're genuinely different games, which is exactly why we treat AI search visibility as its own discipline rather than a bolt-on.

Fan-out queries are the real unlock

Here's the finding that changed how we structure content. Google's AI doesn't run one search. It uses a "query fan-out" technique, firing off multiple related sub-searches to build an answer. Google's own documentation confirms this, describing it as "issuing multiple related searches across subtopics and data sources."

Why care? A Surfer SEO study reported by Search Engine Land, covering 10,000 keywords and over 173,000 URLs, found that pages ranking for fan-out queries were 161% more likely to be cited than pages ranking only for the main query. Pages that ranked for both the head term and at least one fan-out accounted for 51% of all AI Overview citations.

The takeaway is concrete. Don't just cover the obvious keyword. Cover all the questions branching off it, the comparisons, the edge cases, the "but what about" follow-ups. That's the fan-out surface, and that's where citations live.

Off-site mentions carry more weight than on traditional SEO

LLMs form their picture of your brand from the whole web, not just your site. What Wikipedia says about you, what gets discussed on Reddit, what trade publications reference, all of it feeds the model's understanding of who you are and what you're known for.

The Pew data showed this clearly even inside Google's summaries: Wikipedia, YouTube and Reddit together made up about 15% of the sources listed. We unpack why a Wikipedia presence influences LLM citations in more detail, because it's one of the highest-leverage and most underrated moves a brand can make.

What we've actually seen work

Enough theory. Across our campaigns, these are the levers that have produced real citation gains, ranked roughly by impact.

1. Be the most complete source on the topic, not the longest. LLMs favour content that answers the full question, including the sub-questions. We build pages that close every loop a reader could open. Padding doesn't help. Completeness does.
2. Map and cover the fan-out. For every target topic, we list the related searches and "people also ask" branches, then make sure the page answers them with clear sub-headings. This is the single most direct response to that 161% finding above.
3. Front-load a clean answer. A direct one or two sentence answer right after a heading is far easier for a model to lift than a buried gem in paragraph nine. Our analysis of how ChatGPT favours the first 500 words made us ruthless about this.
4. Publish original data. Models love a number they can't find anywhere else. First-party stats, survey results and named benchmarks make you uniquely citable, which is why we keep producing our own AI search visibility research rather than recycling everyone else's.
5. Build genuine entity presence. Earned mentions on trusted sites, accurate profiles on the platforms LLMs lean on, and a coherent story about what your brand stands for. We treat this as entity and knowledge graph work, and it compounds.
6. Keep it fresh and dated. A visible "last updated" and current data tells both Google and the models the page is alive. Stale 2023 content with no refresh gets quietly passed over.

Notice none of that involves a secret AI trick. It's good content and real authority, structured for extraction. That's the honest version of LLM optimization, and it's the version that lasts.

Where do citations actually come from across the platforms?

The split between your own site and everyone else's is the thing most people get wrong. They pour everything into on-page work and ignore the wider web that the models are reading.

A Semrush study analysing over 100 million AI citations across more than 230,000 prompts between July and October 2025 found Reddit and Wikipedia sitting at the top of the citation pile across ChatGPT, Google AI Mode and Perplexity. It also caught something telling: in mid-September 2025, ChatGPT's reliance on Wikipedia dropped from roughly 55% of responses to under 20%, while sources like Forbes and PR Newswire climbed.

Two lessons from that. First, community and reference sites matter enormously, so being talked about in the right places is part of the job. Second, the models change their citation behaviour on a whim, which is why we never let a client's AI visibility hinge on a single platform or a single source type.

What does not work (and what we've stopped doing)

We've burned time on tactics that sounded smart and did nothing. Save yourself the trouble.

• Stuffing "AI recommends" or "as cited by ChatGPT" into your copy. Models don't reward this. They might even distrust it. It just makes your page read like a hostage note.
• Writing "for the training data." You can't reverse-engineer what gets into a training set. Chasing it is a fool's errand. Write for humans, structure for machines.
• Obsessing over a magic file or markup. Plenty of people swear an llms.txt file or some clever schema is the cheat code. Google is blunt about this: there are "no additional requirements to appear in AI Overviews or AI Mode," and "you don't need to create new machine readable files, AI text files, or markup to appear in these features." We tested the file route and shared the underwhelming results in our llms.txt and AI citations data piece.
• Ignoring SEO fundamentals entirely. Yes, ranking and citation are different games. But a page has to be indexed and eligible to appear in Search at all before it can be cited. Solid technical and content SEO is the price of entry, not the whole strategy.

How do you measure LLM visibility?

There's no Search Console for ChatGPT yet, which means measurement is scrappier than we'd like. Here's the stack we use to get directional, honest data.

1. Prompt testing. We run a fixed set of buyer questions through ChatGPT, Perplexity, Gemini and Google AI Mode on a schedule, and log whether the brand appears and how it's described. Same prompts, same cadence, so the trend is real.
2. Referral tracking in GA4. AI tools increasingly pass referral traffic. Filtering for sources like chatgpt.com and perplexity.ai shows whether citations are turning into actual visits.
3. Brand mention monitoring. Alerts for the brand name across the web, because off-site mentions are what feed the models in the first place.
4. AI Overview citation tracking at scale. Tools that monitor which pages get cited in AI Overviews for your target queries, so you can see patterns rather than anecdotes.

It's manual and imperfect. Anyone claiming a clean, complete dashboard for this is overselling. But run it consistently and you'll know whether the work is paying off, which beats guessing.

Our honest take on where this goes

We think LLM optimization becomes as central as traditional SEO within a couple of years, and the brands that win it are the ones building real authority right now. The behaviour change is already here. The Pew numbers on collapsing clicks aren't a forecast, they're a measurement.

But here's the unglamorous truth we keep coming back to. The best LLM optimization strategy looks a lot like the best SEO strategy with the volume turned up: comprehensive content that covers the full fan-out, a clean structure a model can lift answers from, original data nobody else has, and genuine brand presence across the web the AI is reading. Semrush's analysis of AI Overview behaviour reinforces it, with most summaries triggering on informational queries and pulling from a wider, more diverse set of sources than the old top-ten ever did.

This isn't a brand-new discipline that throws out everything you know. It's an extension of doing the fundamentals properly, then pushing them harder than your competitors are willing to. If you'd rather not work out the fan-out maps and citation tracking yourself, that's literally our AI search visibility service, and we're happy to show you what we're seeing in real campaigns. Tell us your category and we'll tell you straight whether you're showing up in the answers or watching a competitor do it.

For a tactical companion to this, our guide on getting cited in ChatGPT and AI Overviews walks through the on-page moves in more detail. The work is doable. The window to be early on it is closing.