GEO for B2B 101
TL;DR
I get it, you’re busy. You can copy this URL into your LLM of choice and have it summarised if you want. Ask it which bits are relevant to you, I won’t be offended. But here’s the main things to remember for yourself:
GEO is important. It’s changing discoverability and validation in B2B buying, but you knew that, it’s why you’re here.
GEO isn’t SEO. They are cousins, not twins. The approach isn’t the same.
B2B GEO is not consumer GEO. Most research, reports, and fellow agencies banging on about GEO do it through a consumer lens. B2B tech has different influences. We’re not selling hairdryers.
Our practical framework is Clarity; Consistency; Consensus. Define what you are, reinforce it, get third parties to validate it. But…
There is no quick fix formula. Being surfaced by LLMs in search results is the end result of a good, fully-functioning comms plan. Sometimes it’s easier depending on the specific search you want to appear in, but there’s no 1+1=2 plan.
A quick note before we start…
There’s a lot of noise out there about GEO. Some of it is good-faith education and experimentation. I have learned from what other agencies are doing because some are doing good stuff. Unfortunately some of it is people selling “5 easy steps” and “quick fixes”. Even more of it is framed for a consumer brand and explains GEO in the context of how people buy hairdryers.
The truth is, nobody has this totally figured out because it’s new, changing all the time, and the platforms themselves don’t fully pull back the curtain on how they work. So, this guide isn’t trying to be the fountain of all knowledge. It’s trying to be useful — especially for B2B tech, where the mechanics of trust are different from the mechanics of consumer popularity.
What is GEO?
Generative Engine Optimisation (GEO) is the practice of improving how visible, and how accurately represented your content (and by extension, your brand) is inside answers produced by generative large language model (LLM) systems like ChatGPT, Gemini, Copilot, Claude or Perplexity.
There has been a bit of a sprint to label it, so you might hear AEO, LLMO (my favourite), or simply AI Search. They describe overlapping territory rather than perfectly interchangeable disciplines, but GEO seems to have won the race for the name after being formalised in academic research authored by contributors from Princeton University, Georgia Tech, and the Allen Institute for AI (among others).
Is GEO the same as SEO?
No, but you’re not the first to ask. This has been a common question in the formative days of GEO, not least because some unsophisticated SEO agencies are desperate to hitch themselves to the shiny new wagon. The truth is they might be cousins, but they are not identical twins.
GEO and SEO solve adjacent problems, but they do it in different ways. SEO optimises for retrieval and is about being returned as a result when someone searches. GEO optimises for synthesis, which means being included, described, and sometimes cited when an engine generates a single answer by combining information from multiple sources.
As a useful shorthand: SEO is often about winning the click. GEO is about winning a place in the answer. The playbooks overlap, but they are not interchangeable.
Good SEO hygiene still forms the foundation of good GEO performance. Generative systems tend to reward content that is easy to extract, easy to interpret, and directly responsive to a question. That can be supported by SEO-style best practices like clear headings, direct language and descriptive subheads. The same Princeton et al report that named GEO as a ‘thing’ also outlined how straightforward textual enhancements on a website can increase measured visibility by up to 40% in generative engine responses, with effectiveness varying by domain.
However, the easiest evidence that they are not the same thing comes from a report by Ahrefs, which analysed 15,000 prompts and compared the URLs cited by AI assistants against Google’s top results for the same prompts. On average, only 12% of the links cited by ChatGPT, Gemini, and Copilot appeared in Google’s top 10 results for that prompt. In other words, ranking well in Google is not a reliable proxy for being cited or referenced in AI answers.
Authority remains a real concept across both disciplines, but it presents differently. SEO historically treated authority as a link-weighted signal on domains and pages, which is why in my early PR days clients were desperate to get backlinks into articles. In GEO, authority behaves more like a trust-weighted signal across an ecosystem. This means repeated, consistent, independent confirmation across sources the model retrieves and feels safe repeating. This is one reason communications and PR strategy suddenly matters in what used to feel like “search territory.”
That last point is especially important for B2B tech. Optimisation is not universal, because the sources retrieved and weighted change with the type of question and the type of buyer, as well as which LLM the searcher is using (more on that later).
A second practical difference is measurement. SEO gives clean outputs; rank, impressions, clicks. GEO gives messy outputs; mentions, citations, narrative accuracy, and inconsistency across engines.
In short, SEO and GEO are related, but they reward different behaviours and they produce different outcomes. SEO remains a foundation. GEO is the next layer, where being understood, validated and repeated across the right sources increasingly determines whether a brand appears in the answer at the moment a buyer is trying to make a decision.
How does GEO work?
Okay, it’s not the same as SEO, how do these large language models (LLMs) produce answers?
They are called “generative engines” because they don’t just retrieve a page and hand it back. They generate an answer from the model’s existing knowledge and, in many AI search experiences, information retrieved from external sources. One common way of combining the two is Retrieval Augmented Generation (RAG).
RAG brings retrieved external information into the model’s context before it generates an answer. That evidence might come from the live web, a search index, a connected database or a company knowledge source, depending on the product. When authoritative information is accessible, well-structured and consistently stated across the sources the system retrieves, it becomes easier for the model to include it accurately. When high-quality information is absent or inaccessible, the system has fewer reliable inputs and can omit, blur or misstate the answer.
RAG is widely used because it allows outputs to incorporate newer, domain-specific or otherwise external information without retraining the underlying model each time the world changes. It is also one reason fresh, retrievable information can matter disproportionately for niche technical questions.
I like to think of it like this. Traditional search behaves like a library. You search something and the engine’s job was to retrieve something useful off the shelf, largely based on keywords. Generative systems are more like a brain, you ask it a question and it thinks about what it already knows (training data) and what it needs to read on the internet to then generate a concise answer.
Why does GEO matter in B2B comms?
In B2B, buyers are using LLMs to compress research and validate their decisions. Forrester’s 2024 research reported that 89% of B2B buyers were using generative AI in at least one part of the purchasing process. More recent G2 research from 2026, focused specifically on more than 1,000 B2B software buyers and decision-makers, found 71% rely on AI chatbots somewhere in software research and 51% start software research with AI more often than Google. G2 also found AI chatbots were the largest reported influence on vendor shortlists, at 54%.
That doesn’t mean every B2B buyer is outsourcing discovery to an LLM, or that being absent from one answer makes you invisible. It does mean a growing share of research, comparison and validation can happen before a prospect lands on your website or speaks to your sales team. Bit of an uphill battle if the answer has already narrowed the field without you in it.
A nuance worth flagging is that B2B buyers often use AI less for pure discovery than consumers do. They often have a list in mind already, and use LLMs to compress complexity and to validate a decision they are already moving toward. Forrester’s own description of the 2024 survey result makes this explicit, arguing buyers were not only using GenAI to discover vendors, but also to evaluate differences and justify the purchase commitment.
Validation questions are psychologically different to discovery questions. A discovery prompt is exploratory. A validation prompt is someone trying to defend their decision. It’s the classic, “nobody got fired for buying IBM” scenario. That makes the output feel authoritative by default, even when it shouldn’t. Research on automation bias consistently shows that humans can over-rely on automated recommendations, particularly when they appear competent or confident, and that users’ understanding of AI does not reliably protect them from miscalibrated trust, with more confident AI users actually trusting the outputs more/interrogating them less than a novice.
B2B decisions are heavily influenced by repeated claims across trusted environments (analysts, trade press, vendor explainers, peer commentary etc etc). Humans are wired to treat familiarity as a cue for truth, and generative systems do something similar at scale. They synthesise from patterns in what they retrieve, and repetition across credible sources becomes a confidence signal for the answer they produce.
This means the earned media consensus to validate where a prospect is already leaning is crucial in B2B buying.
Being cited isn’t the same as being recommended
There’s another distinction worth making. A site can be used as evidence without the brand itself making the answer. A June 2026 Semrush study found that around 62% of AI citations were “ghost citations”, meaning the source was cited, but the brand was not actually named in the generated answer.
For GEO measurement, separate three questions: did the engine cite you, did it name you, and did it actually recommend or shortlist you? Those are not the same thing, and the latter two are usually much closer to the outcome a B2B marketing team cares about.
Where do LLMs pull from to form their answers?
One of the clearest large-scale data points on citation behaviour comes from Muck Rack’s “What Is AI Reading?” research. In the May 2026 edition, analysing 25M+ links across ChatGPT, Claude and Gemini, earned media accounts for 84% of AI citations. Journalism alone accounts for 27%.
I can hear you, surprise surprise the PR people are telling me I need media coverage. Well, we kind of are, but we’re trying to be fair about it, so let me explain more.
Firstly, those percentages are averages across many industries and prompt types. The same Muck Rack write-up is clear that query type materially changes what gets cited. For example, “industry trend” questions drive journalism citations at more than double the rate of “how-to” questions. They also note that for recent queries, journalism can make up a much larger share of citations.
A lot of GEO commentary defaults to consumer behaviour because consumer prompts are easy to imagine, easy to test, and easy to generalise. “Best headphones”, “best hairdryer”, “is this product worth it”, “give me a plan for this holiday”. That prompt shape tends to reward broad consensus sources (large forums like Reddit – though less so since August -, generalist listicles like those “ten best” articles, review aggregators) and brands with mass visibility. B2B tech behaves differently because the questions are different, and the pool of credible sources is narrower.
In B2B tech, the most influential sources are the specialist titles, analyst-adjacent commentary (when not gated), standards bodies, and technical sources that get retrieved for niche prompts. The “trust set” gets narrower as the prompt gets more specialist.
That means authoritative vendor explainers can also carry disproportionate weight. Consumer GEO narratives often underplay the role of brand-owned content because self-published claims look biased and because many consumer prompts are better served by lived experience and comparative reviews. In B2B tech, the equation shifts because many categories are complex, technical, and jargon-heavy, and the web’s clearest explanations are frequently published by vendors, standards bodies, or specialist technical organisations. When a generative system is trying to answer an explanatory prompt (“what is X”, “how does Y work”, “what are the trade-offs”), authoritative technical explainers can become highly retrievable inputs.
This does not mean “owned beats earned” in B2B. It means owned content can be a primary source of clarity and can act as a verification layer, particularly when third-party coverage isn’t present. Consistently publishing informative content as part of your content marketing strategy can increase your chances of being cited by LLMs.
The source mix can also move quickly. In Promptwatch’s July 2026 ChatGPT Search data, product pages accounted for 32.8% of classified citations, up from roughly 18% in March. That looks very different from broader cross-engine studies where earned media dominates. The useful conclusion is not that one study is right and another is wrong, but that engine, prompt type, date and source classification all change the picture. Don’t build a strategy from one thing, and also don’t ignore the role of your own blog and newsroom pages!
Do all LLMs behave the same?
Sadly not, that would be too easy…
The broad principles we have discussed are the same across all and should form the basis of your strategy, but there are some differences worth talking about.
A Yext study from 2025 outlined how Gemini takes more than half its references from brand-owned websites, which makes sense considering it’s owned by Google so has that SEO foundation baked in. Own the owned like you would for Google and Gemini will still support you. It also favours content published on YouTube, which shock horror, is owned by Google. Publishing some educational videos – crucially with the transcript in the description – can really help on Gemini.
Perplexity is designed around live research and often leans heavily on current external sources. ChatGPT’s source mix can change much more quickly than most GEO playbooks admit. In August 2026, Promptwatch measured Reddit’s share of ChatGPT Search citations falling from 3.83% to 0.52% in a matter of days, alongside a sharp increase in site:-restricted searches. YouTube tells a similar cross-engine story: Promptwatch’s February data found it relatively important to Perplexity and Google AI Overviews but almost invisible in ChatGPT at that point. The exact numbers will keep moving. The lesson is that “get on Reddit” or “make YouTube videos” is not a GEO strategy on its own. Platform-specific source preferences are clues, not foundations.
ChatGPT is still the market leader in general so the temptation is to optimise around how it behaves today. But enterprise use is spread across ChatGPT, Gemini, Claude, Copilot and Perplexity depending on a company’s stack and preferences. You need to think across engines, and success in GEO will usually be the result of a mix of owned, earned and social communications rather than one platform hack.
And increasingly, two users may not get the same answer
There is another complication with GEO that is easy to overlook. AI assistants are becoming personalised. Depending on the product, settings and permissions, ChatGPT can use relevant past chats, memory, Library files and connected-app content, while Gemini can personalise from past chats and connected Google apps. In workplace settings, AI can also be grounded in internal company knowledge from sources such as Slack, Google Drive, SharePoint and other connected systems.
That becomes particularly interesting in B2B. A buyer asking an AI assistant to recommend vendors may already have months of context sitting behind the question including companies mentioned in internal documents, suppliers discussed in emails, vendors they have researched before, analyst material they have uploaded, or assumptions established through earlier conversations.
So, the answer is not always a clean-room ranking of the public internet. In some cases AI may reinforce what has already influenced the buyer. Good PR, analyst relations, events, customer advocacy and sales activity can therefore feed the context that later shapes an AI-assisted decision, even if none of those things receives a neat clickable citation. GEO matters, but it sits on top of the wider B2B buying journey rather than replacing it.
Think in topics, not single prompts
One screenshot of one favourable answer does not mean you’ve got it sussed. A 2026 Semrush study of more than 50,000 brands across 1,094 ChatGPT topic clusters found only 15.2% of topics had a clear owner. Their definition of ownership required a brand to appear across at least four of five related prompts, not simply win one carefully chosen question.
That is a useful way to think about B2B GEO. Buyers ask variations of the same underlying question. Think how many different ways you can word ‘how does A compare with B’ just as one example. Consistency across that cluster matters more than getting excited because you appeared once.
Do AI systems trust sponsored content?
Well, do you?
The same MuckRack study mentioned above shows earned media dominates citation share across major models, with paid media accounting for around 0.3%.
However, that shouldn’t be taken as gospel because as with most research, it focussed on broad or consumer-centric searches where ads are much more overt. In B2B, the influence set is often smaller and more specialised.
If you are going to invest in sponsored media coverage, don’t assume an LLM will treat it exactly as a human does. Sponsored content can still be retrieved if it is useful, accessible and relevant, but the citation data we have today suggests paid and advertorial sources are a small part of what major AI systems cite.
If sponsored content gives a genuinely useful answer in a thin information environment, it can still contribute. But treat it as supporting evidence, not as a shortcut to AI visibility.
What about newswires?
Good question. The wires themselves will tell you they are a great way to get into all the major news outlets inboxes, but as a former journalist I can tell you that is codswallop. Most journos have an automatic rule for wires to filter into a folder in their inbox, and because they get so many, it is never opened.
Therefore the general advice from us in the PR industry has generally been if you do not have a clear regulatory reason to use a wire, save your money.
However, there is some thinking that GEO is changing that dynamic. Muck Rack’s May 2026 summary notes that press releases appear almost exclusively in industry trend responses and at 3.5 times the rate they appear in “best-of” queries. At the same time, press releases and wire domains are a tiny slice of overall AI citations.
A BuzzStream analysis using Citation Labs’ XOFU tool tracked 4M AI citations across ChatGPT and Google AI surfaces. In that dataset, syndicated press releases accounted for 0.04% of all citations, and direct wire URLs (e.g., PRNewswire) accounted for 0.21% of the total dataset.
The practical conclusion is not “always wire” or “never wire”. It is in between.
A wire can be worth considering when at least one of the following is true:
The announcement is genuinely time-sensitive, and the likely prompts are “trend-shaped” (“what happened with…”, “latest on…”, “has vendor X launched…”) rather than “best” shaped.
The goal is to create a durable canonical record that other outlets can pick up and that a model can retrieve as a factual reference point, but your own blog can achieve the same thing.
A wire is still usually a waste of money when it is used as a substitute for analytical coverage or as a substitute for a proper newsroom layer on the company site. The citation data suggests that original editorial content dominates “news” citations inside AI answers, while syndicated and wire content barely registers.
So, how do I get into the AI answers?
Okay let’s summarise what we have learned. How to surface in an AI-generated answer will obviously depend on the specific question you are trying to be the answer to and the platform you use. There is no singular ‘right’ answer, but we believe GEO is the outcome of a solid comms plan, that focuses on three pillars; Clarity, Consistency, Consensus.
Clarity means getting your house in order, having clear messaging about what you are, what you do, and what questions you have the answers to. It also means clearing up all the SEO foundations on your website and making sure the LLMs can crawl you easily and find the information they need by un-gating the content you want it to know about. Invisibility can become misinformation very quickly if the AI cannot read all the content. It doesn’t skip, it plugs the gaps by guessing.
Consistency means, well, be consistent. Regularly post informative content that reinforces your position, and keep your messages consistent across owned, earned and paid platforms.
Consensus means having your message repeated and validated by the wider ecosystem. We are using Consensus and not Coverage because it’s not just about media mentions or analyst reports. Consensus in B2B can often come from lots of places.
GEO isn’t a trick. It’s the byproduct of being easy to understand, consistently described, and credibly validated across the sources AI systems see as valuable. For B2B tech, that’s rarely the whole internet. It’s a tight influence set, and increasingly the public web is only part of the context an AI assistant may be working from.
If you want to stop guessing, do the simplest useful thing - map what the engines currently say about you, who they cite, and where the gaps are. Then fix the gaps with clarity, consistency, and consensus.
If you’ve had a play with some prompts and don’t like what you saw, don’t go hunting for a GEO hack. Start with what the answers are telling you.
If an LLM doesn’t understand what your company does, that’s a Clarity problem.
If the right information exists but only in one place, or you aren’t consistently associated with the subjects you want to own, that’s a Consistency problem.
If your own story is clear but competitors keep making the shortlist ahead of you, look at Consensus. What evidence exists elsewhere that makes them easier to trust, recommend and repeat?
Sometimes the answer will be better website content. Sometimes it will be media coverage, analyst relations, customer proof, industry partnerships, thought leadership or simply making better use of the expertise already sitting inside your business. Usually it will be a mix.
And that is probably the main thing to take away from all of this. GEO is not a new marketing channel sitting off on its own. It is increasingly one of the places where the cumulative effect of everything else you do becomes visible.
The engines will change. The sources they favour will change. The acronym might even change. Being clear about what you do, consistently demonstrating what you know and having credible people elsewhere validate it is a rather safer bet.
Not sure what the AI engines currently think about your business? Request an AI Visibility Snapshot.
Or talk to us about GEO, PR and communications.
