Most B2B content is invisible to AI search. I built a framework to change that.

Most B2B content is invisible to AI search because it was built to rank, and ranking is only half the job now. SparkToro’s 2026 zero-click study found 68% of US Google searches ended without a click in early 2026, and G2’s 2026 survey of 1,076 B2B decision-makers found 51% now start software research in an AI chatbot more often than in Google, up from 29% a year earlier. So when I added AI search optimisation to the content operation at a B2B SaaS company serving 200+ higher education institutions, I baked it into how every piece got briefed and written. That operation contributed to 48% of marketing-sourced pipeline in peak months, though AEO was one ingredient among several and I can’t isolate its share. If I did it again, I’d start with the claims a model can cite and fix the formatting after.

What are AEO and LLMO, and how do they differ from SEO?

AEO (Answer Engine Optimisation) is structuring content so a search engine can lift a direct answer straight out of it, which is how you win featured snippets, People Also Ask boxes and other zero-click results. LLMO (Large Language Model Optimisation) is making your content the source a generative AI tool cites by name when it writes an answer, whether that’s ChatGPT, Claude, Perplexity, Gemini or Google’s AI Overviews. You’ll also see GEO (Generative Engine Optimisation), which is the same job as LLMO under the name the Princeton researchers who first measured it gave it.

The industry’s already muddied the terms, so here’s where I draw the lines:

	SEO	AEO	LLMO
The question it asks	How do I rank on page one?	How do I become the answer on the results page?	How do I become the source the AI cites?
Where you show up	Blue links	Featured snippets, People Also Ask, zero-click results	Answers in ChatGPT, Claude, Perplexity, Gemini and AI Overviews
What earns it	Relevance, links, technical health	Direct answers, standalone definitions, question headings	Claims, data and sources a model can’t get anywhere else
How you know it’s working	Rankings, clicks, organic sessions	Snippet and People Also Ask ownership	Citation rate across a fixed set of buyer prompts

Each column needs its own structure and its own scoreboard, and most B2B content teams are still only working on the first one; AEO and LLMO together are what I mean by AI search optimisation. If you’re deciding which of the three to fund first, I’ve written that call up in SEO vs AEO vs LLMO.

Why isn’t SEO enough for AI search optimisation?

SEO still does its job, but on its own it stopped describing how B2B buyers find vendors. At the B2B SaaS company, organic traffic was growing, blog posts were ranking and keywords were climbing, so by every SEO metric the strategy worked. Then I dug into how prospects were actually finding the product and the picture got murkier: more first conversations started from AI, with prospects turning up with their questions half-answered and quoting things they’d “read somewhere” that turned out to be an AI summary of someone else’s content.

The data has since caught up with that hunch. Pew Research Center tracked the browsing of 900 US adults in March 2025 and found people clicked a traditional search result on 8% of visits when Google showed an AI summary, against 15% when it didn’t, and clicked a link inside the summary on just 1%. B2B buying moved even further: G2, which has its own stake in how buyers research software, found 69% of the decision-makers it surveyed picked a different vendor than they’d planned because of what an AI chatbot told them.

Google doesn’t think any of this needs a new name, and Danny Sullivan, then a director in Google Search, told WordCamp US in August 2025:

Good SEO is good GEO, or AEO, AIO, LLM SEO, or LMNOPO.

He’s right about the foundation, and I’d never tell a team to drop it, but I part ways with him on everything after the ranking: an engine that reads and judges sources before it writes needs pages it can quote cleanly and trust, and a page built only to rank often gives it neither. I’ve made the longer version of that case in why AEO won’t save a weak content strategy, which comes at it from the other side: AEO bolted onto thin content doesn’t work either.

What’s in the AEO/LLMO framework I built?

The AEO/LLMO framework is five working rules for AI search optimisation that I built from zero and applied across blog posts, product pages, whitepapers, landing pages and FAQ sections at the B2B SaaS company, with everything documented. The current, fuller version lives on my AEO/LLMO content optimisation framework page and the line-by-line tactics are in the tactical guide to AI-citable content, so here’s the version I ran, in the order I’d run it now.

Make every claim citable. This is where most B2B content falls over, because a model checks how specific and well-attributed a claim is before it picks a source. “Our platform helps universities improve admissions” gives it nothing to quote, while “Institutions using the platform cut average admissions processing time by 40% across 12 institutions between 2023 and 2024” does (that one’s illustrative, so please don’t quote it). I set up a verification process where I traced every stat and timeline, plus any comparative claim, back to help documentation, internal project material or a stakeholder who’d confirm it. The editorial discipline behind that is in why writing quality is a competitive moat in B2B.

Define things at first use. An AI system has to know what something is before it can use it as an answer, so anything that introduces a concept or product gets a standalone definition near the top, saying what it is and who it’s for in a sentence that still makes sense pulled out of the page. Most B2B SaaS content buries its definitions under three paragraphs of preamble, and AI systems don’t have patience for that (frankly, neither do your readers).

Then structure for extraction, since a model lifts a section at a time and every section has to survive being the only one it reads. In practice that meant:

• Subheadings that say what’s coming, with no clever wordplay
• One complete idea per paragraph
• FAQ answers that repeat enough of their question’s context to work without the article around them, which made each pair citable on its own
• A summary at the top of every long piece compressing the core argument into a few sentences

Build clusters, because models judge domains. AI systems look at whether your site knows a topic, which means depth across connected pages. I mapped clusters to buyer personas and buying stages so each one did double duty, giving SEO its keyword coverage and AEO/LLMO its topical authority. A cluster for university IT leaders looked like this:

• A whitepaper on enterprise software selection as the pillar
• Blog posts on specific evaluation criteria (security, integrations, scalability)
• FAQ pages on integration requirements and technical specs
• Case study content on implementation outcomes at comparable institutions

Each piece stood up alone, and together they gave a model a body of evidence it could recognise as authoritative.

Last, track AI visibility as its own metric. Most analytics setups weren’t built to see AI-driven discovery, so I used Amplitude to build dashboards for referral patterns from AI tools and shifts in zero-click behaviour, and to check whether changes in structure lined up with changes in where traffic came from. The attribution was rough, but it still gave the company a baseline nobody had before, and it’s the part I’ve changed most since.

What happened when I deployed the framework?

Across all content types at the B2B SaaS company, organic traffic grew 33% quarter on quarter and blog content beat its targets by 12% in the first two weeks. The content operation contributed to 48% of marketing-sourced pipeline in peak months and influenced €230K+ in pipeline in a single reporting period, and I’ve broken down where that 48% came from separately.

Were those results purely down to AEO/LLMO? No: they came from a full-funnel strategy for five buyer personas, a content strategy rebuilt for the company’s pivot from admissions-led to IT-first positioning, a LinkedIn programme pulling 2,900+ interactions a quarter, and editorial standards applied ruthlessly across every channel. The multi-persona strategy and the ICP pivot each get their own write-up because they’re their own stories.

What AEO/LLMO did was tie the rest together: it changed how I structured blog posts and product pages, right down to how I formatted FAQ sections and whitepapers, and my definition of good content moved with it.

I’ve since run the same framework a second time, at OpinionX, where I’m Principal Content Strategist: I audited and updated the whole blog library for AEO/LLMO, and average Google search position went from 18.4 to 8.8 in five weeks, with search CTR up 54% on the Q3 average. The same caveat applies, since a library-wide refresh fixes more than AEO, but the framework moving two very different B2B companies the same way is a stronger signal than one.

What would I do differently with the framework?

I’d flip the order: when I first published this framework it led with definitions and extraction structure, and I’ve since rebuilt it so substance comes first, meaning the claims, data and named sources a model can’t find anywhere else, with the formatting that delivers them after. The research backs the change: the Princeton team behind the GEO study tested optimisation methods across 10,000 queries and found that adding statistics, quotations and source citations lifted visibility in AI answers by up to 40%, while keyword stuffing did worse than doing nothing at all.

Formatting is the cheap, checkable end of AI search optimisation, which is why teams start there, and also why plenty of thin content passes the audit and still gets zero citations. I’ve written up the AEO implementation mistakes that come from running it backwards.

I’d track citations from day one. My Amplitude dashboards showed me referral traffic, which is the smallest slice of AI search, since most of the value sits in answers nobody clicks through. At OpinionX I set up citation tracking instead, running 20 buyer prompts across ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews, and the first baseline had the brand named in 38% of answers and cited as a source in 37%. That’s the scoreboard I’d want from the start, and the method is in how I track AI citations and my prompt research process.

I’d stop leaning on conversion multiples. I originally quoted a 5.1x conversion premium for AI search traffic, and the only multi-site study I’ve found that tested for significance, Amsive’s analysis of 54 sites, saw no significant overall difference between AI and organic traffic. Its B2B sites did see AI referrals convert at 2.17% against 1.16% for organic, though even that gap didn’t pass the test on the stricter sample. I’ve gone through the rest of that evidence in tracking AI traffic in GA4.

And I’d be blunter about attribution from the start: I still can’t tell you what share of that 48% AEO earned, and nobody running AEO inside a working content strategy can, so anyone selling you a clean AEO-only uplift either ran it in a lab or wants your budget.

Where should a B2B team start with AI search optimisation?

Start with an audit of your ten best-performing blog posts before you write anything new. For each one, ask:

• Does it make at least one claim a model can’t get elsewhere, backed by your own data or a named source?
• Is there a standalone definition in the first few paragraphs?
• Would each section make sense if it were the only part a model read?
• Does every statistic carry a visible link to where it came from?

Every “no” goes on your to-do list. If you want a scored version, my AI readiness audit framework gives you one, and the five checks I run on a content library go through how to fix what it finds.

Then put AI search optimisation inside the workflow you already run, so it lives in your style guide and editorial checklist, and in every brief you write, because a separate workstream is the fastest way to get it cut at the next planning cycle.

Measure it, even badly: pick a fixed set of buyer prompts, record a baseline before you change anything and re-run it monthly, so you’re looking at a trend by the time someone asks whether the work pays. If you want the traffic side too, you can give AI referrals their own channel in GA4 with a custom channel group.