LLM SEO for B2B SaaS: what actually gets you cited

There is a particular kind of frustration I have watched B2B SaaS founders go through in the last eighteen months.

The rankings are fine. Position three, position four, the same places they have held for a year. The technical audit comes back clean. And the traffic keeps falling.

The first instinct is that something is broken. So you check again. You rewrite a few pages. You wonder whether the agency is any good.

Nothing is broken. The question is being answered above you, by a summary that quoted two other companies, and nobody told you the rules had changed.

That is what happens in LLM SEO.

What is LLM SEO?

LLM SEO is the practice of getting a brand cited inside AI-generated answers rather than ranked on a search results page.

For B2B SaaS this matters more than for most industries, because your buyers research in exactly the way AI answers serve. They ask comparative questions. They want to know what breaks, what it costs, who else uses it. Those are questions a model answers in one response, naming two or three vendors, and you are either one of them or you are not in the conversation.

How B2B SaaS buyers actually use AI now

Your buyer’s research has moved, and it moves in three stages.

1. Problem identification. They describe a symptom rather than search a category. “What are the signs we need better project management software.” The model names the problem and the solution categories, and you are not in that conversation yet.
2. Solution education. Feature comparisons, implementation considerations, rough costs. They get detailed answers without visiting a single vendor site. This is where most of your website’s job used to happen.
3. Vendor evaluation. Shortlists, alternatives, who suits which situation. The brands cited here get the mindshare, and the ones that are not cited are not considered.

By the time a B2B SaaS buyer reaches your site, two of those three stages are done.

You will also see this called GEO or AEO, generative engine optimization or answer engine optimization. Same discipline. I use LLM SEO because it names the thing doing the deciding.

What actually gets a B2B SaaS brand cited in AI answers?

This is the question I wanted answered, and for most of 2025 nobody had a real answer. Everyone had theories.

Then Ahrefs did the obvious thing and measured it. They took 75,000 brands and ran correlations between different signals and AI search visibility. These were the four strongest:

Signal	Correlation
YouTube mentions	0.737
Branded web mentions	0.664
Branded anchor text	0.527
Brand search volume	0.392

Look at that list again. Not one of them happens on your website.

I will be honest about why this was uncomfortable to read. Most of what agencies sell, including a lot of what we have sold, is work on your own domain. Content, structure, technical fixes. That work is not worthless. It is just not what the data says moves the needle most.

The thing that moves the needle is being talked about somewhere else.

Do you still need to rank in Google’s top ten?

It helps, although it does not carry you the way it used to.

Ahrefs looked at 863,000 search results and around 4 million cited URLs in March 2026. 37.10% of cited pages ranked in Google’s top ten. Another 36.70% did not rank in the top 100 at all.

Roughly a third from the top ten, roughly a third from nowhere near it.

And if your buyers are using ChatGPT rather than Google, which for B2B SaaS evaluation they increasingly are, the link is weaker still. A separate Ahrefs study of 15,000 long-tail queries found only about 12% of AI assistant citations rank in Google’s top ten.

One number you will see everywhere that I do not fully trust

You will read that top-ten citations “collapsed from 76% to 38%”. I nearly put that in this post as a headline, and I am glad I checked.

The two studies did not measure the same thing. The 2025 figure counted only the three most visible citations in each AI Overview. The 2026 figure counted every citation Ahrefs could parse, after they improved their parsing.

The direction is real. BrightEdge, measuring independently, puts the overlap nearer 17%. But the clean before-and-after story is tidier than the evidence, and I would rather tell you that than repeat it.

Why is your organic traffic falling when your rankings held?

Because the click is disappearing, not the ranking.

68.01% of US Google searches ended without a click in the first four months of 2026, up from 60.45% two years earlier. When an AI Overview appears, click-through on the top organic result drops by roughly 60%.

So you can be indexed, ranked, technically healthy and completely invisible to the person deciding.

We went deeper on what decides AI Overview citations specifically in the AI Overview post, and on the wider click collapse in the zero-click piece.

Is AI traffic worth having when the volume is so small?

This is the fair objection, and I want to answer it properly rather than hand-wave.

The volume is small. Microsoft Clarity data from November 2025 put AI-referred traffic at under 1% of total site traffic. Anyone telling you this replaces your acquisition is selling something.

But the people who do arrive are different. Similarweb measured ChatGPT referrals converting at 11.4% against 5.3% for organic search. They arrive having already done their research inside the conversation, which for a B2B SaaS buyer means they turn up knowing what you do, roughly what it costs, and who else they are considering.

Twice the conversion rate on under 1% of visitors is a real thing and a small thing. The reason I still think it matters: that 1% is growing, and the buyers inside it are the ones furthest along in their evaluation.

Being absent does not cost you traffic. It costs you the shortlist.

Put another way: the brands cited most often in AI answers get the mindshare, and mindshare is what decides who gets considered before anyone types a search at all.

How do you get your B2B SaaS cited in AI answers?

Given the data above, the work sorts itself into an order.

1. Get mentioned off your own domain

If the top four signals all happen elsewhere, this is the job, not an extra.

Review sites, comparison pages, podcasts, communities, other people’s writing. For B2B SaaS specifically, G2 and Capterra pages, integration directories of the tools you plug into, and roundups in your category.

YouTube is the one I would push hardest on, and the one most B2B SaaS teams ignore. It tops the correlation table, and among AI Overview citations from pages not ranking in Google’s top 100, 18.2% were YouTube URLs. If you have demo recordings sitting in a folder, that is the cheapest thing on this list.

2. Build clusters, not single pages

Models break a question into sub-questions and retrieve across all of them. A company that covers a topic properly across several connected pages gets reached more often than one with a single strong page on the head term.

This is why we stopped writing standalone posts and started building clusters. Our GTM strategy guide is the hub for one of them, and this post is the hub for another.

3. Say what category you are in, plainly

A model can only describe you clearly if you describe yourself clearly.

B2B SaaS is unusually bad at this. If your homepage says you are a “revenue acceleration platform”, a model has nothing to anchor to and will describe you vaguely or not at all. Say the category in the words your buyers use, even if it sounds less impressive.

4. Name your competitors

Buyers ask models comparative questions constantly. “X vs Y”, “alternatives to Z”. Content that names alternatives honestly gets retrieved for those. Content that pretends competitors do not exist removes you from the comparison entirely.

I know this feels wrong if you have been trained never to mention a rival. Do it anyway.

5. Answer the question in the first two sentences

A model lifting an answer takes the sentence that states it plainly. Bury it three paragraphs in and the passage is harder to extract. This is also why FAQ pages still earn their place.

6. Keep it current

Recently updated content gets cited more than old content. One of the few genuinely easy wins, and the reason you are reading a rewritten version of this post rather than the 2025 one.

What does not work in LLM SEO?

I want to start this section with something awkward.

1. Watermarked sentences, which I recommended

The earlier version of this post told you to write “watermarked sentences”; phrases engineered to carry your brand name so models would repeat them. That was the advice circulating in early 2025 and I passed it on.

It does not hold up. A 2026 citation-preference study by Ando and Harada, built on a benchmark of 5,192 sentences, found models under-select sentences containing personal names by up to 20.1% and numeric sentences by up to 22.6% compared with how humans cite.

Planting your name in a sentence does not make it more likely to be quoted. It may make it less so. I got that wrong, and if you built anything on that advice, this is me telling you to stop.

2. Publishing more

More posts optimised for a system that rewarded coverage.

Citation rewards being the clearest source on a question, not having the most pages about it.

3. Schema on its own

Worth having. It helps a machine parse your page. It does not make a model trust you, and trust is accumulated somewhere else entirely.

4. Keyword density

Weak for ranking already. For citation it is irrelevant.

How do you measure LLM SEO?

Not with traffic. Pew Research found only about 1% of AI Overview views produce a click on a cited source, so judging this work by sessions will make good work look like failure.

Track prompts instead. A fixed set of questions a buyer would actually ask, run on a schedule, recording which brands get named and in what order, with competitors tracked alongside you.

The part people skip is the baseline. Run it before you change anything, or you will never be able to prove what worked. We wrote up how to do that yourself in the AI brand visibility post.

What can nobody guarantee?

Three promises that should end the conversation, whether they come from an agency or a tool.

1. Guaranteed citations. Nobody controls what a model retrieves.
2. A fixed position in an answer. There is no position. If someone says this, they are still thinking in rankings.
3. A timeline. When Gemini 3 became the default behind AI Overviews in January 2026, SE Ranking found it replaced roughly 42% of previously cited domains within weeks. Anyone promising you a date is guessing.

If you are evaluating outside help, the questions worth asking separate firms doing this work from firms that renamed what they already sold.

Where should a B2B SaaS team start?

In this order, because each step depends on the one before.

1. Weeks 1-2: Take a baseline. Ten questions a buyer would type, run across the models your buyers use, written down. Most teams skip this and then cannot tell whether anything worked.
2. Weeks 2-4: Fix your category language. One sentence, no hedging. If you cannot write it, nothing downstream works.
3. Weeks 4–16: off-domain presence. Reviews, podcasts, comparison content, communities, YouTube. The long one, and the one that matters.
4. Weeks 8 onward, in parallel: your own pages. Direct answers, honest comparisons, clear FAQs, kept current.
5. Ongoing: re-run the baseline monthly. Same prompts, same record.

That order is deliberate. Most teams do it backwards, spend six months on their own site, and wonder why nothing moved.

If you want to see how we run it, our SEO and AI visibility work sets it out.

Common questions