AI Search Visibility for B2B: The Difference Between Being Ranked, Cited, and Recommended

On a Saturday afternoon in October, I typed a question into Google's AI Mode that one of our buyers might type on any given morning: "best fractional CMO firms for mid-market B2B companies."

That's our category. Marketri has been doing exactly that work for more than twenty years.

AI Mode gave me five firms, with a short write-up and a "best for" line for each. It was a confident, useful answer, and we weren't in it.

So I switched to the regular web results for the same query. Two of the top ten pages Google ranked for that search named Marketri by name, one of them specifically for mid-market B2B. We were in the sources Google ranked. We just weren't in what Google told the buyer.

That gap is the most important thing I can tell a B2B marketing leader about AI search visibility right now. For twenty-five years, being ranked and being chosen were close enough that you could manage them as one thing. They've split into three.

Three layers of AI search visibility

After that first prompt, I kept going. I ran seven buyer-style prompts about our own categories through Google AI Mode, logged out, and compared each answer with the regular search results. The pattern was clear enough that I now think about AI search visibility as three separate layers.

Ranked. Your page shows up in the classic search results. This is what your SEO dashboard measures, and it still matters, because it's part of the pool AI systems draw from.

Cited. The AI uses your content as a source. Your page gets a link or a footnote inside the answer, and your ideas shape what the buyer reads.

Recommended. The AI names you as an answer to the buyer's question. You're on the shortlist.

Here's how Marketri did across those layers, unedited, including the parts I'm not thrilled about:

Read the "fractional CMO firm that can help a B2B company adopt AI" row twice, because it surprised me most. Our CEO, Deb Andrews, wrote a LinkedIn article on the fractional AI CMO role. It ranked first in the organic results. AI Mode used it as the first source for the opening of its answer. Then it recommended five other firms.

We supplied the expertise, and the shortlist went to other firms.

And we're a firm that spends a lot of time on this. We're a Platinum HubSpot partner, and Marketri didn't come up anywhere in the answer about HubSpot partners. If this is happening to us, I'd bet money it's happening to you.

A caveat before going further: this was one engine, on one afternoon, with seven prompts. AI answers vary from one run to the next. Treat it as a pattern worth checking against your own brand. I wouldn't put these numbers in a board deck. The broader research, though, points the same direction.

Why ranking stopped predicting citation

In July 2025, Ahrefs found that about 76% of the URLs cited in Google's AI Overviews also ranked in the top ten for that query. When Ahrefs re-ran the analysis this March across 863,000 keywords and 4 million AI Overview URLs, that number was 38%. Another 31.2% of citations came from pages ranking between 11 and 100. The other 31% came from pages that didn't rank in the top 100 at all.

BrightEdge, using a different method, puts the overlap closer to 17%.

You'll also see a statistic that seems to contradict this: around 90% of AI Overviews include at least one top-ten URL (seoClarity). Both are true. An AI Overview cites several sources, so one page-one result can appear in nearly every answer even while most of the individual citations come from deeper pages. Page one still gets you into the pool. It just accounts for a shrinking share of what the AI picks from that pool.

The mechanism behind this is called query fan-out. Google's AI features don't run your buyer's question as a single search. They break it into related sub-questions, search for all of them in parallel, and then combine the results into one answer. On Google's Search Off the Record podcast, Google engineer Nikola Todorovic described it as forking the query and doing "the retrieval for multiple search queries" at once.

Think about what that means for a buyer asking about fractional CMO firms. Behind the scenes, Google might also be searching for pricing models, firms that serve private-equity-backed companies, and differences between a fractional CMO and an agency. The page that wins any of those sub-questions can earn a citation, even if it never ranks for the head term your SEO agency reports on every month.

So you can be on page one and still lose the answer, and a page you've never tracked can win it.

The traffic is shrinking, and the buying decision is moving upstream

Most CMOs I talk to have noticed this first as a traffic problem.

Pew Research tracked the actual browsing of 900 U.S. adults in March 2025. When a Google results page included an AI summary, people clicked a traditional result on 8% of visits. Without a summary, it was 15%. People clicked a link inside the summary itself on just 1% of visits. And 26% of visits to pages with a summary ended the browsing session right there.

That's the zero-click problem, and it's the half of the story your dashboard can see.

The other half is where the decision is being made. G2 surveyed 1,076 B2B software buyers this March. 51% now start their research in an AI chatbot more often than in a search engine. And 69% said they ended up choosing a different vendor than the one they originally planned on, based on what the AI told them.

Sit with that second number. The AI assistant is changing the shortlist, and many vendors never find out they were on it or that they dropped off.

The visitors who do click through also seem to be worth more. Ahrefs looked at its own analytics and found that AI search sent just 0.5% of its visitors but produced 12.1% of its signups, roughly 23 times the conversion rate of traditional organic search. That's one company's site, and Ahrefs says so plainly. But it fits what you'd expect: a buyer who has already had a long conversation about requirements, budget, and options with an AI shows up ready to talk.

For a CMO, this changes how to read your dashboard. A drop in organic traffic might mean demand is falling. It might also mean your buyers are getting their answers, and making their shortlists, somewhere your analytics can't see.

What earns a citation

The best research on getting cited comes from a 2024 study by researchers at Princeton and IIT Delhi, presented at KDD, a major data science conference. They tested specific content changes against generative engines and measured how much more visible a source became in the answers.

The changes that worked were concrete: adding quotations from credible people, adding statistics, and citing outside sources. The strongest methods improved visibility by about 40%. Keyword stuffing, the old SEO reflex, did slightly worse than leaving the page alone.

That matches what happened in my test. Outside of our own brand name and our home city, the prompts where Marketri got cited were all informational: how to choose an AI strategy consultant, what a fractional AI CMO actually does. The content that got pulled had specific frameworks, named criteria, and real numbers. When AI Mode laid out questions to ask a consultant in a first meeting, Marketri was one of the sources it cited.

Here's what that looks like on the page:

• Put the answer first. Under a heading phrased like the buyer's question, write two or three plain sentences that answer it directly. Fan-out retrieves pieces of pages, so each section should make sense on its own.
• Include real evidence. Use specific numbers, named sources, and quotes from named people. AI systems choose content that looks like it can be trusted, and specifics are how a page shows that.
• Publish data that only you have. A statistic that originates with you gets repeated by journalists, analysts, and eventually AI answers, with your name attached. One honest survey of your customers will do more for your citations than a dozen recycled "ultimate guides."

These steps get you cited. On their own, they didn't get Marketri recommended.

What earns a recommendation

The one prompt where Marketri was recommended was the Philadelphia query. AI Mode listed us first. When I looked at the sources behind that answer, most of them weren't our pages. They were other firms' roundups of Philadelphia fractional CMO providers that included us.

That's the pattern across every vendor shortlist I saw. The recommendations were built mostly from third-party lists, directories, and reviews. Several of those lists were written by competitors who, naturally, ranked themselves first. The AI used them anyway.

The research says the same thing at much larger scale. Meltwater analyzed 9.5 million AI citations and found that LinkedIn, Reddit, and YouTube together accounted for 47.5% of them. Company websites accounted for 18.7%. LinkedIn alone was the second most-cited source overall, and it ranked in the top five for B2B searches across categories including technology, financial services, and consulting and professional services.

Two more details from that study matter for B2B leaders:

• About 75% of LinkedIn's AI citations came from individual member profiles. Only about 25% came from company pages.
• 51% came from members with fewer than 10,000 followers. You don't need to be an influencer for this to work.

That's exactly what happened with Deb's article. One senior person, writing clearly about a topic she knows, shaped Google's answer more than anything on our company website.

So at the recommendation layer, the most important content about your company is written by other people: the firms that publish category roundups, your clients when they leave reviews, the journalist who quotes your data, and your own executives writing under their own names. You can't fully control that layer. You can show up for it.

The technical floor

None of this works if AI systems can't read your site in the first place.

Vercel and MERJ studied how AI crawlers behave and found that the major ones, including OpenAI's GPTBot, Anthropic's ClaudeBot, and PerplexityBot, didn't execute JavaScript at the time of their late-2024 study. If your key pages rely on JavaScript to load their content, those crawlers may see an empty page. Server-side rendering fixes this. Ask your web team one question: "If you turn JavaScript off, does our homepage still show what we do?"

Structured data helps too. Organization, Person, Article, and FAQ schema tell machines who you are and who wrote what, using a format they're built to read.

Two newer ideas get a lot of attention. I'd give them less of your budget until you address the critical pieces above.

llms.txt is a plain-text summary of your site written for AI systems. SE Ranking looked at about 300,000 domains and found no measurable link between having one and getting cited. It's cheap, so do it. Just don't expect it to move anything yet. (We have an AI-facing company information page on marketri.com, and it's clearly getting read: AI Mode drew on it when describing us in detail. I'd put that kind of page well ahead of llms.txt on your list.)

WebMCP and Google's new Lighthouse "Agentic Browsing" audit are about the next step, when AI agents fill out forms and complete tasks on your site. Chrome announced the audit in June, and WebMCP is still a proposed standard. Keep watching it, but it probably doesn't belong in this quarter's plan for most mid-market companies.

What to tell your CEO

If you're a CMO, your CEO is going to ask why organic traffic is down and whether marketing is to blame. Here's the honest version, in three sentences:

1. Fewer buyers click through because the AI answers their question on the search page, so traffic is a weaker signal of demand than it used to be.
2. More than half of B2B buyers now start research with AI, and most say it changed which vendor they chose, so the real question is whether we're on the AI's shortlist when someone asks about our category.
3. We're going to measure that directly every month, on a fixed set of buyer questions, and report how often we're ranked, cited, and recommended.

That last point matters more than anything else in this post. Google Analytics and Search Console can't tell you what an AI said about you. You have to go and look.

The afternoon audit

Here's how to run the test I ran, on your own company. You don't need a tool or an agency to start. It takes an afternoon.

1. Write 15 to 20 "golden prompts." Phrase them the way a buyer would actually type or say them. Split them across three groups:

• Category prompts: "best [your category] firms for [your buyer]," "top [category] providers in [your region]."
• Problem prompts: "how do I [problem you solve]," "how should a [buyer] choose a [category] partner."
• Brand prompts: "what does [your company] do," "[your company] vs. [competitor]," "is [your company] good at [your specialty]."

2. Run each one through Google AI Mode, ChatGPT, and Perplexity. Use a logged-out or private window so your own history doesn't skew the answer. Save the full answer and the source list.

3. Score every answer on four questions. Were we ranked (in the regular results)? Were we cited? Were we recommended? Was what it said about us accurate?

4. Collect the sources behind every shortlist you're missing from. This is the most useful list you'll make all quarter. It's your PR and partnership target list: the roundups, directories, and review sites the AI trusts for your category.

5. Fix the gaps where you can.

• Rewrite your most important pages answer-first, with real numbers and named sources.
• Publish a plain, factual company-information page: what you do, who you serve, where you are, who leads it.
• Ask happy clients for reviews on the platforms that showed up in your source lists.
• Get your executives writing on LinkedIn under their own names, about what they actually know.
• Check that your key pages show their content with JavaScript turned off.

6. Re-run the same prompts every month. Keep the wording identical. Individual answers will vary, so watch the trend across all your prompts and ignore movement on any single one.

I'm doing this for Marketri this month. The Philadelphia result tells me what works: other credible sites agreeing about who we are. The national category result tells me where that agreement doesn't exist yet. Each of the roundups that left us off is a specific, fixable gap.

Your traffic report will keep telling you about the first layer. The buyers making shortlists in AI chats are reading the third, and you'll only know where you stand there if you go and ask.

Sources

• Ahrefs, via Search Engine Journal: AI Overview citations diverge from rankings (March 2026)
• Comparison of overlap studies (Ahrefs, BrightEdge, seoClarity): AI Overview Citations vs Google Rankings
• Google on query fan-out, via Search Engine Journal: Google explains query fan-out
• Pew Research Center: Google users are less likely to click on links when an AI summary appears (July 2025)
• G2, via Demand Gen Report: Half of B2B software buyers now start their research with AI chatbots (May 2026)
• Ahrefs: 0.5% of visitors drove 12.1% of signups
• Aggarwal et al., KDD 2024: GEO: Generative Engine Optimization
• Meltwater: How AI is changing the way information is found (May 2026)
• Vercel and MERJ: The rise of the AI crawler (December 2024)
• SE Ranking, via Search Engine Journal: llms.txt shows no clear effect on AI citations
• Chrome for Developers: A developer toolkit to make your website agent-ready (June 2026)
• Marketri prompt test, Google AI Mode, October 10, 2026 (internal log)