Your buyers built the shortlist without you.

The same G2 research found two things that are genuinely uncomfortable if you run marketing.

Around 69% of buyers ended up choosing a different vendor than the one they had in mind when they started. And about a third bought from a company they had never heard of before the research began.

Read that again. A third of deals went to a vendor who was invisible to that buyer at the start and became visible during an AI-assisted research session. That is either the best news you have had all year or the worst, depending entirely on whether the machine mentions you.

At the moment, most companies are on the wrong side of it. DerivateX ran 1,400 buyer-intent prompts across ChatGPT, Perplexity, Claude and Gemini against fifty B2B SaaS companies earlier this year. The average visibility score came out at 56.9 out of 100, and 44% of the companies scored below 50. These are not startups with no marketing. These are funded companies with real teams, structurally invisible at the exact moment their category gets discussed.

Here is the trap. If you look at AI referral traffic as a percentage of your total, it looks trivial. Most studies put it around 1% to 5% of sessions. Easy to ignore.

Now look at what that traffic does. Multiple independent analyses have found AI-referred visitors converting at somewhere between four and five times the rate of traditional organic. One agency dataset put AI-referred sessions at 14.2% conversion to a qualified action against 2.8% for Google organic on the same landing pages. Ahrefs found AI search visitors driving 12.1% of signups while accounting for 0.5% of visits.

The reason is obvious once you say it out loud. Someone arriving from an AI answer has already been told you are worth a look. They are not browsing. They are checking.

And the visits are only the visible part. Analysis from The Digital Bloom suggests around 70% of AI traffic arrives with no referrer header, so it lands in your analytics as direct traffic. Then there is the influence that never produces a click at all: the buyer who reads 'the main options are A, B and C', closes the tab, and types your competitor's name into their browser next week. Zero-click search on Google went from 56% to 69% in a single year. The dark funnel got a lot darker.

I test this constantly, and I would encourage you to do the same this afternoon. Take the ten questions your buyers ask before they shortlist. Not 'what is [your company]'. Real ones: best [category] for [industry] in [region]; alternatives to [the market leader]; what should I look for when choosing a [category] platform; is [competitor] any good for [use case].

Ask each of them in ChatGPT, Claude, Perplexity and Google's AI mode. Write down who gets named, in what order, and whether the description of you is accurate. That is your real market position in 2026, and it will probably differ from what your rankings dashboard tells you.

One warning from doing this a lot: the answers move. Research suggests only around 30% of brands stay visible from one regeneration of the same prompt to the next. Do not panic at a single bad result and do not celebrate a single good one. Run it monthly and watch the trend.

Most of what gets you cited does not live on your website.

This is the part that annoys marketers most, and I understand why, because we spent fifteen years being told the website was the asset we controlled. But AI systems corroborate. They favour claims that appear in several credible places, ideally places you do not own. Comparison articles, review platforms, analyst and trade coverage, community threads, directories.

G2's own 2026 buyer research found 92% of enterprise software buyers consult peer review platforms before shortlisting, and 72% will drop a vendor entirely over a negative review. That is a marketing channel and a citation source at the same time.

So the work splits three ways. On your site: answer real questions in plain, quotable sentences. Not 'we reimagine the workflow experience' but 'we do X, for Y companies, in Z markets, and here is what it costs'. FAQ blocks with schema. Comparison pages that are honest about who you are not for. Facts in text, not trapped in images.

Off your site: reviews, third-party lists, trade coverage, analyst mentions, and consistent facts about your company everywhere it appears. Earned media is doing double duty now: it is PR and it is machine corroboration.

Original numbers: models love citing statistics, and citing a statistic means naming its source. One decent annual benchmark for your category will out-earn fifty blog posts. This is the single most underrated move available to a mid-sized B2B company right now.

Gartner has forecast that by 2028 the large majority of B2B buying will be intermediated by AI agents, with an enormous share of spend moving through them. Whether or not that specific number lands, the direction is not in question, and there is a detail in it that people keep missing.

When a human researches with AI and gets a bad answer about you, there is a recovery path. They might click through, meet you at an event, hear about you from a colleague. When an agent runs the research inside a procurement workflow with no human validation step, there is no recovery path. You are in the retrieval data or you are not in the deal.

That is why I keep telling people this is not a content project to slot into next year's plan. The companies building citation authority now are building something their competitors cannot buy back quickly, in a window measured in months.

Part two looks at the team and the systems you need to actually do this, and why the marketing org chart everyone copied in 2019 has quietly stopped making sense.