Your buyer asked an AI for a shortlist. What did it find on your site?

Think back to the last time you had to shortlist a supplier for something you'd never bought before. There's a decent chance you opened ChatGPT or Claude before you opened Google, typed out the problem in a sentence or two, and got three or four names back with a line on each. Your buyers are doing the same with whatever you sell.

Most companies respond by asking whether they need an AEO agency. The quicker first step is to ask the assistants yourself, the way your best customer would, and see whether you're named. If you aren't, the fix usually starts with proof you already have and haven't published.

How are B2B buyers using AI to build a shortlist?

They use it early, to narrow the field before anyone talks to a vendor. Forrester's Buyers' Journey Survey 2025 found 94% of business buyers now use AI in the process. The bigger shift is in trust: twice as many buyers as the year before named generative AI or conversational search as more important than any other source, ahead of vendor websites, product experts and salespeople.

The traffic shows it too. Demandbase's platform data, covering more than 11 billion visits to B2B sites, recorded monthly ChatGPT referrals rising from about 645,000 in June 2025 to 2.6 million in June 2026.

There's a detail in the Forrester piece that matters more than either number. 61% of buyers use private AI tools provided by their employer. Those conversations leave no referrer in your analytics and no trace in any dashboard. A buyer can rule you out in one of those chats and you'll never see it happen.

The four-AI test: see whether you make the list

This is the quick version, from the buyer's side. It won't give you a trend line (our guide to measuring whether AEO pays covers the monthly method with a fixed prompt set). It will tell you, today, whether you're in the answer.

1. Write the question your best customer would type. Picture the last client you'd happily clone. Write their problem in their words, with their size and location, and ask who they should shortlist. For example: "We're a 60-person B2B software company in the UK and our outbound isn't booking meetings. Which agencies should we shortlist?" Keep your own category jargon out of it.
2. Open ChatGPT, Claude, Gemini and Perplexity in fresh sessions. Log out, or use a private window. A logged-in assistant that knows you work at your company will flatter you.
3. Paste the identical question into all four. Same wording every time. One changed word changes the answer.
4. Record four things per assistant. Whether you're named, where in the list, which competitors appear, and which sources it cites or links.
5. Ask one follow-up: "Why did you pick those?" The reasons it gives tell you what kind of evidence it found on the companies it chose.
6. Then ask about yourself by name. "What does [your company] do, and who uses them?" Copy the answer word for word. Wrong prices, old services and a vague description all point to pages that need fixing.
7. Score yourself out of four. Note the date, and run the same question again in a month.

A simple sheet does the job:

If you score zero, look at who came back instead and open the pages the assistant cited for them. Those pages are your brief: they show the kind of evidence the assistant went looking for and found somewhere else.

What does an AI quote when it names a supplier?

It quotes pages where a specific fact is easy to lift. The best published look at this so far is a June 2026 study by DerivateX, which asked ChatGPT for software recommendations across 40 B2B categories and classified every source it cited. About half the citations (51%) went to vendor-owned content, most often a "best tools in this category" post on a vendor's own blog. Independent niche sites took 23%, recognised media 15% and communities like Reddit and LinkedIn 10%. G2 and Capterra were cited zero times.

Treat that with care. DerivateX sells AI search work, it tested one model with one prompt style, and software isn't every market. The direction is still useful, and it has an awkward implication. When a competitor writes a well-structured comparison of your category, that page may be what the assistant reads when it builds your buyer's shortlist.

Now compare that with the sites we see on first calls. A version of this comes up a lot: founders describe their own website as a "digital business card". It has a logo and a contact form, and not much in between. That's fine for someone who already knows you. An assistant looking for a fact to support a recommendation finds nothing it can use.

The proof checklist: what gives an AI something to quote

Run your site against this list. Each line is something an assistant can quote and a buyer can check.

You don't need all six at once. Most companies already have the raw material for the first row sitting in old proposals and client emails.

For a sense of the format, our own Descartes case study states £5M+ pipeline in the first six months, from 120+ qualified meetings, and our Jua case study records three days from kick-off to the first qualified meeting. Each gives a number, the period it covers and what produced it, and that's the shape to copy.

Before any of this, make sure the assistants can read the pages at all. Plenty of sites block the AI search crawlers by accident through a robots.txt setting or a firewall rule. Our guide to which AI crawlers to allow has the lines to check.

Buyers check what the AI told them

Getting named puts you on the list. Your site and your customers decide whether you stay there. TrustRadius surveyed more than 2,500 tech buyers and vendors in 2025. Of the buyers who saw an AI summary while researching, 90% clicked through to the sources it cited. 77% read user reviews, and 54% spoke to an existing customer before buying. Vendors guessed only 38% would.

So the proof has to hold up when someone clicks. A case study that says "significant growth" might get you named once, and then the buyer lands on the page and finds nothing to check. I'd rather see three honest numbers than ten vague claims, because the buyer arriving from an AI answer is already in checking mode.

That 54% figure is worth sitting with too. If you can name two customers who'd take a ten-minute call, mention on the case study that references are available. It costs nothing and answers the question a serious buyer is about to ask.

Who can skip this for now

Some companies can park it. If almost all your business comes from a small, known set of accounts, through framework agreements or formal tenders, the shortlist is set by procurement rather than a chat window. The same goes if you sell into a market so small that every buyer already knows every supplier by name.

It's also worth being clear about what being named won't do. AI answers send very few clicks compared with how often they read and mention a company (the Cloudflare figures in our AEO measurement guide put some crawlers at tens of thousands of pages read per visit sent back), so don't expect a traffic spike. Treat a mention as an introduction you didn't have to make, and look for it in what new enquiries tell you about how they found you.

And be wary of anyone offering a shortcut. Fake forum recommendations and bought listicle placements get removed, and a buyer who catches one won't come back. Our guide to choosing an AEO agency has the questions that separate real work from that. We sell website and content work ourselves, for what that disclosure is worth. The four-AI test costs nothing, and if it comes back blank, the write-ups above are the first thing we'd tell you to do whether or not you hire anyone.