The shortlist used to be a fortress

For most of the history of B2B software, the shortlist was decided before any vendor knew the deal existed. A buyer would ask a colleague, skim an analyst quadrant, remember a conference sponsor, and write down three or four names — almost always names they already knew. Everyone else was invisible. Getting onto that list required years of brand building, a review-site moat and enough marketing budget to be the name someone happened to remember on a Tuesday.

If you were a new vendor, your realistic path was to outspend, out-wait, or find the rare buyer willing to look past page one. The system structurally favoured incumbents, because human memory favours incumbents.

AI-assisted research breaks that structure — not by making buyers more adventurous, but by removing memory from the process entirely.

What G2’s Answer Economy research actually found

In March 2026, G2 fielded an online survey of 1,076 B2B decision makers responsible for, or influencing, software purchase decisions, drawn from North America, EMEA and APAC. The results were published as The Answer Economy: G2’s 2026 AI Search Insight Report. Three findings matter for this post:

• 69% of software buyers said an AI chatbot led them to select a different vendor than initially planned. Not “considered” — selected. The chatbot’s recommendation changed the outcome of the purchase.
• 33% purchased from a vendor they’d never previously heard of. One in three buyers signed with a company that was not in their mental map at the start of the process.
• 54% of buyers ranked generative AI chatbots the #1 source influencing which vendors make their shortlist — ahead of software review sites, market research firms and vendor sites. And 85% said they think more highly of a vendor cited by AI in its answer.

Two scoping notes, because numbers travel badly. First, these are survey responses, not purchase telemetry — they measure what buyers report, not observed behaviour. Second, the survey’s most-quoted finding, about where buyers now begin their research, is a separate claim with its own complications; we audited it line by line in our breakdown of G2’s chatbot research stat, so we won’t re-litigate it here.

What this post cares about is the second-order effect: if the assistant builds the longlist, the shortlist stops being a memory test.

Why AI melts the shortlist

An AI assistant answering “what tools can qualify inbound leads over WhatsApp in Portuguese?” doesn’t consult a mental brand hierarchy. It retrieves. It assembles candidates from whatever it can read and ground: product documentation, comparison pages, review data, forum threads, structured markup. Brand familiarity — the single biggest historical advantage — is simply not an input the model weights the way a human does.

That’s why one in three buyers in G2’s survey bought from a vendor they’d never heard of. The assistant heard of them, because the vendor was legible to machines. Here’s the structural shift, side by side:

The window won’t stay open forever

Call it a discovery window rather than a new normal, because the conditions that make it generous to unknown vendors are temporary. Right now, assistants lean heavily on open-web sources, and coverage of most software categories is shallow — a well-documented small vendor can out-rank a poorly documented large one. As engines sign more licensing deals, integrate more structured commercial data, and as every incumbent hires someone to do exactly what this playbook describes, gravity will reassert itself.

There’s an obvious irony in a waitlist-beta company writing this post: unknown vendors are precisely what AI-led discovery surfaces, and vendors like us exist in the market because buyers’ assistants no longer require a decade of brand before considering a product. We’re not a neutral observer of this window. We’re standing in it. That’s also why we’ve done the work below on our own site — and can tell you which parts are real effort and which are an afternoon.

Apply For Partnership

The operational playbook for being the discovered vendor

1. Make your entity unambiguous

Before an assistant can recommend you, it has to be certain who you are. If your company name, product names, category description and founding facts differ between your website, LinkedIn, directories and press coverage, retrieval systems see several weak entities instead of one strong one. Schema.org’s sameAs property exists for exactly this: it takes the “URL of a reference Web page that unambiguously indicates the item’s identity”, letting you tie your Organization markup to your official profiles so machines can merge the signals. We’ve written a full working guide in our post on entity consistency for AI search — it’s the highest-leverage, lowest-glamour item on this list.

2. Publish comparison content that names competitors fairly

Buyers ask assistants comparative questions — “X versus Y”, “alternatives to X for teams that need Y”. Assistants answer them by retrieving comparison content, and most categories have almost none that’s honest. A comparison page that names real competitors, concedes where they win, and is precise about where you win is far more likely to be retrieved and cited than a page that pretends rivals don’t exist. Fairness here isn’t ethics theatre; it’s what makes the content usable as an answer. If your comparison reads like an ad, it gets treated like one.

3. Show up where category prompts draw their answers from

When an assistant answers a category-level prompt, it draws on a mix of review platforms, forums, documentation and independent coverage. You don’t control that mix, but you can be present in more of it: public docs, genuinely useful guest content, community answers, and directory listings with consistent facts. Notably, a review-site profile is an input, not a prerequisite — we’ve documented how AI citations happen without a review-site profile, which matters if you’re early-stage and review volume is the one thing you can’t shortcut.

4. Ship machine-readable brand facts

Assistants reward sites that state facts in forms machines can lift cleanly: JSON-LD for your organisation and products, consistent product naming everywhere, and plain-language pages that answer one question each. There’s also llms.txt — a proposal by Jeremy Howard, first published in September 2024 with a v2 released in August 2026, to place a curated, concise guide for AI agents at a site’s root. Be clear-eyed about it: it’s a proposal, not an adopted standard, and no major engine has committed to honouring it. It costs an afternoon; treat it as cheap optionality, not a strategy.

What this means if you’re the buyer

The same window cuts the other way. If one in three of your peers is buying from vendors they’d never heard of, your evaluation process needs to handle unknowns: check the entity facts the assistant relied on, ask vendors directly for the evidence behind claims, and treat an AI recommendation as a longlist input rather than a verdict — remembering that 85% of buyers already admit an AI citation makes them think more highly of a vendor. That’s a bias worth knowing you have.

Sources & ownership

Frequently asked questions

What is the AI discovery window?

It’s the current period in which AI assistants build B2B vendor longlists from retrieval rather than brand memory, so well-documented unknown vendors can be recommended alongside incumbents. The window exists because assistants lean on open-web sources and category coverage is still shallow; it will narrow as engines add licensed data and incumbents optimise for the same systems.

How many B2B buyers actually purchase from vendors they’d never heard of?

According to G2’s 2026 AI Search Insight Report, 33% of surveyed software buyers “purchased from a vendor they’d never previously heard of” after AI-assisted research. The figure comes from an online survey of 1,076 B2B decision makers conducted in March 2026 across North America, EMEA and APAC — it’s self-reported survey data, not purchase telemetry.

Does an AI recommendation actually change how buyers see a vendor?

G2’s same March 2026 survey found 85% of buyers think more highly of a vendor cited by AI in its answer, and 69% said a chatbot led them to select a different vendor than they initially planned. Both figures are self-reported, but together they suggest AI citations now function as a trust signal comparable to analyst or peer endorsement.

Should we name competitors in our own comparison content?

Yes, if you do it fairly. Assistants retrieve comparison content to answer “X versus Y” prompts, and content that names real alternatives and concedes their strengths is more likely to be treated as a usable answer than a page that reads as advertising. The risk of naming competitors is smaller than the risk of being absent from the comparisons buyers’ assistants actually retrieve.

Is llms.txt required to be discovered by AI assistants?

No. llms.txt is a proposal — first published by Jeremy Howard in September 2024, with a v2 in August 2026 — for a curated guide to your site placed at its root for AI agents. No major AI engine has committed to honouring it, so treat it as cheap optionality alongside proven fundamentals like consistent entity data and structured markup, not as a requirement.

Apply For Partnership