What happens when buyers let an AI agent build the shortlist?

Your best salesperson has a gift for the first call. She reads the room, finds the real problem, and by the end of the conversation the prospect trusts her. Now picture a procurement manager who never makes that call, because she asked an AI agent to research the market, compare the options against her requirements and come back with three names. Your firm was not one of them, and nobody in your business knows the exercise happened.

How common is this? Using AI for supplier research is already close to universal. In 6sense's 2025 Buyer Experience Report, a survey of more than 4,000 buyers, 94% used large language models, mostly to summarise reviews or analyse data, and 94% of buying groups had ranked their preferred vendors before first contact. They bought from that first choice 77% of the time. Handing the whole research task to an agent is much newer, and no survey we trust yet measures how often it happens. But the tools now exist, and the firms that make those shortlists are the ones whose evidence is already public.

This guide covers which agents can do the job today, how one builds a shortlist, what it reads, whether it can contact you, and seven steps a founder-led firm can take this year to be on the list.

#Which AI agents can research suppliers today?

Several, though most sit on paid plans and some are available only in the US. An agent here means an assistant given a job rather than a question: it browses, clicks, compares and reports back.

Researcher matters most for B2B firms, because many buyers already have Copilot at work and it can combine a web search with the buyer's own notes and emails about the purchase.

#What does an AI agent actually do with a buying task?

It turns the brief into criteria, searches, reads each candidate and what others say about it, drops firms that fail a hard requirement, and writes up a comparison. No agent maker publishes how it weighs sources, so read what follows as how a careful researcher would work.

The difference from asking a question is delegation. Instead of "who are the good payroll providers for construction firms", the instruction becomes "find payroll providers that handle CIS deductions for construction firms with 100 to 500 staff, UK-based, with support hours that cover site start times, and give me three to compare". The agent then:

1. Turns the brief into criteria: some explicit (CIS, staff numbers, support hours), some inferred from the type of purchase.
2. Searches for candidates, using live search and what the model already knows. Firms that only appear for their own brand name often do not come up.
3. Reads each candidate's website, looking for the specific facts the criteria require. A page that says "flexible pricing tailored to your needs" gives it nothing to record.
4. Reads what others say: directories, reviews, trade press, partner listings and client case studies.
5. Checks whether the facts agree. A firm that describes itself differently on three pages, or whose reviews contradict its claims, is harder to recommend with confidence.
6. Writes a short comparison: this firm meets everything; this one is cheaper but lacks a capability; this one is strongest on support but smaller.

The buyer reads the comparison. Some will contact all three. Some will ask follow-up questions. Few start again from a blank page. By the time your sales team hears about the opportunity, if it hears at all, the field is down to a handful. What that does to your website is covered in what your website should do now visitors arrive pre-sold.

#Why do relationships arrive too late?

Because relationships, referrals and persuasion win deals among the firms already on the shortlist, and they cannot get you onto it. The 6sense figures show the shortlist was already largely settled before first contact, and an agent moves that moment earlier still.

In a human-run process there are many ways to win from a weak start: a warm introduction, a good first call, persistence when the buyer goes quiet. None of those exist in the research phase of an agent-run process. The agent has either found the evidence it needs about you or it has not.

Sales still matters. Once the buyer contacts the three firms, everything your team is good at comes back into play, with a warmer, better informed prospect. What has moved is the work of getting into the room: from your sales team to your published evidence.

#What does the agent read and weigh?

It reads your website first, then everything else that mentions you, looking for facts it can match against the buyer's criteria and evidence that others confirm them.

The explicit criteria come from the buyer: what you do, who for, where you are based and where you serve, what it costs or a range, accreditations, how long onboarding takes and how support works. If any is missing from your site, you fail that line of the comparison by default.

The implicit criteria are where firms separate:

• Results: outcomes for clients like the buyer, with numbers and context, or just logos and adjectives?
• Expertise: are senior people named, and do they publish or get quoted on the buyer's problem?
• Standing: do trade press, professional bodies, awards and independent reviews mention you?
• Transparency: is what a buyer needs easy to find, or behind a form?
• Consistency: do your website, directories, profiles and coverage tell the same story?
• Currency: is your latest evidence from this year, or from 2022?

The sector criteria vary: planning success for an architect, software and specialisms for an accountant, certifications and response times for an IT provider. The thread is the same: public proof of success with organisations and problems like the buyer's.

Consistency is where many established firms fall down. For TCC Group, a compliance firm with three brands, the finding was blunt: "Three brands, one group. AI could only see the pieces." Part of the cause was "machine-readable data that declared a name, a logo and nothing else."

#Can an agent contact us or book a call?

Increasingly, yes, so make it easy. Chrome auto browse can already fill in online forms and get quotes from tradespeople for US subscribers, and since July 2025 Google Search has been able to call some local businesses in the US to ask about pricing and availability. Expect B2B enquiries to go the same way in time.

• Publish a plain booking link to a named person's diary, readable as ordinary text on the page.
• Keep the enquiry form short, with labelled fields and no hidden steps.
• Check your spam protection. A CAPTCHA wall that stops every automated visitor also stops an agent acting for a real buyer, so prefer protection that filters spam after the form is sent.
• Show a phone number and email address that someone answers, and state prices or ranges and lead times so a call or an agent gets a straight answer.

#How should we prepare? Seven steps

Make everything an agent needs specific, evidenced, consistent and easy to reach, starting with the pages that say what you do and for whom.

#Step 1: Audit what an agent can find today

Write the brief a procurement manager might give an agent to find a firm like yours, with three hard requirements, then read your own site looking only for those facts. Put the same brief to ChatGPT, Copilot, Gemini and Perplexity and note who is named and which sources are cited.

#Step 2: Publish the basics you hold back

Price ranges and what drives them, onboarding timescales, who you are and are not a fit for, security and compliance information, integrations. An agent treats "not disclosed" as a gap, and a gap is a reason to prefer a competitor who filled it.

#Step 3: Turn results into case studies with numbers

Ask your ten best clients what changed, in numbers. Anonymise where you must, but publish the sector, the size, the starting position, the work and the result, one page each.

#Step 4: Make your experts visible

Name your directors and senior people, write proper biographies, and have them publish on the problems clients bring, on your site and in trade titles. This is the core of Reputation That Wins.

#Step 5: Gather third-party evidence

Independent reviews, trade press mentions, professional body listings, awards and speaking slots each tell the agent that someone other than you finds you credible. Why ChatGPT sounds unsure about some firms explains how assistants cross-check.

#Step 6: Make the facts agree everywhere

Check your name, address, services, sectors and claims on your website, Companies House, LinkedIn, Google Business Profile, directories and press. Agree the facts once and repeat them everywhere.

#Step 7: Make the site readable by agents

Put the answer at the top of each page, one page per service and sector, key text in the page's HTML rather than loaded by JavaScript, and accurate structured data. ID-Pal's rebuilt industry pages show the pattern: "Each opens with the question a compliance lead types, names the regulator, and answers inside the first ninety words." The technical checklist for an AI-ready website covers crawler access and the rest.

#What is agentic commerce, and will it affect B2B services?

Agentic commerce means agents that complete purchases, not just research them. So far it has centred on consumer shopping and simple bookings, such as the quotes and appointments Chrome auto browse handles. For B2B services, expect research, shortlisting and first contact to be handed to agents well before contracts are, which is why the steps above matter now. Firms that prepare early also build the evidence that keeps them recommended, as why first movers are hard to dislodge explains.

#What to do this month

1. Write the three-requirement brief a buyer might give an agent, and read your site looking only for those three facts.
2. Pick your three biggest gaps. For most firms they are prices, case studies and named experts.
3. Write to your five best clients this week and ask for a case study with numbers.
4. Draft a page on what your work costs and how long it takes, with honest ranges, and get it live.
5. Test your enquiry form and booking link as a stranger would, and remove any step an agent could not complete.
6. Book thirty minutes with your sales lead to list every question buyers ask that your website does not answer. That list is next quarter's content plan.

#Frequently asked questions

#How soon will buyers delegate supplier research to agents?

Some already do, and almost all use AI somewhere in the process. Fuller delegation, where the agent applies criteria and returns a ranked comparison, is arriving as the tools mature. We would not put a date on widespread use. The preparation is the same either way, and it improves how assistants describe you today.

#Will a human still make the final decision?

For most B2B purchases of any size, yes. The agent researches, compares and recommends, and a person approves, often after speaking to the firms. The change is that the person is approving a shortlist you may not be on, for reasons you did not get to argue with.

#What does this do to our sales team?

It moves their best conversations later and makes them better: prospects from an agent's shortlist already know what you do and how you compare, so the first call is about fit, edge cases and terms. Every question the agent could not answer is a page marketing should write. Researched outbound also reaches buyers before any agent is briefed. Our Pipeline That Fills works this way, with every message written and approved by a person; for Helm Club: "Around 2,000 well-matched founders approached a month across three LinkedIn accounts, each checked against their live profile first. The first message asks a question. No pitch, no link, and every reply gets a same-day answer from a person."

#How can I tell if an AI agent has visited my website?

Check your server logs for the user agents ChatGPT-User, Claude-User and Perplexity-User, which fetch pages when a person asks, alongside the search crawlers OAI-SearchBot, Claude-SearchBot and PerplexityBot. Agents that run inside the buyer's own browser, such as Chrome auto browse or Comet, can look like an ordinary visit. Add "an AI assistant" as an option on your enquiry form and ask on first calls.

#Do AI agents check review sites like G2 or Trustpilot?

They read what search surfaces, and review sites surface often in B2B answers. In G2's April 2026 survey of 1,076 software buyers, 45% said citations of review sites were the most confidence-inspiring signal in an AI answer. Pick the one or two review sites your buyers use and keep them current.

#We are a small firm. Can we compete here?

Yes. Agents reward specificity and evidence more than size. A small firm that says precisely who it serves and publishes real outcomes for that segment can beat a national firm whose website is written for everyone.

#Is publishing our pricing really necessary?

Publishing how pricing works is, even if the final figure depends on scope. An agent that finds "from £X a month, depending on the number of sites" can use it; one that finds "contact us" has nothing to compare. It also filters out poor-fit enquiries.

The buyer who hands the research to an agent will not call to explain why you were left off. The only way to know is to look at what the agent would read. For an outside view, book a free Growth Review: 30 minutes with a founder, with your site, two competitors and what Google and ChatGPT say about all three checked beforehand, and three fixes in writing afterwards.

#Sources