How AI Assistants Change the B2B Vendor Shortlist

What AI assistants actually change in a B2B shortlist

AI assistants can introduce a vendor, expand a known list, narrow options, summarise evidence or validate an existing preference. To understand how AI assistants change the B2B vendor shortlist, measure their role in a specific buying decision. A mention alone is not evidence of influence.

A shortlist is the small group of suppliers a buying group actively evaluates. The consideration set is broader: suppliers discussed, researched or ruled out. AI-mediated B2B buying includes any point where an assistant contributes to those activities.

An answer that names your company shows model visibility. It does not show that a buyer acted on the answer or that it caused a purchase. Investigate the gap with three evidence streams:

• observed assistant outputs;
• buyer accounts and contemporaneous artefacts; and
• commercial records, such as CRM notes and win-loss interviews.

Each stream answers a different question. Together, they can show whether AI altered a buyer’s route to a shortlist. See how B2B buyers use AI to research vendors.

AI changes the route, but buyers rarely start from zero

Available surveys suggest that AI can affect consideration. They do not show that buyers routinely begin with a blank slate.

Semrush reports that 92% of 519 US respondents who used AI for work said it had shaped their vendor shortlist, and 45% said it did so significantly (Semrush). This is a survey of AI users, not all B2B buyers. G2’s March 2026 global survey of 1,076 B2B software decision-makers found that respondents rated AI chatbots as the leading source influencing which vendors made their shortlist (G2).

Prior knowledge still matters. In 6sense’s European research, the eventual winner appeared on the day-one shortlist 92% of the time (6sense). Separate 6sense research reports that 26% used AI tools to discover new vendors (6sense).

These studies are not interchangeable. Their populations, category scopes, geographies and definitions of influence differ. Identify which job AI performed:

1. Discovering an unfamiliar vendor.
2. Expanding a list of known options.
3. Narrowing or reranking alternatives.
4. Validating a preferred vendor.

Do not assume AI has replaced peers, review sites, vendor websites or sales conversations. Test what it did in the buying episodes that matter to your category.

Frame the research around a shortlist-change question

Start with one question: which vendors were considered before the buyer used AI, which appeared or disappeared afterwards, and what evidence changed the buyer’s view?

Use one buying episode as the unit of analysis. Define the category, buying group, decision date and shortlist stage. This stops a broad claim about AI-assisted research becoming an unsupported claim about the final purchase.

Segment episodes by buyer role, use case, procurement constraints, geography, company size, purchase stage and prior category experience. Record the assistant, prompt, mode, account state and test date. A web-enabled fresh chat is not equivalent to a logged-in conversation with memory enabled.

Let interview language shape the test programme. That will help you find the questions buyers ask AI rather than relying on generic category prompts.

Step 1: Reconstruct the shortlist in buyer interviews

Recruit recent wins, losses and no-decisions. Where practical, interview more than one member of the buying group. A technical evaluator, commercial sponsor and procurement lead may each have seen a different part of the journey.

Build the timeline before asking about AI. Cover:

• the purchase trigger;
• vendors known on day one;
• research sources used;
• suppliers added, removed or reranked;
• the first sales contact; and
• when evaluation became serious.

Then ask which assistant the buyer used, the approximate prompt wording, follow-up questions, vendors named, sources cited and actions taken. “Did you use AI?” is too broad to establish a change in the consideration set.

With consent, ask for saved chats, copied summaries, internal messages, evaluation sheets or meeting notes. Redact personal and commercially sensitive material before analysis. Treat a buyer’s account of discovering a vendor through AI as recall until an artefact or subsequent action supports the sequence. Where support is unavailable, record that limitation rather than upgrading recall to fact.

This takes win-loss analysis for B2B SaaS beyond the final decision and into early consideration.

Step 2: Run a reproducible AI vendor-discovery test

Build prompts from buyer language, not generic “best software” searches. Interview wording reveals constraints that shape a real shortlist: implementation model, geography, integrations, compliance, team capability and switching risk.

Create a compact prompt set covering unbranded discovery, use-case fit, operational constraints, vendor comparison, risk checks and shortlist narrowing. Run those prompts in the assistants buyers named. For every run, record the date, location, product, account or memory state, search mode and fresh-chat policy.

Repeat each prompt. One response is a snapshot, not a stable account of assistant behaviour. Save the complete answer, cited links and screenshots or exports. Code:

• vendor presence and order;
• how each vendor is described;
• qualifying language and caveats;
• cited domains;
• factual errors; and
• material omissions.

Open every citation and check it against the source page. OpenAI warns that ChatGPT search results and citations can be incomplete, outdated or incorrect (OpenAI). An assistant’s explanation does not reveal why a vendor appeared.

Have two coders assess a sample of outputs. Compare their decisions, then document how ambiguous categories will be resolved before analysing the full set. This can improve qualitative coding with intercoder reliability.

Step 3: Triangulate AI evidence with deal data

Add structured questions to win-loss and post-decision research. Ask which assistant the buyer used, what task it performed, whether they first encountered a supplier there, whether the list changed and what happened next.

Compare the interview account with CRM source notes, call transcripts, branded-search timing, direct traffic and referral data where available. No single record proves causation. It can corroborate the buyer’s timeline or expose a contradiction.

Label each item consistently:

• observed artefact;
• corroborated buyer recall;
• uncorroborated buyer recall; or
• model-only simulation.

Last-touch attribution will not answer this question. Higher AI visibility or more direct traffic does not demonstrate that AI created pipeline. Report influence by task and stage: discovery, expansion, narrowing or validation.

Use the AI Shortlist Influence Evidence Matrix

Use this completed illustrative matrix as a worksheet for each buying episode. Score each stream from 0 to 2: 0 = absent, 1 = present but limited or uncorroborated, and 2 = directly observed or corroborated. The total makes confidence visible; it is not causal proof.

Apply these decision rules:

• Discovery influence: an unfamiliar vendor appears and receives subsequent consideration.
• Narrowing influence: AI-supported research is followed by a documented removal or reranking.
• Validation only: the list remains unchanged, but the buyer reports increased confidence and takes a documented next step.
• No demonstrated influence: a brand appears in assistant outputs without buyer evidence.

Look for repeated patterns. Consistent factual gaps may point to unclear source material. Recurring errors on third-party sites may need correcting at the original source. Weak evidence should trigger more interviews, not a stronger claim. This complements product research methods for B2B SaaS.

Act on evidence without confusing visibility with influence

Fix verifiable factual gaps and unclear use-case evidence that recur across buyer research and cited sources. Prioritise issues found across multiple buying episodes, assistants and commercial records. A single missing mention should not dictate the programme.

Retest a fixed prompt set on a declared cadence, while keeping interviews and win-loss evidence in the mix. This research can diagnose shortlist influence. It cannot reveal an assistant’s undisclosed retrieval, ranking or generation systems.

Use the findings to sharpen evidence and positioning, then validate value propositions with message testing.

Frequently asked questions

Do AI assistants create B2B vendor shortlists from scratch?

They can introduce an unfamiliar vendor, but current studies do not establish this as the usual route. Separate new discovery from expansion, narrowing and validation.

How can you tell whether AI changed a buyer’s shortlist?

Reconstruct the lists before and after AI use, identify the task the assistant performed, and seek a saved answer, internal message or subsequent evaluation action that supports the buyer’s account.

Which AI assistants should a vendor test?

Begin with the assistants buyers name. Record the product, mode, account context and date, then run the same buyer-derived prompts across more than one assistant.

Is appearing in an AI answer proof of shortlist influence?

No. It shows model visibility. Influence requires evidence that a buyer added, removed, reranked or validated a vendor because of the answer.

Can conventional win-loss interviews capture AI-mediated buying?

Yes, if the guide reconstructs early consideration and probes for specific AI tasks and artefacts. A generic question about research sources is unlikely to capture the sequence.

---

About Glasgow Research — Glasgow Research helps B2B SaaS teams turn customer and market research into product decisions. Work with us.