Software Buyers Use AI to Add Vendors, Not Cut Them: 63% Grew Their Shortlist

Summary

62.7% of B2B buyers who opened an AI chat with a vendor already in mind added another vendor to their shortlist, so AI extends the list instead of compressing it.
Buyers use follow-ups to eliminate vendors: 57.5% ask about pricing, 54.5% compare vendors, 46% ask about integrations and 40% ask about setup.
71% of surveyed buyers used Claude for software research versus 62.5% for ChatGPT and 38.5% for Gemini, so audit Claude first and track each engine monthly.
Fix the three pages that decide the shortlist: a pricing page with a real starting number, a maintained comparison page and an integrations and setup page with named tools.
Map your follow-up prompt chains, test them on every engine and repeat the audit monthly.

The dominant assumption in B2B marketing is that AI compresses the buying journey. The story goes: a buyer opens a chat, the model surfaces one or two leaders, and everyone else disappears. G2's Chief Innovation Officer described it as "the third compression era of the buyer journey."

A November 2026 survey of 200 B2B software buyers, conducted by Wynter and analysed by Kevin Indig for Search Engine Land, says otherwise.

Of the 158 respondents who already had a vendor in mind when they opened an AI chat, 62.7% — nearly two thirds — added another vendor to their shortlist. AI did not narrow their list. It extended it.

The implication is concrete: the job is not to be first. It is to be retrievable at the moment a stranger opens a chat, and then to survive four or five rounds of follow-up questions about price, integrations and setup.

What Buyers Do Before the Chat

The Wynter data shows that most buyers arrive with prior knowledge. 79% of respondents already had at least one vendor in mind before typing their first prompt. Only 40 of the 200 started from a blank slate.

This matters because it reframes what AI actually does in the buying journey. It is not replacing discovery. It is extending due diligence on a list that was already forming.

Buyers bring their existing assumptions to the chat — and AI adds to them. For the 40 who started with no vendor in mind, 82.5% had a vendor added to their shortlist by AI. Yet only 40% of that group went on to report a vendor choice.

For buyers who arrived with a vendor already in mind, 72% of those who added another vendor reported making a choice. Starting with context accelerates the decision. Starting from zero does not.

What Happens During the Chat

The survey recorded actual prompt sequences, not just first queries. The picture that emerges is not a single lookup. It is an interrogation.

41.5% of respondents sent 3 to 5 follow-up messages after their opening prompt. One third sent 6 or more. The follow-up topics reveal exactly what buyers are trying to resolve:

Pricing: 57.5% of follow-ups
Vendor comparisons: 54.5%
Integrations: 46%
Setup: 40%

Each of these is an elimination question. A prompt sequence such as "Which tools integrate with HubSpot?" followed by "Of those, which has transparent pricing?" followed by "Remove any that take more than two weeks to set up" is a filter, not a search. Vendors who cannot supply a concrete answer at each step get removed.

The first answer still carries weight. Of the 126 respondents who reported choosing a vendor, 62% said that vendor appeared in the AI's first response. Another 34% said it appeared after follow-ups. But 75 of those 78 who got an answer in the first response still continued asking follow-ups — meaning appearance in the first answer earns inclusion, not a decision.

Conversion data supports the value of depth. Respondents who sent 6 or more follow-ups converted to a demo or signup at 41.8%, compared with 32.3% for those who sent 1 to 5. The rate peaked at 6 to 10 follow-ups and fell above 10, suggesting that sustained engagement signals intent, but very long sessions may reflect unresolved uncertainty.

What Buyers Do After the Chat

The AI chat does not close the loop. After the conversation ends, buyers leave to verify the shortlist elsewhere:

Vendor websites: 72.5% visited
Google search: 48%
Peer recommendations: 31%
Reddit: 29%
Review sites like G2: 23%

AI is a feed into existing channels, not a replacement for them. The buyer leaves the chat with a refined shortlist and then verifies it through your site, through search results and through social proof.

Ahrefs' own site data points the same way: AI-search visitors represented 0.5% of total site traffic but 12.1% of signups — a conversion rate 23 times higher than organic search visitors. The volume is lower. The intent is not.

68.5% of respondents started each chat with no prior context — a fresh session. 67.5% used more than one AI platform across their research. This is not a single-engine, single-session process.

The Engine Split That Most Teams Miss

The Wynter survey recorded which AI platforms respondents used for their software research:

Claude: 71%
ChatGPT: 62.5%
Gemini: 38.5%
Google AI Overviews or AI Mode: 15%

Claude led. Most marketing teams are not optimising for Claude.

This finding sits alongside a larger G2 survey of 1,076 buyers, which found ChatGPT dominant at 63%. The difference reflects panel composition and question design rather than a settled answer. Both findings are worth holding.

What FAIND's research adds is the distinction between referral share and recommendation share. ChatGPT drives the majority of AI referral traffic to websites — 92.4% of standalone AI referral clicks, citing data from Previsible. Claude drives far fewer clicks.

But in FAIND's prompt panels across two adjacent B2B fields, Claude supplied 56% and 74% of actual vendor recommendations. The engine that sends the most traffic is not the engine that names the most vendors.

The technical reason matters. Claude's answers increasingly depend on its dedicated web crawler, Claude-SearchBot, which Cloudflare Radar data ranked as the largest dedicated AI search crawler by Q2 2026. Claude fetches live page content to build its answers. If your pages are not cleanly structured and publicly accessible, Claude cannot retrieve them — and may not name you at all.

Demo and signup rates in the Wynter data also split by engine: Claude users converted at 35.2%, Gemini at 33.8%, ChatGPT at 31.2%. The differences are not large, but they are consistent with Claude driving higher-intent, more deliberate research sessions.

The Three Pages That Decide the Shortlist

The follow-up data produces a direct specification for three pages every software vendor needs. These are not suggestions. They are the pages buyers most frequently use to eliminate vendors from their shortlist.

Pricing Page

57.5% of follow-ups asked about price. DataDab's audit of 479 YC-backed B2B companies found that 51% would not show a buyer a price:

Contact sales gate: 24%
No findable pricing page: 18.8%
Pricing page with no number: 8.4%

A "contact sales" gate makes you invisible to AI on the most common follow-up question. If the model cannot retrieve a starting price, it either omits you from the answer or flags the gap. Either outcome removes you from the short list.

Publish a pricing page with a real starting number, named tiers and stated contract terms.

Comparison Page

54.5% of follow-ups asked to compare vendors. Comparing software is also the top use case for AI in software research overall, at 41%, per G2's research.

A study by Hasmord of B2B buying questions found that vendor-owned comparison pages made up a large share of citations in AI answers. In a separate analysis of "X vs Y" prompts, ChatGPT cited the vendors' own sites in 25 of 36 answers. LymLyt's comparison testing found that 10 of 15 answers cited nothing at all. When AI answers without retrieving, it relies on what it already knows — and vendors without a retrievable comparison page lose control of how they are described.

Treat the comparison page as a permanent, maintained asset — not a blog post that gets updated once.

Integrations and Setup Page

46% of follow-ups asked about integrations. 40% asked about setup. These are binary questions. A buyer running an elimination prompt — "remove any vendor that does not integrate with Marketo" — needs a named, findable answer. If the page does not list specific integrations, the model cannot confirm compatibility and may exclude you by default.

The requirements are simple:

Named integrations: list the tools you support, by name
Setup time: state how long implementation typically takes
Contract terms: publish the standard length and commitment

These are the facts that survive elimination prompts.

Six Steps to Act on This

Map prompt sequences, not just keywords. Take your top 20 commercial keywords and write out the five most likely follow-up prompts a buyer would send after the opening question. Use the Wynter follow-up distribution as your template for which questions dominate.
Audit across engines. Test each prompt sequence on every major engine using logged-out or fresh-session accounts. Personal history contaminates results. AI answers are volatile — Perplexity's cited sources changed by 38% within two weeks in one study — so the audit captures a point in time, not a permanent state.
Build the page that answers each follow-up. For every step in the prompt chain where you are missing or incomplete, build or fix the page that supplies the answer. The goal is to hand the model a retrievable fact, not to outrank a category page.
Add hard facts to pricing and integration pages. Models can only quote what your page states plainly. Strip the marketing language and publish the specifics buyers use to eliminate vendors.
Promote the comparison page to a first-class asset. Assign an owner, set a review cadence and treat it as infrastructure. A comparison page that is 18 months stale will not survive a prompt that asks which vendor added a Salesforce integration in the past year.
Track per engine, monthly. A single blended score obscures which engine is recommending you and which is not. Claude led software research in the Wynter panel and builds answers from a live crawl, so your inclusion status there can shift with your page structure rather than your brand. Track each engine separately and re-run the check every month.

Make Transparency Your Retrieval Strategy

The Wynter survey is self-reported data from 200 respondents. The sample is not representative of all B2B software buyers. Recalled prompt sequences may differ from what buyers actually typed. The survey establishes correlation, not causation: it cannot confirm that AI caused a purchase, or that the vendor AI introduced was the one ultimately chosen.

Kevin Indig, who analysed the results for Search Engine Land, disclosed that he is an investor in Wynter. That disclosure belongs in any fair reading of the data.

These limits do not invalidate the findings. They bound them. A pattern in 200 self-selected software buyers is a hypothesis worth testing against your own pipeline data — not a market-wide benchmark to cite without qualification.

What the data does establish clearly: AI is not closing the shortlist. It is reopening it. Buyers arrive with a vendor in mind, and more than six in ten leave with a longer list than they started with. The vendors who benefit are the ones whose pricing, comparison and integration pages are built to supply the specific facts that follow-up questions demand.

Retrievability is the entry condition. Transparency is what keeps you on the list.

Running this workflow by hand does not scale past a handful of prompts. LLM keyword tracking in Synscribe covers Google and every major AI engine, one engine at a time — the recurring check this workflow depends on. See Synscribe's tools for how this applies to your own shortlist.

Frequently Asked Questions About AI Search Visibility

What percentage of B2B buyers use AI chatbots to research software?

About half now start their research with an AI chatbot, according to G2. In Wynter's 2026 survey of 200 software buyers, 79% arrived at the chat with at least one vendor already in mind — and 62.7% of that group added another vendor to their shortlist during the conversation.

Does AI search shorten the B2B buying journey?

No. It usually lengthens the shortlist. Nearly two thirds of buyers who arrived with a vendor in mind added another one during the chat. AI does not replace discovery; it extends due diligence. The goal is to be retrievable at every follow-up prompt, not just present in the opening answer.

Which AI engine should B2B companies optimize for?

All three majors, but audit Claude first. 71% of buyers in the Wynter survey used Claude for software research, versus 62.5% for ChatGPT and 38.5% for Gemini. Claude also produced a disproportionate share of vendor recommendations, while ChatGPT sent the most referral traffic. Track inclusion per engine, monthly.

How do I get my software recommended by AI chatbots?

Publish the facts buyers ask about in follow-up prompts: a real starting price, named integrations, setup time and contract terms. Pricing alone drove 57.5% of follow-ups, with vendor comparisons close behind at 54.5%. If a model cannot retrieve the answer, it drops you.

What pages do I need to be cited in AI answers?

Three pages carry most of the weight:

Pricing page: a real starting number
Comparison page: current and maintained
Integrations page: named tools and typical setup time

These are the pages buyers use to eliminate vendors, not the ones they read for inspiration.

Does appearing first in an AI answer win the deal?

No. 62% of buyers who chose a vendor said it appeared in the AI's first response, yet 75 of the 78 whose first answer named a vendor kept asking follow-ups. Appearing first earns inclusion, not a decision. Transparency across the follow-ups is what closes the gap.

Is AI search replacing Google for B2B buyers?

Not yet. After the chat ends, 72.5% of buyers visit vendor websites and 48% run a Google search. 68.5% start each session with no prior context and 67.5% use more than one AI platform. AI feeds existing channels rather than replacing them, so visibility has to hold up in both.

How is GEO different from SEO?

GEO optimizes for retrieval inside an AI answer rather than ranking in a list of links. The unit of competition is the fact, not the page — a named integration, a stated setup time. Ranking first still helps, but only if the model can fetch and quote the detail it needs.

How often should I audit my AI search visibility?

Monthly, per engine. AI answers are volatile — one study found Perplexity's cited sources changed by 38% within two weeks — and Claude builds answers from a live crawl, so inclusion can shift when your page structure does. A single blended score hides which engine is naming you.

Is the Wynter survey data reliable?

Treat it as directional. It is self-reported data from 200 respondents, not a representative sample, and recalled prompts may differ from what buyers actually typed. The analyst who published it also disclosed an investment in Wynter. Test the pattern against your own pipeline data before citing it.