How B2B software buyers shortlist with AI, and the seven pages a SaaS site must get right

How do software buyers use AI to build a shortlist?

They ask an AI chatbot which tools fit their problem, read the answer, and start their evaluation from the names it gives. For many buyers that now happens before a search engine or a vendor's site is involved at all.

G2, which runs a software review marketplace, surveyed 1,076 B2B decision makers in North America, EMEA and APAC in March 2026. It found 51% start their software research with an AI chatbot more often than with Google.¹ The wording matters: that is "more often than Google", not "only in AI". In the same survey, 69% said AI chatbots had surfaced information that led them to choose a different vendor than they expected.²

US data points the same way. Semrush, which sells SEO and AI visibility software, surveyed 622 US B2B professionals in March and April 2026. Among the 519 who use AI at work, 92% said AI had shaped their vendor shortlist.³ Our note on that survey covers what it can and cannot tell you.

Forrester's broader buyer research points in the same direction. Its State of Business Buying 2026 survey of nearly 18,000 buyers worldwide, fielded in 2025 with exact dates not stated, found 94% use AI somewhere in their buying process.⁴ Different questions, different samples, one direction.

B2B software buying in 2026

The shortlist now starts in an AI answer

51%start research in an AI chatbot more often than GoogleG2, Mar 2026 ¹

69%chose a different vendor than expected after AI informationG2, Mar 2026 ²

92%of US B2B professionals who use AI at work say it shaped their shortlistSemrush, Mar–Apr 2026 ³

83%shortlisted three or fewer productsTrustRadius, Jan 2026 ⁵

The shortlist is short, and it is set early

A buyer who starts in AI does not carry ten names forward. TrustRadius, a software review site, surveyed 1,862 technology buyers in January 2026. It found 83% shortlisted three or fewer products, and 63% used AI at some point in their purchase.⁵ Of the buyers who used AI, 94% said they fact-check its answers at least some of the time.⁵

Put those together and the arithmetic is unkind. If an engine names three tools and you are the fourth, you may never be evaluated. If it names you with a wrong price or a missing integration, you may be dropped at the fact-check.

The group doing the checking is large. Forrester's same 2025 fieldwork, exact dates not stated, found buying decisions involve on average 13 internal stakeholders and 9 external participants.⁶ Each of them can ask an engine a different question about you: the developer about the API, the security lead about certifications, finance about price.

Our view: that is why a single "AI visibility" number hides the real work. A SaaS company is not visible or invisible in general. It is cited or missing on specific questions asked by specific people in the buying group, and each question maps to a page.

Where do the engines find answers about software?

From pages that state the answer plainly, on your site and on everyone else's. Engines differ in how many sources they draw on. In Semrush's 2026 AI Visibility Index of 126 million US prompts (January to April 2026), ChatGPT cited about 15 sources per response and Gemini about 3.⁷

That gap matters for a SaaS company. On an engine that cites many sources, your page competes with review sites, comparison articles and forums in the same answer. On one that cites few, a single page decides what the buyer reads. Either way, the page has to answer the exact question in words an engine can lift.

From prompt to signed contract

1. Buyer asks AI
2. Engine quotes pages
3. Shortlist of three
4. Fact-check with peers and sales
5. Finance review

Our view: most SaaS sites are built for the visit, not the quote. They lead with a value proposition and hide the specifics behind demos and PDFs. An engine asked "does this tool integrate with NetSuite and what does it cost for 50 users" finds nothing quotable and moves on to a review site that does answer.

The seven pages a SaaS site must get right

Seven page types answer most of the questions a buying group puts to AI. Each one should open with its answer, carry figures with dates, and match what your sales team says.

1. Pricing

Publish prices, or at least the structure: the unit you charge on, the tiers, what is included and where the price starts. An engine asked what a tool costs will quote somebody, and "contact sales" gives it nothing to quote from you.

Finance cares about this page more than it used to. G2's 2026 Buyer Behavior Report found finance involvement in software decisions rose from 31% to 46% in a year (data period and region not stated).⁸ Our note on that report covers what finance asks. A pricing page that states the unit, the minimum term and what usage costs answers it before the call.

2. Integrations

Give each major integration its own page or section: what syncs, in which direction, how often, on which plans, and what does not work. A logo wall answers "do you integrate with X" with a picture. Engines need the sentence.

Be specific about limits. An invented example: "Two-way sync of contacts and deals every 15 minutes on the Business plan; custom objects are read-only." That is a sentence an engine can quote and a buyer can trust. Most integration questions in a buying group come from IT, and IT is the role most likely to drop a tool for a missing detail.

3. Comparisons

Write fair "you vs competitor" pages for the comparisons buyers actually make. Fair means a criteria table, the places the competitor is stronger, and the date you checked. A page that claims you win on every row reads as an advert, and the buyer's fact-check will catch it.

Pick the comparisons from evidence, not ego. Your sales team knows which competitor appears in most deals, and your prompt runs show which names engines put next to yours. Write those pages first, link to the competitor's own documentation for the facts you cite about them, and set a date to recheck each one.

4. Security and compliance

State certifications with their dates and scope, where data is stored, who your sub-processors are and how to get the reports. Security questions come from people who can veto, and engines answer them from whatever page states the facts.

A trust centre behind a request form protects the reports, which is reasonable, but it should not hide the summary. Put the list of certifications, the audit period each covers and the hosting regions on an open page. The security lead will still ask for the full report; the engine only needs the facts to keep you on the list.

5. Alternatives

Publish an honest "alternatives to [your product]" page that says who should buy something else. Buyers ask engines for alternatives to the market leader and to you. If you do not describe your own place, a listicle written by someone with an affiliate deal will.

6. Implementation

Say how long setup takes, what the customer does, what you do, and what it costs. "Time to value" is one of the first questions a buying group asks, and one of the last that SaaS sites answer in writing.

Give a range with its conditions rather than a promise: a typical setup for a team of a given size, the data migration steps, the training offered and what slows projects down. The project owner in the buying group is the person who has to defend the timeline internally. A page that helps them do so becomes the page an engine quotes when they ask.

7. Reviews and customer proof

Keep your profiles on review sites current and consistent with your own pages, and publish case studies with named customers and dated results. Engines draw heavily on third-party review pages for software, so a stale profile with an old price undercuts a perfect pricing page.

Ask customers for reviews at the moments they have something specific to say, such as after go-live or a renewal, and answer critical reviews in public with facts. Engines read the responses too, and a calm correction of a wrong claim is quotable in a way a defensive one is not.

Our page map for a B2B SaaS site. Illustrative; adjust the roles to your buying group.

Example prompts for US buyers

Write your test prompts the way a US buyer would ask, naming a need rather than a vendor. These are illustrative, built for a fictional project management tool from Northwind Digital.

• "What project management software integrates with Salesforce and Jira for a 200-person company?"
• "Is Northwind Digital cheaper than its main competitor for 50 users?"
• "Which project management tools are SOC 2 Type II certified and store data in the US?"
• "What are the best alternatives to Northwind Digital for agencies?"
• "How long does it take to implement Northwind Digital, and do we need a consultant?"

Notice that only some of them name the brand. The unbranded prompts decide whether you reach the shortlist. The branded ones decide whether you survive the fact-check. Test both, on six engines, and on more than one day, because answers change from run to run.

Do buyers trust what AI tells them?

Partly, and they check. Gartner surveyed 645 B2B buyers in August and September 2025 (geography not stated) and found 69% prefer to validate AI-generated insights with sales reps.⁹ In the same survey, 51% said they were more likely to meet misleading information from generative AI and 49% from a sales rep.⁹ Neither source wins buyers' trust outright.

That has a practical consequence for SaaS teams. A buyer arrives at the first call with an AI answer about your price, your integrations and your weak spots. If your rep contradicts your own pages, or the AI answer contradicts both, the buyer has three versions and a reason to doubt all of them.

Our view: treat your seven pages as the single source of truth for sales as well as for engines. When a price or an integration changes, the page changes first, the review profiles next, and the sales deck after that. If you find an engine stating something wrong about you, our guide to fixing what ChatGPT gets wrong sets out how to trace and correct it.

Which of the seven pages to fix first

Effect on the shortlist →

Comparisons, alternativesStrong effect, but need care to stay fair. Second wave.

Pricing, integrationsStrong effect, mostly facts you already have. Start here.

ImplementationNeeds input from delivery teams. Plan it.

Security, review profilesQuick to correct; they matter most at the fact-check.

Ease of fixing →

How to audit the seven pages in two weeks

Start from the answers, not the pages. Run the prompts first, see what the engines say and cite, then fix the pages behind the gaps.

A two-week page audit

1. Days 1–2

   Write prompts for each page type and each buying-group role, branded and unbranded.

2. Days 3–7

   Run each prompt on six engines on three separate days; record names, claims and cited sources.

3. Days 8–10

   Mark wrong or missing facts and trace each to the page the engine quoted.

4. Days 11–14

   Rewrite the first lines of the weakest pages; update review profiles to match.

At Sigzen AI, our AI answer monitor runs this across ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI. It uses official APIs where they are offered and a licensed AI-answer data provider otherwise. Our AI Visibility Audit runs 150 buyer prompts on three separate days, against five competitors, and maps which sources the engines trust in your category.

Once the pages are fixed, connect the result to pipeline rather than stopping at citations. Our guide to proving AI visibility in B2B pipeline covers the signals finance will accept.

What these pages will not fix

Clear pages make you quotable. They do not make you the obvious choice in a category where a competitor has years of reviews, coverage and community discussion. Engines weigh what others say about you, not only what you say.

The survey figures in this post also have limits. They measure what buyers report, not what they did, and several come from companies that sell into software buying or search marketing. G2's buyer report, for example, says 82% of buyers sourced software recommendations from an AI chatbot in the last 24 months, but its data period and region are not stated.¹⁰ Read them as direction, not precision.

Our view: the seven pages are the part of AI visibility a SaaS company fully controls, and most sites have at least three of them wrong. Fix those first. Then work on the sources you do not control, starting with the review profiles that engines already cite.