B2B software buyers used to build a shortlist by reading vendor sites, review platforms and analyst reports. Increasingly they ask an assistant to do it for them, and the list they get back is the one they evaluate.
That creates an awkward gap for SaaS marketing teams. The shortlist now forms before the buyer visits any website, on the strength of sources the vendor mostly does not control. This guide covers how buyers ask, which sources shape the answers, how to keep your facts straight, how to design a prompt set, and what to measure.
How do B2B buyers use AI assistants to choose software?
B2B buyers use assistants to build a shortlist and then to narrow it. Their questions fall into four patterns, and each carries a different risk for a SaaS brand.
Category prompts build the shortlist
Best project management software for agencies. What CRM should a small sales team use. Top tools for security ticket triage.
This is where a buyer with no shortlist gets one. Answers name a handful of products, and evaluation often starts from exactly that list. Absence here can mean never being considered.
Comparison prompts narrow it
X versus Y for mid-market teams. Alternatives to X. Is X or Y better for compliance reporting.
Comparison prompts come later and carry high intent. The answer can promote you or quietly remove you, often on the strength of a single sentence about pricing, a missing feature or a reputation for difficult onboarding.
Integration prompts filter it
Which help desk tools integrate with Salesforce. Does X work with Slack. Best analytics tool for a Snowflake stack.
Integration questions are distinctive to software. A buyer’s existing stack is often a hard constraint, so these prompts cut the field sharply. Engines answer them from integration directories, documentation and marketplace listings, and the quality of those pages decides whether you survive the cut.
Problem prompts define the category
How do I stop onboarding emails going to spam. Why does our month-end close take so long. How do we see which features customers actually use.
Here the buyer describes a symptom and may not know a product category exists. Whichever tools the assistant names become the category in the buyer’s mind. These prompts are often less contested, because competitors focus on the obvious category terms.
Your homepage is rarely the page that decides the answer
For software questions, engines lean on review sites, comparison articles, documentation, integration listings and community discussion. Your own marketing pages are one input among several, and often not the strongest.
- Software review sites. Profiles on the major B2B review platforms give engines structured, third-party descriptions of what a product does and who uses it. Incomplete or outdated profiles are a common cause of thin or wrong answers.
- Comparison and “best of” articles. These map closely to category and comparison prompts, so they are cited often. Being included and described correctly matters more than the article’s search ranking.
- Documentation. Docs are where engines find specific, verifiable answers: supported features, limits, setup steps, API behavior. Public, well-structured docs are one of the strongest owned assets a SaaS company has.
- Integration listings. Marketplace and app directory pages on partner platforms confirm that an integration exists and what it does, which makes them a primary source for integration prompts.
- Community threads. Practitioner discussions show how real users talk about tools, complaints included. Engines draw on them for “what do people actually think” questions.
- Pricing and feature pages. These anchor branded and comparison answers, but only when the facts are in plain text rather than images, gated forms or a bare “contact sales” button.
The practical step is to look at which URLs engines actually cite for your prompts rather than guessing. For a given category that pool tends to be small and fairly stable, and it becomes your outreach and content plan.
Split what you find into two groups. Owned sources, such as your docs and pricing page, you can fix this week. Earned sources, such as review profiles, comparison articles and community threads, take longer: you can update your own listings, supply accurate information to publishers, and take part honestly in discussions, but you cannot edit what others write.
Software facts go stale faster than answers update
SaaS products change constantly, and AI answers lag behind. Inaccurate answers about software almost always trace back to an outdated source somewhere.
Training knowledge describes the product you used to be
An engine answering from memory knows your product as it was when its training data was gathered. Renamed plans, new features and changed pricing may simply not exist in that picture.
Retrieval finds whatever page ranks, not whatever page is current
An engine that searches live may land on an old review, a stale comparison article or a forgotten help page. It has no way to know that page is out of date unless something on it says so.
Third-party copies drift out of step
Every review profile, marketplace listing and partner directory holds its own copy of your product facts. Each one was accurate on the day it was written, and each one ages independently. An engine that finds two conflicting descriptions has to pick one, and it will not necessarily pick yours.
Wrong answers cost deals quietly
A buyer told that you lack an audit log, or that your entry plan costs more than it does, rarely contacts you to check. They move on to the next name on the list, and the error never surfaces in a lost-deal report.
How do you keep pricing, feature and integration facts accurate?
Publish the facts clearly in one authoritative place, keep every third-party copy in step with it, and check what engines say on a schedule.
- Make your own pages the clearest source. State plan names, starting prices, key limits and supported integrations in plain HTML text, with a visible “last updated” date and structured data where it fits.
- Update third-party profiles at the same time. Treat review profiles, marketplace listings and partner directories as part of every pricing or integration release.
- Answer common objections directly. If buyers believe you lack a feature you now have, say so plainly in a question-and-answer format on a page an engine can retrieve.
- Audit branded prompts regularly. Ask “how much does [you] cost” and “does [you] integrate with Z” on each engine and note every wrong claim.
- Trace each error to its source. Check the cited URLs. Fixing the page the engine relied on usually works better than publishing yet another page of your own.
A fuller method for correcting wrong claims is in what to do when AI gets your brand wrong.
How should a SaaS company design its prompt set?
Build the set from questions buyers actually ask, weight it toward category and comparison prompts, add integration and problem prompts, and report branded prompts separately. Then fix the set so the numbers stay comparable.
- Size. Large enough that one answer changing does not visibly move the total. Small sets turn normal run-to-run variation into false trends.
- Segments. If you sell to distinct segments by company size, industry or role, write prompts that include those constraints. A prompt about a startup and a prompt about a regulated enterprise can produce entirely different answers.
- Integrations. Cover the few integrations buyers treat as non-negotiable, not every integration you support.
- Competitors. Include “alternatives to” prompts for your main competitors, and one for your own brand.
- Branded prompts. Keep a small number, as an accuracy check. Blended with category prompts they inflate visibility.
- Stability. Once the set is live, change it rarely and in deliberate batches, and note the date on the trend line. Never drop a prompt because the answer is unflattering; that prompt is telling you something.
The best raw material comes from your go-to-market team: first-call questions from sales, onboarding questions from customer success, the wording of support tickets, and the reasons churned customers gave. Keyword tools are a poor source, because typed queries strip out the constraints that decide an assistant’s answer. The general method is in which prompts to track.
What should SaaS companies measure?
Measure whether answers name you, how they position you against competitors, which sources they cite, and whether branded answers are accurate. Segment by engine and by prompt type before averaging anything.
- Visibility. The share of answers naming your brand, reported separately for category, comparison, integration and problem prompts.
- Share of voice. Your share of brand mentions in those answers, against the competitors you actually lose deals to.
- Position. Whether you are the first product named or the last.
- Sentiment. Whether the description is favorable or carries caveats such as “powerful but expensive.”
- Sources. Which URLs are cited, and whether they are your pages, review sites, articles or community threads.
- Gaps. Prompts where competitors are named and you are not, the most actionable list you will produce.
- AI-referred visits. Traffic that arrives from assistants, treated as directional because many influenced buyers never click.
Report per engine. ChatGPT, Gemini, Perplexity, Claude and Grok regularly answer the same software question with different shortlists, and an average hides where you are strong and where you are missing. Noma tracks these metrics by asking each prompt on each engine once a day from a fresh session, and its gap analysis lists the prompts where competitors appear and you do not.
What to do first
In rough order of return on effort:
- Audit branded prompts on every engine and fix wrong pricing, feature and integration claims at their source. Cheapest change, most direct effect on deals.
- Bring review profiles, integration listings and public documentation up to date.
- Build a fixed prompt set from sales, support and customer success questions, and record a baseline by engine and prompt type.
- Work through the highest-intent gaps, starting with comparison and integration prompts, by creating or earning content that answers them.
None of this replaces good product marketing; it points it at the places buyers now form their shortlists. For the broader framework read the AEO guide, and to see how your product shows up today, run a free brand audit.