For B2B SaaS
Understand how AI evaluates your category
When a buyer asks an assistant which tool to use, an answer gets composed from sources you did not pick and competitors you did not invite. CiteTrail asks those same questions on a schedule and records what comes back: which prompts you appear in, who is recommended instead, and what the model read to decide.
01: The problem
The evaluation now happens before the visit
You get described, not listed
There is no results page to rank on. There is a paragraph, and either it contains a sentence about you or it does not. What that sentence says matters as much as whether it exists.
The model picks the comparison criteria
A buyer asks which tool handles X. The model decides that price, integrations and setup time are the axes worth mentioning. You did not choose those axes and you will not be told what they were unless you read the answer.
Most of it leaves no referrer
The shortlist gets built in a conversation you cannot see, and the visit that follows often arrives direct or by brand search. By the time your analytics notices, the decision is made.
02: Buyer prompts
The questions that decide a shortlist
CiteTrail classifies prompts across 13 intent categories. Eight of them decide software shortlists, and they are where a B2B SaaS team should start. Each is scored out of 100 for buyer value, so a comparison question and a definition question do not get the same budget.
Category discovery
“what tools do X”
The buyer does not know your category by name yet. They describe a problem, and the model decides which kind of tool solves it.
Comparison
“A vs B”
You against a named rival. The model picks the criteria, and the criteria usually decide the winner.
Alternative
“alternatives to A”
Someone asks for alternatives to a competitor. Appearing here is how you enter a deal you were never invited to.
Competitor replacement
“switching from A”
Active switching intent. The buyer already wants out. These prompts convert, and they are the ones worth losing sleep over.
Integration
“works with our stack”
Whether you work with the stack they already run. A wrong answer here removes you before a human ever reads your docs.
Pricing and buying
“what does it cost”
What you cost and how you are sold. Models answer this from whatever they can find, which is not always your pricing page.
Use case
“best for this specific job”
Whether you fit a specific job. Narrow fit questions are where a smaller competitor beats a better-known brand.
Enterprise
“can it handle our scale”
Security, scale and procurement questions that decide whether you survive a shortlist at a larger account.
Prompts are also placed on a five-stage funnel: awareness, consideration, evaluation, purchase intent and post-purchase. That way a high-volume awareness prompt never outranks the evaluation prompt that actually closes deals. See how prompt portfolios work.
03: Competitive intelligence
Losing is a fact you can read
Displacement analysis shows the prompts where a competitor is recommended and you are absent, how consistently that happens across runs and engines, what the answer claimed about them in its own words, and which sources cited them and not you.
The useful part is the last one. A competitor winning an answer usually means a set of pages you are missing from, and those pages have URLs.
Illustrative
Prompt
What is the best alternative to [incumbent] for a mid-market team?
Recommended instead
Why they win
The answer cites a review roundup and a community thread, neither of which mentions your brand. Your own documentation is not among the sources the model reads.
An illustration of the shape of a displacement finding. It is not a customer result, and the brands are placeholders.
04: Coverage
The engines behind the answers
Every engine here has a production adapter and is exercised by real prompt runs. We name exactly what we query, because the surface changes what the number means.
ChatGPT
OpenAI API
Perplexity
Perplexity API (sonar)
Gemini
Gemini API
Azure OpenAI
Azure OpenAI Service
Claude
Anthropic Messages API
Gemini with Search grounding
LimitedGemini API + google_search tool
CiteTrail measures answers from provider APIs — the models behind these assistants — not scrapes of their consumer apps. Read the full engine list and its caveats.
05: Actions
Gaps become owned work, not a slide
Every gap becomes a typed action with a priority score out of 100: business impact 35%, visibility opportunity 25%, competitive gap 20%, ease 10%, confidence 10%. The weights are visible, which means the ranking is arguable.
■ Weights published, not hidden06: Questions
What B2B SaaS teams ask first
Do you scrape ChatGPT?
No. We call provider APIs, the models behind the assistants. Results will differ from a consumer chat window, and we would rather explain that than imply we are reading over your buyer's shoulder.
How do you decide which prompts matter for a software category?
Every prompt is classified by intent and funnel stage and scored out of 100 for buyer value. Comparison, alternative and competitor-replacement prompts usually score highest in B2B SaaS, because that is where a shortlist gets made.
What evidence do I get when a competitor is recommended instead of me?
The prompt, the stored answer, the brands named, the verbatim claim made about them, the sources cited, and how many runs and engines that pattern held across.
Answers change every time I ask. How is this measurable?
By sampling rather than looking up. Each prompt runs repeatedly, across engines, over a window, and confidence reflects both consistency and sample size. A finding from fewer than two runs is always labelled Low.
Which engines do I get?
Six in total. Free includes ChatGPT and Perplexity. Growth includes all six. The full list, with the exact API surface behind each one, is on the engines page.
Start with your own category.
Run a free audit to see how assistants answer for your brand today, or talk to us about the prompts and competitors worth tracking.