When we run an AI visibility audit on a B2B SaaS product, the founder almost always asks the same question first: "Are we even in there?" And the answer is almost always: "Sort of."
Being "in there" is not binary. There is a spectrum from completely invisible, to mentioned in a list, to cited with specifics, to actively recommended when a buyer describes the exact problem you solve. Most B2B SaaS brands sit somewhere in the middle, and they have no idea where.
This post covers the 4 metrics that define an AI visibility baseline, how to run the measurement yourself, and what the numbers typically mean for a B2B SaaS product in a competitive category.
Why standard analytics do not capture AI-driven buyer behavior
Google Analytics tells you how many people visited your site and what they clicked. It does not tell you how many buyers asked Claude or ChatGPT "what is the best tool for [your category]" and got directed to a competitor instead.
AI-assisted buying has become a standard part of the B2B research process. Buyers prompt AI models at the awareness stage, the comparison stage, and the decision stage. If your product is absent or misrepresented at any of those stages, you lose deals you never knew were happening.
The 4 metrics below are designed to close that visibility gap.
The 4 AI search visibility metrics
1. Mention rate
How often does your brand appear in AI responses when buyers ask category-level questions?
To measure: run 10 to 15 category queries across Claude, ChatGPT, Gemini, and Perplexity. Examples: "best tools for [your category]," "top [your category] software for [target company size]," "[your category] for [your ICP role]." Record whether your brand appears in each response.
Benchmark: In competitive B2B SaaS categories, well-positioned brands appear in 60 to 80% of category queries. Under 40% is a signal that your entity recognition needs work. Under 20% means you are effectively invisible to AI-assisted buyers.
2. Citation rate
Of the responses that mention your brand, how many include a specific claim about what you do rather than a list mention?
There is a meaningful difference between "Acme is one of the tools in this category" and "Acme is used by mid-market SaaS ops teams to automate invoice reconciliation, with a typical setup time of two weeks." The first is a mention. The second is a citation. Citations drive buyer intent. Mentions rarely do.
Benchmark: A brand with strong content structure and entity signals typically sees 40 to 60% citation rate on mentions. Under 20% means the model has low confidence in specific claims about you, which is usually fixable with structured content changes.
3. Recommendation rate
How often do AI models recommend your product when a buyer describes your ideal customer problem, without naming your product category?
This is the hardest metric to move and the most valuable to track. Example prompt: "I run a 50-person B2B SaaS company and I need to automate [specific workflow]. What should I use?" The model should surface your product if you have adequate AI visibility. Most B2B SaaS products do not make it into these unprompted recommendations.
Benchmark: Under 15% recommendation rate is typical for brands that have not invested in AI visibility work. Over 35% is a sign of strong positioning and content authority. The gap between these two is where most of the dollars are lost.
4. Accuracy rate
When AI models do describe your product, how accurate is the description?
We consistently find brands with significant AI visibility that are being described incorrectly. Wrong pricing, wrong target customer, wrong primary use case. This is often worse than low mention rate because it actively misdirects buyers who found you.
Benchmark: Any accuracy rate below 70% should be treated as a P0. Incorrect AI descriptions are usually caused by thin or inconsistent product descriptions across third-party sources like G2, Capterra, Crunchbase, and LinkedIn.
| Metric | What it measures | Typical range | Target |
|---|---|---|---|
| Mention rate | Presence in category queries | 40 to 80% | 75%+ |
| Citation rate | Specificity when mentioned | 15 to 60% | 50%+ |
| Recommendation rate | Unprompted problem-to-product match | 5 to 35% | 35%+ |
| Accuracy rate | Correct description of product | 50 to 95% | 85%+ |
How to run your own baseline audit in 45 minutes
You do not need a paid tool to get a rough baseline. Here is the manual method:
- Choose 4 AI models: Claude, ChatGPT, Gemini, Perplexity.
- Write 3 query types for your product: (a) category query, (b) problem query, (c) comparison query.
- Run each of the 3 queries in each of the 4 models. That is 12 data points.
- For each response, record: (a) did your brand appear, (b) was there a specific claim, (c) was it accurate?
- Calculate your rates from the 12 data points.
The 45 minutes includes the time to run 12 prompts and tally the results. It does not include diagnosis or fix planning, which typically takes another session.
What the manual method misses: Model responses vary between sessions. A manual audit gives you a point-in-time snapshot, not a reliable trend line. It also cannot tell you which content changes moved which metric, or how your baseline compares to competitors in your category. For that, you need structured repeat testing or a dedicated audit.
What low scores actually mean (and the fix sequence)
Low mention rate is an entity recognition problem. The models do not have enough consistent signals about who you are. Fix: publish more structured content about your category, get consistent descriptions across third-party directories, and add FAQPage schema to your product page.
Low citation rate is a content structure problem. The models know you exist but cannot extract specific claims. Fix: add a dedicated FAQ section to your product page with specific, factual answers. Add a pricing page. Add a comparison page with quantifiable differentiators.
Low recommendation rate is a positioning signal problem. The models do not associate your product with the specific problem it solves. Fix: publish content that names the problem you solve in the exact language your buyers use when they search. This usually requires a content audit against actual buyer queries, not the keywords you optimized for in 2022.
Low accuracy rate is a source consistency problem. Third-party sources disagree about who you are. Fix: audit your G2, Capterra, LinkedIn, and Crunchbase descriptions and make them consistent with your current positioning. This is often the fastest and cheapest fix in the sequence.
Want the full 47-point AI visibility audit for your product?
The LLMRadar Audit covers all 4 metrics above plus 43 additional checks across 5 AI models, 3 query types, and your full content and entity signal stack. You get a prioritized fix list with estimated impact per change.
Get the $197 LLMRadar AuditThe compounding effect of measurement
The 4-metric baseline is most useful when you run it quarterly and track the delta. A single snapshot tells you where you are. A trend line tells you whether your content and positioning changes are actually working.
Most B2B SaaS brands that invest in AI visibility work see mention rate move first, citation rate second, and recommendation rate third. Accuracy rate tends to improve quickly once source consistency is fixed. The full improvement cycle from first audit to hitting all 4 benchmarks typically takes 8 to 16 weeks depending on publishing cadence and how many third-party sources need updating.
The brands that close the gap fastest are the ones that treat AI visibility as a measurable discipline with a baseline, a target, and a fix sequence, not a vague goal to "do better at LLM SEO."