How to Set an AI Visibility Baseline for Your B2B SaaS Brand

"We've been optimizing our SEO for two years. Now our sales team is telling us buyers are coming in having already heard of three competitors from ChatGPT, and we're not one of them. I have no idea where we stand."

That is the AI visibility problem in one sentence. You cannot fix what you have not measured. And most B2B SaaS brands have never measured their AI visibility at all.

This post covers what metrics to track, how to run the baseline query set across Claude, ChatGPT, Perplexity, and Gemini, and what the numbers mean once you have them.

TL;DR: An AI visibility baseline requires four metrics across five query types run on four LLMs. The whole thing takes 90 minutes manually. If your citation rate is below 30% on category queries, you have a structural problem, not a content quality problem.

What an AI visibility baseline actually measures

AI visibility is not the same as SEO rank. A brand can hold page-one Google rankings and be completely absent from AI-generated shortlists. The two systems pull from different signals.

An AI visibility baseline measures four things:

1. Citation rate. Of the queries most relevant to your category, what percentage surface your brand in the AI's answer? This is expressed as a percentage. If you run 20 category queries and appear in 6 of them, your citation rate is 30%.

2. Ranking position. When you do appear, where in the answer does your brand land? First mention carries meaningfully more buyer weight than fourth mention in a list of five. AI-generated lists function like organic search results: buyers click the first two or three and form their shortlist from there.

3. Framing accuracy. Does the AI describe your product correctly? This covers category (are you described as what you actually are), audience (is the target customer stated correctly), and capabilities (are the features and differentiators accurate). Stale framing can be worse than absence. A buyer who sees your brand described as a 2022 feature set forms a wrong impression before they ever visit your site.

4. Coverage spread. Which query types surface you and which do not? A brand might appear on direct "what is X" queries but be absent from problem-first queries ("I need a tool that helps me track X without Y"). Coverage spread tells you whether your visibility is broad or narrow.

These four metrics give you a complete picture of where you stand. Citation rate is the headline number. The other three explain the headline.

The five query types that define your baseline

Not all queries are equal for baseline purposes. The query types below represent the actual patterns buyers use when researching B2B SaaS tools in AI assistants. Run each one across all four engines.

Query type 1: Category recommendation query.

Format: "What are the best [category] tools for [target customer type]?"

This is the highest-stakes query for buyer intent. If a buyer has a defined problem and is building a shortlist, this is often the first question they ask. Your citation rate on category recommendation queries is the number that matters most.

Query type 2: Direct brand query.

Format: "What is [your brand name] and who is it for?"

This tells you how AI assistants describe you when they already know about you. It measures framing accuracy more than citation rate.

Query type 3: Problem-first query.

Format: "I need a tool that [specific problem your product solves]. What are my options?"

Problem-first queries are where query vocabulary alignment matters most. If your product page uses brand language but buyers frame the problem in different vocabulary, you will be invisible on problem-first queries even if you appear on category queries.

Query type 4: Competitor comparison query.

Format: "[Your brand] vs [main competitor] -- what are the main differences?"

This query type tells you whether AI assistants have enough information to position you accurately against your competitive set. If the AI can answer this with specifics, your entity signals are solid. If the answer is vague or incorrect, you have a gap.

Query type 5: Use case query.

Format: "What tool should I use for [specific use case your ICP has]?"

Use case queries are how buyers with a defined workflow search. Your visibility here depends on whether your content explicitly addresses that use case, not just the category.

How to run the baseline across four LLMs

Open four browser tabs: Claude (claude.ai), ChatGPT (chat.openai.com), Perplexity (perplexity.ai), and Gemini (gemini.google.com). Run each of your five query types in each tab. That is 20 query runs minimum.

For each run, record three things:

• Did your brand appear? (Yes / No)
• If yes, what position in the answer? (1st, 2nd, 3rd, 4th, 5th+)
• If yes, is the framing accurate? (Accurate / Partially accurate / Inaccurate)

Use a simple spreadsheet with rows for each query and columns for each LLM. It takes about 90 minutes. When you are done, you have a baseline.

One important note on LLM variability: AI assistants do not return identical answers every time you run the same query. Temperature and retrieval variability mean the same query can produce slightly different results across runs. To reduce noise, run each query twice with slight wording variation and average the results.

Step-by-step: establishing your baseline in five steps

Step 1: Define your three most important query categories.

Before you run anything, identify the three query categories your buyers are most likely to use. For most B2B SaaS products this is: category recommendation, problem-first, and comparison queries. These are where citation rate gaps cost you the most pipeline.

Step 2: Write five specific query instances per category.

Vague queries produce noisy results. "What is a good SaaS tool?" tells you nothing. "What are the best API monitoring tools for engineering teams under 20 people?" tells you exactly where you stand for a specific buyer profile.

Write out 15 queries total (5 per category). Use the language your buyers actually use, not your internal product language.

Step 3: Run all 15 queries across Claude, ChatGPT, Perplexity, and Gemini.

That is 60 individual query runs. Log the results in your spreadsheet. Include a timestamp because AI visibility changes over time and you will want to re-run this in 60-90 days.

Step 4: Calculate your citation rate per query type and per LLM.

Sum the "Yes" count per query type and divide by 15 (the number of queries in that category across three LLMs). That gives your citation rate. Do the same aggregation by LLM to see whether you have uneven coverage across engines.

Step 5: Flag framing problems separately from absence problems.

Framing problems and absence problems require different fixes. Absence means your entity signals are too thin or your query vocabulary does not match. Framing problems usually mean a high-authority source has stale or incorrect information that AI assistants are pulling from. Fix the source that is generating the wrong framing.

What good looks like (and what needs fixing)

Here is how to read the numbers once you have them.

Citation rate on category queries:

• Below 20%: You are structurally absent. The signals AI assistants use to recommend you are missing or too thin. This is a structural fix, not a content fix.
• 20-40%: You appear sometimes but not consistently. Your signals exist but are incomplete. This is the most common position and is fixable with targeted changes to schema, entity signals, and vocabulary alignment.
• 40-70%: Solid presence. You are in the consideration set for most buyers. Focus on improving framing accuracy and ranking position.
• Above 70%: Category-dominant. Maintain by keeping entity signals current and monitoring for competitor improvements.

Coverage spread:

If you appear on direct brand queries but not on problem-first queries, your product page uses your internal vocabulary but not the vocabulary buyers use. The fix is rewriting your product description and FAQ answers to match buyer language. The LLMRadar Audit shows this gap explicitly by comparing your citation rate across query types.

LLM-specific gaps:

Different LLMs pull from different citation stacks. Perplexity leans heavily on real-time web retrieval. ChatGPT combines training data with retrieval. Claude and Gemini have their own weighting systems. A brand with strong G2 and Capterra profiles often shows up well on Perplexity but poorly on training-data-weighted queries. A brand with extensive community discussion shows the reverse pattern.

Framing accuracy:

If AI assistants describe your product as the wrong category or list capabilities you do not have, the most common cause is a stale or inaccurate high-authority source. Check your G2 description, your Capterra profile, and any press coverage from the last 18 months. One stale article describing a deprecated feature can anchor AI framing for months.

The two things you can fix before the week is out

Running the baseline is fast. Full remediation takes weeks. But two fixes are available immediately and they move the citation rate within 30 days.

Fix 1: Update your review aggregator profiles.

G2, Capterra, and Trustpilot are weighted heavily by AI retrieval systems. If your profile description uses your 2022 positioning, AI assistants pull that description when buyers ask about you. Log in and update every field today: category, use case tags, target customer, product description. This takes 30-45 minutes.

Fix 2: Add a direct "what it is" paragraph to your product page.

AI assistants cite pages that state in plain language what the product is, who it serves, and what it does. Most product pages lead with a benefit headline ("Grow revenue faster") that provides no extractable information. Add one paragraph below the hero that reads: "[Product name] is a [category] tool for [target user]. It [core mechanic] to [specific outcome]." Two sentences. That single paragraph becomes your most cited content.

When to get a professional baseline

The manual baseline process above works. The limitations are: 60 query runs is still a small sample, one person logging results introduces inconsistency, and four LLMs across 15 queries misses the long tail of query variations buyers actually use.

The $197 LLMRadar Audit runs 40 buyer-intent query variations across ChatGPT, Perplexity, Claude, and Gemini. You get a structured citation rate score by query type and by LLM, a framing accuracy breakdown, specific gaps ranked by pipeline impact, and a prioritized fix list. It delivers in 48 hours.

If your marketing budget has room for one research investment before you start publishing content, knowing exactly where you stand is the highest-value starting point. Content that does not address your actual citation gaps improves organic traffic and does nothing for AI visibility.

Christine Johnson is the founder of OperatorIQ. The LLMRadar Audit methodology has been run across 50+ B2B SaaS sites across project management, sales enablement, API tooling, and marketing automation categories.