You have probably seen a vendor pitch you a single "AI visibility score" and felt unsure whether it meant anything. That instinct is right. A blended score that averages your brand's performance across ChatGPT, Perplexity, and Gemini into one number hides the exact information you need to act on: which platform cites you, which ignores you, and why.

The tools to track brand mentions in LLMs are maturing, but most teams still do not know what a good score looks like for their category, or how to build a benchmark they can trust. This piece walks through how to set one up yourself, what the numbers actually mean, and where manual tracking breaks down.

What "AI Visibility Score" Actually Means

An AI visibility score measures how often your brand appears when AI engines answer prompts relevant to your category. That sounds simple, but the way it is measured changes what the number tells you.

Citation Rate vs. Share of Voice

These are two different numbers, and conflating them leads to bad decisions. Citation rate is the percentage of prompt runs where your brand appears at all. Share of voice is your brand's portion of total mentions relative to named competitors on the same prompt set.

A brand can have a 60% citation rate (it shows up in 60 of 100 runs) but a low share of voice if three competitors appear in 80 of those same runs. Citation rate tells you whether you are in the conversation. Share of voice tells you whether you are winning it.

Why a Single Number Without Platform Breakdown Is Misleading

ChatGPT, Perplexity, and Gemini use different retrieval systems, different source preferences, and different response generation architectures. A brand that appears consistently in Perplexity may be nearly invisible in Gemini because Gemini weights different authority signals.

Any tool or platform that evaluates platforms monitoring brand visibility should report per-platform results. A blended average masks exactly the gaps you need to close. Passionfruit's ​AI visibility tracking breaks results down by platform, prompt theme, and competitor for this reason.

How to Benchmark It Yourself: Step by Step

You can build a working benchmark without any paid tool. It takes discipline and consistency, not software.

Building a Prompt Set That Reflects Real Buyer Questions

The prompt set is the foundation. If your prompts do not match how real buyers ask questions in your category, your benchmark brand visibility measures the wrong thing.

Start with 15 to 25 prompts that cover three layers: category-level prompts ("best project management tools for remote teams"), comparison prompts ("X vs Y vs Z for small businesses"), and recommendation prompts ("what should I use for [specific use case]"). Pull phrasing from your sales team's notes, from Reddit threads in your category, and from actual customer questions. Avoid generic prompts that no real buyer would type.

Running It Across ChatGPT, Perplexity, and Gemini Consistently

Run each prompt a minimum of 5 times per platform per measurement cycle. AI responses are probabilistic, so a single run tells you almost nothing. Record whether your brand was mentioned, where in the response it appeared (first recommendation, mid-list, or passing mention), and whether it was linked or just named.

Keep the prompt wording identical across platforms and across months. Changing the phrasing between cycles breaks comparability. Run your benchmark on the same week each month so seasonal variation does not distort trends.

Scoring Sentiment and Position, Not Just Presence

A mention is not automatically a good mention. Record whether the AI engine described your brand positively, neutrally, or with caveats. Track whether you appeared as a primary recommendation or as an also-ran. Over time, this sentiment layer reveals whether your content strategy is shifting how AI engines frame you, not just whether they name you.

What Good Looks Like: Benchmarks by Category

There are no universal benchmarks for AI visibility yet, because the field is too young and too variable across industries. But directional ranges are forming from the brands Passionfruit tracks.

For competitive B2B SaaS categories, a citation rate above 30% on category-level prompts is a strong starting position. For DTC e-commerce brands, where product recommendations are more fragmented, a 15 to 25% citation rate on comparison prompts is a realistic initial target. Market leaders in well-defined categories may see citation rates above 50%, but that is the exception, not the baseline.

The more useful benchmark is your own trend line. A brand moving from 10% to 25% citation rate over three months is making measurable progress, regardless of where competitors sit. Our ​luxury womenswear brand case study shows what sustained GEO work looks like across real metrics.

Where Manual Tracking Breaks Down at Scale

The process described above works for a small prompt set. It does not scale. Running 25 prompts across 3 platforms with 5 runs each generates 375 data points per cycle. Doing this monthly, tracking competitors alongside your own brand, and scoring sentiment turns a manageable task into a full-time job.

Manual tracking also cannot control for the variability built into AI responses. A brand that runs the same prompt on different days, at different times, or from different geographic locations may get materially different results. Statistical confidence requires sample sizes that manual processes cannot sustain.

How Passionfruit Labs Automates This

Passionfruit Labs runs repeated-sample measurement across ChatGPT, Perplexity, Google AI Overviews, and Gemini on tracked prompt sets built for each client's category. Citation rate, share of voice, competitor benchmarking, and sentiment analysis are tracked per-platform with enough sample depth to separate signal from noise.

The platform connects AI visibility data to our ​GEO service, so the tracking directly informs which pages to optimize, which entities to strengthen, and which prompt themes to prioritize. The measurement and the action run as one loop rather than two disconnected workstreams.

Start With the Prompt Set, Not the Tool

The most common mistake is buying a tracking tool before building the prompt set. The prompt set is the input that determines whether any measurement, manual or automated, reflects real buyer behavior. Get that right first. Build it from real buyer language, run it consistently, and track per-platform trends before worrying about scale.

When manual tracking hits its limits, see how Passionfruit Labs handles the measurement at scale, and ​talk to the team about where your brand stands across AI search.

Frequently Asked Questions

What Is an AI Visibility Score?

An AI visibility score measures how often your brand appears in AI-generated answers across platforms like ChatGPT, Perplexity, and Gemini. It is typically expressed as a citation rate (percentage of prompt runs where your brand is mentioned) or as share of voice relative to competitors.

How Many Prompts Should I Track for a Reliable Benchmark?

Start with 15 to 25 prompts covering category, comparison, and recommendation queries. Run each prompt at least 5 times per platform per cycle. Fewer prompts or single runs per prompt produce unreliable data because AI responses are probabilistic.

Why Does My Brand Appear in Perplexity but Not in ChatGPT?

Each AI engine uses different retrieval systems and source preferences. Perplexity retrieves from live web search results. ChatGPT blends training data with retrieval. Gemini weights different authority signals. Per-platform tracking is the only way to diagnose platform-specific gaps.

What Citation Rate Should I Target?

There are no universal benchmarks yet. For competitive B2B SaaS, above 30% on category prompts is strong. For DTC e-commerce, 15 to 25% on comparison prompts is a realistic starting point. Your own month-over-month trend line is the most useful benchmark.

Can I Improve My AI Visibility Score Without Changing My Website?

Some gains are possible through off-site signals like third-party mentions and review volume. But the largest improvements typically come from on-site changes: structured content, entity clarity, FAQ blocks, and validated schema. Most brands need both.