How to Benchmark Your Brand in AI Search (+ Template)

How to benchmark your brand against competitors in AI search

AI search benchmarking answers a question your rank tracker cannot: when a buyer asks ChatGPT, Perplexity, or Gemini for a recommendation, do they hear your name or a competitor’s? This guide gives you the method, the five metrics that make a benchmark real, and a free template so you can start today without any tool.

Why AI search benchmarking is not rank tracking

In classic SEO, you benchmark by comparing positions for the same keywords. Your competitor is at position three, you are at position seven, and the gap is obvious and stable. AI search does not work that way. There are no fixed positions, only generated answers, and those answers vary from run to run, from model to model, and from market to market. The same prompt can name you today and skip you tomorrow.

That difference changes the method. A benchmark in AI search cannot be a single check, because a single check catches one draw of a variable process. It has to be built from repeated sampling of the same prompts over time, so you are measuring a pattern rather than a coincidence.

The five metrics that make a real benchmark

A credible AI search benchmark rests on five measurements, and each answers a distinct question.

Mention rate is how often you appear at all across your prompt set. It tells you whether the engines know you exist in this category.

Share of voice is your mentions compared with your competitive set on the same prompts. It tells you how much of the conversation you own versus the brands you compete with.

Answer position is where you sit inside the answer, whether you are the first recommendation or a footnote near the end. Being mentioned last is not the same as being recommended.

Citation share is whose sources the engine cites when it answers. It tells you which domains are shaping the category’s answers, which is often where the real battle is.

Framing, sometimes called sentiment, is how you are described when you do appear. Being named as “a budget option” and being named as “the market leader” are very different mentions.

Step by step: build your first benchmark manually

You can do this without buying anything, and doing it once by hand teaches you what the metrics mean. Pick roughly 20 prompts that reflect how buyers actually ask about your category, including recommendation, comparison, and price-and-quality questions. Run each prompt across ChatGPT, Perplexity, and Gemini. Log every result in the template: whether your brand was mentioned, where in the answer, which competitors appeared, and which sources were cited. Repeat the run weekly, because one run is a snapshot and the trend is the point. Then tally the five metrics.

Be honest with yourself about the cost. Twenty prompts across three engines, run weekly and logged carefully, is roughly an afternoon of work every week, and if you only run it once the variance in AI answers will mislead you. That real cost is exactly why teams eventually automate it, but doing it manually first means you will trust and understand the automated numbers when you get there.

The free benchmarking template

The template is a Google Sheet you can copy with no email gate. It has three tabs. The first is a prompt log with dropdowns for engine and for yes or no columns, plus one pre-filled example row so it is instantly clear how to use it. The second is a dashboard that auto-computes mention rate per engine and share of voice against each competitor you name. The third is a one-page set of instructions. Open it, choose File and then Make a copy, and start logging.

Automating it

Running this by hand works for one brand and 20 prompts. It stops working when you need daily data, many markets, or a wide competitive set. GetMentioned runs the same benchmark automatically, every day, across ChatGPT, Perplexity, and Gemini and across the markets and languages you sell in, and every number drills down to the stored answer behind it so you can see exactly why a metric moved. The share-of-voice view shows you, prompt by prompt, who wins and who is missing. See /features/competitor-benchmarking.

How to read your results

Three patterns come up again and again. If you are invisible almost everywhere, the problem is usually retrievability, so fix whether AI can find and use your content before anything else. If you are mentioned but never first, the problem is usually comparison content and reviews, so strengthen the pages and third-party signals that push a brand from “also exists” to “recommended.” If you are strong in one engine and absent in another, the problem is engine-specific sources, so look at which domains that engine cites in your category and where your gaps are. The engine tracker pages for ChatGPT, Perplexity, and Gemini go deeper on each.

Frequently asked questions

How many prompts make a valid benchmark?

Around 20 well-chosen, buyer-realistic prompts is enough to start and to see clear patterns. What matters more than the count is that you run the same set repeatedly, because the trend across runs is the signal.

How often should I re-run it?

Weekly at a minimum for a manual benchmark, and daily if you automate it. AI answers vary between runs, so a single measurement cannot tell you whether a change is real.

Can I benchmark in languages other than English?

Yes, and you should if you sell in more than one language. Answers and cited sources differ by language, so an English-only benchmark can miss how you actually perform in your real markets.

Why do my results differ between runs?

Because large language models are non-deterministic. The same prompt can produce different answers on different runs, which is why repeated sampling, not a one-off check, is the basis of a real benchmark.

Get your baseline in five minutes

If you would rather not build the benchmark by hand, run the free visibility report at /visibility-reports for an instant baseline with no signup. When you are ready for the automated, daily version across your full competitive set, book a demo from there.