Share of LLM: the metric your boss will ask about next quarter 📏

Quick answer: share of LLM is the percentage of AI answers that mention your brand across a fixed list of buyer prompts, compared with your competitors. Pick 30 to 50 real buyer questions, run each one several times in ChatGPT, Perplexity and Gemini, count how often you’re named, and divide by total brand mentions. Track that one percentage weekly, not a “position”.

Your boss has seen the Google Analytics report. It says AI sends a sliver of traffic. Then they asked ChatGPT for the best tool in your category and you weren’t in the answer.

Next question: “So how are we doing in AI?” You need a number ready.

The idea 💡

Track share of LLM: the share of AI answers that name you, measured on a fixed prompt list, every week, with a workflow instead of a person.

Why it works 🧠

Traffic hides the problem. Conductor analysed more than 3.3 billion sessions and found AI referral traffic is 1.08% of all website traffic, with 87.4% of that coming from ChatGPT. The same report tells brands to treat AI visibility, meaning citations and mentions, as a core KPI alongside SEO. People get the answer inside the chat, so you can be recommended a lot and still see almost no clicks.

Rank tracking doesn’t work either. SparkToro had 600 volunteers run 12 prompts through three AI tools a combined 2,961 times. They found less than a 1 in 100 chance that the same list of brands appears in any two responses. Their advice: visibility percentage across dozens to hundreds of prompts, run multiple times, is a reasonable metric. A “ranking position in AI” is not.

That’s why this is a share, not a rank. One answer is noise. Fifty prompts run three times each is a signal.

The formula is simple. LLMpulse puts it as your brand mentions divided by total mentions for you and your chosen competitors, times 100. They suggest 30 to 100 prompts and weekly checks, since daily is too noisy and monthly misses changes.

How to do it ✍️

1. Write 30 to 50 buyer prompts. Real questions, not keywords. “Best marketing automation consultant for a 10-person B2B team in Melbourne.” “HubSpot vs Salesforce for a small agency.” Mix category, comparison and “alternative to” prompts.
2. Pick 3 to 5 competitors. Your share only means something against someone.
3. Run each prompt 3 times on each engine. ChatGPT, Perplexity and Gemini is plenty. Save the raw answers.
4. Count mentions. For each run, note which brands were named. Your share is your mentions divided by all tracked-brand mentions.
5. Log it weekly in a sheet. Add a column for the date, one for each engine, and one for what you changed that week.

Fifty prompts is enough to see a trend. One prompt is a coin toss.

Here’s a worked example with made-up numbers, so you can sanity check your maths. Say you run 40 prompts three times on one engine, which is 120 answers. Your brand is named 18 times, competitor A 30 times, competitor B 24 times and competitor C 12 times. Total mentions are 84, so your share is 18 / 84, about 21%. If next month it’s 27%, something worked.

Automate it ⚙️

Doing this by hand once is fine. Doing it every Monday is where it dies, so let a workflow do it. This is a perfect n8n automation job.

1. Schedule trigger: every Monday at 7am.
2. Google Sheets node: read the prompt list and the competitor list.
3. Loop with HTTP Request nodes: send each prompt, three times, to the OpenAI and Perplexity APIs (add Gemini if you like). The API isn’t identical to the chat app, so treat it as a consistent yardstick, not a perfect mirror.
4. AI or Code node: extract which brands each answer names, using the prompt below.
5. Google Sheets node: append one row with the date, share per engine and the change since last week.
6. Slack or email node: send yourself a three-line summary, plus any prompt where a competitor appeared and you didn’t.

Copy this into the extraction step:

You are a brand-mention extractor. Below is an AI answer to a buyer question.
List every company or product name recommended or compared in it.
Return JSON only: {"brands": ["Name", ...]}
Treat "HubSpot" and "HubSpot CRM" as the same brand. Ignore generic terms.

Tracked brands: {{ $json.tracked_brands }}
Answer: {{ $json.answer }}

The workflow does the counting. You do the reading.

Want it to go further? An AI agent can read the “you were missing” list and draft which page to write or refresh next. Pair it with the page types in the Best X for Y page ChatGPT keeps quoting and the quick-answer approach in how to rank in AI Overviews.

Quick wins to try 🧪

• 🔍 Run 10 prompts by hand today. It takes 20 minutes and you’ll find a surprise.
• 🏷️ Add a “named but wrong” column. Wrong pricing or the wrong service is worth fixing.
• 🧾 Log the sources each engine cites. Those are the pages you need to be on.
• 📆 Screenshot the first week. Your baseline makes every later win visible.
• 🆚 Track one “alternative to [competitor]” prompt. Buyers ask it more than you’d think.
• 🗣️ Add your own brand name as a prompt. What does AI say when someone asks about you directly?

How to measure it 📊

• Share of LLM: your mentions divided by all tracked-brand mentions, per engine.
• Prompt coverage: the percentage of your prompts where you appear at least once in three runs.
• Citation rate: how often your own domain is linked as a source.

Frequently Asked Questions

What is share of LLM?

Share of LLM is the percentage of AI-generated answers that mention your brand compared with competitors, across a fixed set of buyer prompts. It’s the AI-search version of share of voice.

How many prompts do I need to track?

Start with 30 to 50. LLMpulse suggests 30 to 100 prompts, and SparkToro’s research supports running them multiple times because answers vary between runs.

Can I track AI visibility for free?

Yes. A spreadsheet and manual prompts work for a first baseline. To run it weekly without the effort, build it in n8n, which you can self-host, and pay only for the API calls.

Why not just track my ranking in ChatGPT?

Because there isn’t a stable ranking. SparkToro found you’d almost never get the same list of brands twice, so a percentage across many runs is the honest measure.

Count first, then fix what the count shows. That’s the whole trick.

Sam

If you want this built for you, book a call.

Keep reading: The more AI names you, the more it invents links to you. Catch those lost visitors with a 404 page that converts.