How to monitor competitors in ChatGPT, Gemini and Perplexity

The most useful AI visibility question isn't "are we mentioned?" It's "who is mentioned instead of us, on which questions, and why?" That's what clients ask after the first report, and it's what turns AI visibility from a curiosity into work an agency can do something about.

Competitor monitoring in AI answers comes down to four outputs: who appears, how often, on which prompts, and which sources the engines lean on when they mention them.

Choose competitors the AI actually names

Clients usually list the competitors they worry about. AI assistants often name different ones. Do both:

1. Start with the client's list — the 3–5 companies they lose deals to.
2. Run the prompt set once and read the answers. Note every company named that isn't on the list. Marketplaces, directories and review sites count too; "Booking.com" can be the real competitor for a hotel.
3. Keep the final list short. 3–5 tracked competitors per client is enough to make every chart readable. Add aliases for each (product names, abbreviations), so mentions aren't missed.

Track competitors on the same answers

Competitor numbers only mean something if they're measured on the same set of answers as the client's. For every answer, record whether each tracked competitor was mentioned. Then for each brand:

• Mention rate: answers mentioning it ÷ all answers.
• Share of voice: its mentions ÷ mentions of all tracked brands. The details — and the denominator mistakes to avoid — are in how to calculate AI share of voice.

Here is what that looks like for a fictional client (the same example data as our sample report):

That's a sentence a client understands: "Veloz is recommended more often than you — 40% of answers to 30%."

Find the prompts where competitors win

Averages hide the action. Break the results down by prompt (fictional example below) and sort for this pattern: the client is never mentioned, and at least one competitor is. Those prompts are the gaps.

The first row is a content and positioning problem. The second is a strength worth defending. The third tells you who the AI considers the obvious substitute.

Look at the sources behind competitor mentions

When an engine searches the web before answering, it cites sources. Collect the domains cited in answers where a competitor is mentioned and the client isn't. Patterns appear quickly:

• Review and comparison sites that list the competitor and not the client.
• Directories and marketplaces where the competitor has a complete profile.
• The competitor's own pages that answer the exact question — a pricing page, a comparison page, a city page.

This is where agency work starts: getting the client into the sources that keep getting cited, or publishing the page the engine is missing. Our explainer on why ChatGPT recommends a competitor covers the common causes and how to talk about them with clients.

Compare engines separately

ChatGPT, Gemini and Perplexity search differently and cite different sources, so a competitor can dominate one engine and be absent from another. Report competitor rates per engine as well as overall — otherwise a strong Perplexity result can hide a weak ChatGPT one. See ChatGPT vs Gemini vs Perplexity for why they differ.

A monthly competitor routine

1. Same prompts, same competitors, same schedule. Change them only deliberately, and note when you do.
2. Report the competitor table — mention rates and share of voice for client and competitors.
3. List the gap prompts — client absent, competitor present — with the competitors named.
4. List the top cited domains in those gap answers.
5. Note movement: which prompts the client gained or lost since last month, and which competitor gained ground.

Done by hand for one client, this is a long afternoon. Across a roster it's the reason agencies automate. Ranxii tracks up to five competitors per client on the same answers as the client, shows share of voice, lists prompts where the client is absent but competitors appear, and extracts the cited domains — every figure linked to the stored answers.

What not to do

• Don't report a competitor's "rank". Order changes from answer to answer; frequency is what's stable.
• Don't compare across different prompt sets. Two numbers measured on different questions aren't comparable.
• Don't treat one dramatic answer as news. If a competitor appears in a single answer for the first time, check the next few weeks before you call it a trend.