How to Find Out Which Competitor Is Winning AI Citations in Your Category

Ask ChatGPT the exact question your buyers ask, then Claude, Gemini, and Perplexity, then log who gets named and who does not. Do it for at least ten buyer-shaped prompts. The competitor whose name shows up most often across the four engines is winning your category's citations, and the fix is to publish answer-first content aimed at the specific prompts where you are invisible and they are not, then re-check those same prompts weekly until your name replaces theirs.

That is the whole loop. The rest of this post is how to run it without fooling yourself.

Step 1: Write down the prompts your buyers actually use

Do not start with keywords. Start with sentences. Sit with a whiteboard for twenty minutes and list the questions a real buyer would type into an assistant when they are three weeks from a purchase decision. Not "CRM software," but "what is the best CRM for a 15 person B2B sales team that already uses HubSpot for marketing." Aim for ten to twenty of these. If you have a sales team, mine their discovery-call notes. If you have a support inbox, mine that.

Split them into three buckets: category questions ("best X for Y"), comparison questions ("A vs B for Y"), and problem questions ("how do I solve Z"). You will need all three later.

Step 2: Ask all four engines, and record what they say

Open ChatGPT, Claude, Gemini, and Perplexity in four tabs. Paste each prompt into each engine. For every answer, log three things in a spreadsheet: which competitors were named, in what order, and whether you were named at all. Save the raw text. Screenshots help when a name gets stripped later.

Do this on two separate days at least a week apart, because AI answers drift. A competitor named on Monday may vanish on Friday, and one appearance is noise. Two appearances across two days is a signal.

Step 3: Score share of voice, not just presence

Count per competitor: how many prompts named them, across how many engines. A brand named in eight of twenty prompts across all four engines is beating a brand named in twelve prompts but only on ChatGPT. Multi-engine consistency is what compounds, because it means the assistant retrieval layers are drawing on overlapping source material about that brand.

Rank the top three. Those are the competitors you are actually losing to in AI, regardless of what any SEO tool tells you.

Step 4: Reverse engineer why they are winning

Click through to the sources the assistants cite when they name your competitors. Perplexity and Gemini show them; ChatGPT and Claude will often reveal them if you ask. You are looking for two things. What content earned the citation, a blog post, a review roundup, a G2 page, a comparison article, and what structural moves that content made. Almost always it opens with a direct, self-contained answer, uses question-shaped headings, and gives concrete numbers or specifics rather than hedged marketing copy. That is what the retrieval layer grabs.

Step 5: Publish the answer they should have gotten from you

For each prompt where you are invisible, publish one post whose first paragraph answers that exact prompt cleanly, in a way that makes sense quoted out of context. Question-shaped H2s. Real specifics. Your own first-party detail. No throat-clearing. If the prompt is a comparison, write the honest comparison, including where a competitor is genuinely better. Balanced comparisons get cited more than one-sided ones because assistants trust them more.

If you already have a page targeting the topic and it is not being cited, do not write a new one, republish that page with a rewritten opening paragraph, tighter headings, and stronger authorial credentials at the top.

Step 6: Re-ask the same prompts, weekly

This is the step most teams skip and it is the whole game. Every seven days, re-run your prompt list across the four engines and log the deltas. Which prompts now name you. Which competitor lost ground. Which prompts still ignore you and need a second, more specific piece. Citation authority compounds early, so a page that gets picked up in week three will usually keep getting picked up in week seven.

What this looks like when it works

Bosten Shoes ran essentially this loop. Its founder had watched ChatGPT recommend competitors for "best leather shoe brand in El Salvador" while never mentioning him. Once answer-first content aimed at those exact buyer prompts started publishing, Bosten went from unmentioned to the top recommended brand in that market, with the first ChatGPT-attributed sale showing up in revenue in under 30 days.

The honest version of the shortcut

Running steps 2, 3, and 6 by hand across four engines and twenty prompts every week is a real job, and most founders drop it by week three. That is what our Prompt Monitoring agent, Clark, does daily: it asks your buyers' questions verbatim across ChatGPT, Claude, Gemini, and Perplexity, tracks competitor share of voice on a leaderboard, feeds the gaps directly into its content queue, and then ties the resulting AI-referred sessions to actual revenue so you can see which prompt earned which dollar. If you want the manual version, the six steps above are the playbook. If you want the automated version, start with a free visibility check and see the gap before you decide anything. Either way, the competitor winning your category today is not winning because their product is better. They are winning because they answered the question first, and the fix is available to you.