How to Track Competitor Mentions in AI Overviews: A Step-by-Step Method

A B2B software company ranking number one for forty relevant keywords recently reported that ChatGPT consistently names competitors when users ask for solutions in the same space. The exact quote: “I can’t put a clean number on what we’re actually losing.” That problem has a method.

Tracking competitor mentions in AI Overviews, ChatGPT, and Perplexity is not complicated, but it requires a different instrument than rank tracking. Your position in Google tells you one thing. Who the AI recommends tells you another.

This article walks through the complete method: how to build a competitor prompt set, run it across AI platforms, log what you find, and calculate a share-of-voice number that shows where you stand. If you want to understand why competitors win AI citations in the first place, that question is answered in a separate guide. This guide covers what to measure and how.

---

Key takeaways

• Google rank and AI citation are decoupled systems. A brand can rank number one for forty keywords while ChatGPT names competitors for the same queries.
• AI answers are non-deterministic. Run each prompt two to three times and average the results. A single check is a data point, not a signal.
• Track fifty prompts minimum across category, comparison, use-case, and recommendation query types.
• Calculate share-of-voice per platform separately. A 40% ChatGPT presence and a 15% Perplexity presence require different fixes.
• When a competitor wins citations, the gap is almost always Access, Understanding, or Authority. Each points to a different action.

---

What you are actually tracking (and why it is not the same as rank monitoring)

AI citation tracking and rank position tracking measure different things. Your Google rank tells you where your page appears in a list of results. AI citation tracking tells you whether a model names your brand, or a competitor’s, when someone asks for a recommendation or comparison.

These two numbers can move in opposite directions. A practitioner tracking AI citations alongside Google rankings found that an article sitting at position eleven was being heavily cited in Perplexity, while a position-one page had no AI presence at all. That inversion is now confirmed by multiple independent practitioners.

There is a second important difference: AI answers are non-deterministic by design. A practitioner tested the same query repeatedly with no modification. The brand that appeared first varied across runs; sometimes the target brand did not appear at all. This is not a bug. It follows from how language models work. Sampling temperature means the model draws slightly different output each time.

The practical implication: a single check produces a data point, not a signal. Reliable tracking requires repeated runs over time, with results averaged.

The three metrics that actually matter:

1. Are you cited at all across your tracked prompt set?
2. How often are you cited versus each competitor, per query and per platform?
3. Is the citation pattern moving your branded search volume over time?

Sentiment scores and coverage dashboards beyond those three are largely noise at the current maturity of AI citation measurement. For a practical overview of how brands track AI Overviews mentions together with competitor share-of-voice, there are several approaches worth reviewing before you commit to a logging format.

Step 1: Define your competitor set and target query list

Start with who and what. The tracking method only works if you use the same competitors and the same prompts across every run.

Choose three to five competitors. These should be brands you already know appear in AI answers for your category, or that you suspect do. If you are not sure, run a handful of category queries across ChatGPT and Perplexity, note every brand named, and pick the ones that appear most often.

Build your prompt set from real demand. The most reliable source for prompts is Google Search Console. Pull your top fifty queries by impressions, then rephrase each as a natural-language AI question. If a query is “best ai visibility tool,” the prompt becomes “What are the best tools for tracking AI visibility?” Building from GSC data means you are tracking queries where real demand already exists.

Build prompts from real demand, not guesses

Fifty prompts is the minimum before patterns become reliable. Cover four types:

• Category queries: “What are the leading [category] tools?” or “Who are the top [category] agencies?”
• Comparison queries: “[Your brand] vs [Competitor A]: which is better for [use case]?”
• Use-case queries: “Which tool should I use to [specific job]?”
• Recommendation queries: “What do most [audience type] use for [problem]?”

How to word prompts to surface competitor citations

AI models surface competitors most reliably when a prompt invites a recommendation, not a factual lookup. “Who provides AI visibility audits in the US?” surfaces more competitor names than “What is an AI visibility audit?” The first asks the model to name brands. The second asks for a definition.

Avoid prompts that name yourself. The goal is to see what the model reaches for unprompted.

Step 2: Run prompts and log what you find

Run each prompt across ChatGPT Search, Perplexity, and Google AI Overviews. If Gemini is material to your category, add it. Spread runs across a few hours to reduce session-level caching effects.

For each prompt, run it two to three times before recording. Note every variation. If your brand appears in two of three runs, log it as “2/3” rather than a clean yes or no. Averaging across runs is what turns noise into signal.

If you are new to AI visibility tracking and want to build a baseline for your own brand first, the AI search visibility tracking guide covers the full tracking setup in detail. This guide focuses specifically on the competitor-facing layer of that same method.

What to record in your tracking log

Column	What to record
Prompt	Exact wording. Never paraphrase between runs.
Platform	ChatGPT Search / Perplexity / Google AIO / Gemini
Run number	1, 2, or 3 for that session
Your brand cited	Yes / No / Partial
Competitor A cited	Yes / No / Partial
Competitor B cited	Yes / No / Partial
Citation source URLs	Every domain the AI linked or referenced
Answer sentiment	Positive / Neutral / Negative toward each brand
Notes	Unusual behavior: disclaimer added, refusal, different mode

Keep the prompt wording frozen across every run. Even minor rephrasing shifts the model’s output, making runs incomparable.

Free tools that help

You do not need a paid platform to start:

• Bing Webmaster Tools AI Performance Report includes query intent categories, topic clusters, and share-of-voice data vs competitors for Bing AI. It does not cover Google AI Overviews or ChatGPT, but it is the most transparent native AI visibility data currently available from any major search engine. Google Search Console has no equivalent.
• Google Search Console impressions/click ratio gives a useful proxy. High impressions with low clicks on an informational query often means your content is surfacing inside an AI answer but not getting the click. Low impressions with low clicks typically means you are not being mentioned at all.
• Microsoft Clarity Citations Dashboard is free and shows which of your pages are being cited by AI systems at the page level.

If you evaluate a paid tool, apply one filter: does it separately identify (1) where you stand, (2) why competitors are winning, and (3) what to do about it? Most tools blend these into a single dashboard that tells you the score without telling you how to change it. Several platforms cover competitor mention tracking for teams that want to automate the logging step, though the underlying method is the same as the manual approach above.

Step 3: Calculate your competitor citation share

Once you have run your full prompt set, calculate a share-of-voice ratio for each brand.

The formula:

Your citation share = (prompts where your brand was cited) / (prompts where any tracked brand was cited) x 100

Run this calculation per platform. A brand that appears in 40% of ChatGPT runs may appear in only 15% of Perplexity runs. Platform breakdowns matter because the source ecosystems differ. Ahrefs found 88.46% of ChatGPT citations trace to the Google index, while Perplexity scrapes Reddit and live sources, sometimes within hours of a post going live.

For more on what this metric represents and how it connects to brand authority in AI answers, the share of model concept covers the underlying logic.

Read the number by platform, not just in aggregate

An aggregate citation share across all platforms can hide large platform-level gaps. A brand with strong Google-indexed authority may dominate ChatGPT but barely appear in Perplexity if it lacks Reddit and third-party editorial presence. Those gaps point to different fixes, so aggregate numbers can mislead.

Run your full prompt set on a consistent cadence. Monthly is the minimum for most categories. Fast-moving categories may warrant weekly runs. The trend across three to four cycles matters more than any single run number. For a full framework on structuring that cadence, see how to continuously monitor AI visibility.

Step 4: Diagnose why your competitor wins the citation

When you see that a competitor is cited more often across your prompt set, the next question is why. The answer almost always falls into one of three categories.

Access: can AI bots reach your content?

AI crawlers, including GPTBot, PerplexityBot, and ClaudeBot, follow robots.txt rules and respect noindex directives. If your pages are blocking them, content quality will not help. Check:

• Your robots.txt does not block AI-specific crawl agents
• Redirect chains are short. High 301 redirect rates burn AI bot crawl budget and reduce how many of your pages get indexed.
• Your important pages return clean 200 responses, not redirect loops or soft 404s

If a competitor has cleaner crawlability, that alone can explain a citation gap even when your content is stronger.

Understanding: is your content structured for extraction?

AI models extract and cite content that is easy to parse. The structural signals that help:

• A direct answer or clear claim at the top of each major section. Models pull from the first quotable sentence, not the middle of a paragraph.
• FAQ schema (FAQPage JSON-LD) on pages with question-and-answer content
• Organization and WebSite schema with correct entity signals
• HowTo schema on step-based guides
• Clear, specific headings that match the vocabulary searchers use

Structured data is confirmed as an emerging AI citation signal. A solo founder who reached 1.54M impressions and 1,200 monthly AI referral sessions credited structured data on every page: “I didn’t ask for that. The structured data made it happen.”

Authority: does the AI trust your brand as a source?

AI citation authority is built mostly off your own site, not on it. Approximately 95% of LLM citations come from unpaid, earned sources. About 30% come from journalistic and editorial sources specifically (Muckrack data from PR community research). Paywalled publications are increasingly blocked by LLMs, so coverage behind a subscription wall may not be accessible for citation at all.

What builds the kind of authority that drives AI citations:

• Non-paywalled third-party editorial coverage in your category
• Reddit community presence. Perplexity indexes Reddit content within approximately three hours of a post going live. Reddit is currently one of the highest-leverage surfaces for AI citations, and it is still underused by most brands.
• LinkedIn, YouTube, and industry newsletter mentions. Entity mentions across these surfaces build the citation footprint models use to judge whether your brand is a credible, known entity in the category.
• Getting onto the lists AI is already citing. Look at the citation source URLs column in your tracking log. If an AI is pulling from a “top ten” listicle that includes competitors but not you, getting onto that list is the next action. Writing another blog post is not. For a broader look at how to calculate AI share of voice against a competitor set, that framing is covered well in several independent analyses.

One guardrail: do not buy or manufacture brand mentions. Google confirmed in June 2026 that it treats manipulated brand mention signals like paid links: detected and disregarded. The only durable path is earned third-party presence.

When you know which gap is largest, Access, Understanding, or Authority, you have a prioritized fix list rather than just a visibility score.

When tracking alone is not enough

Manual tracking tells you what is happening. It surfaces the gap: which queries competitors win, which platforms they dominate, how large the citation share difference is. What it cannot easily tell you is why, especially when the answer involves technical issues on your site (crawl problems, schema errors, redirect chains) or structural gaps in your source ecosystem (which publications the AI is pulling from, where you are absent, what the model is inferring about your brand from what it can see).

An AI Visibility Audit maps the full picture across Access, Understanding, and Authority, across your site and your wider source ecosystem, and identifies specifically where the gap is largest and what to address first. If your tracking shows a persistent citation gap and the fix is not obvious from the log, an audit is the faster path to knowing what to do next.

FAQ

How do I know if my competitors are being mentioned in AI Overviews?

Open Google and search a category query. If a Google AI Overview appears, read it and note which brands are named. For systematic tracking, build a prompt set of fifty or more queries, run them across ChatGPT Search, Perplexity, and Google AI Overviews, and log which brands each platform cites. A single check tells you almost nothing. The AI’s output varies per run, so you need repeated runs before the pattern becomes reliable.

Can I track competitor citations in ChatGPT and Perplexity without a paid tool?

Yes. The core method is manual: build a prompt set, run it across platforms, and log results in a spreadsheet. Free options include the Bing Webmaster Tools AI Performance Report (native share-of-voice data vs competitors for Bing AI), Google Search Console impressions/click ratios as a proxy signal, and the Microsoft Clarity Citations Dashboard for page-level citation data. Paid tools add automation; manual tracking is sufficient to establish a baseline and identify the largest gaps.

Why does my competitor appear in AI answers when I rank higher on Google?

Google ranking and AI citation are decoupled systems. ChatGPT draws 88.46% of citations from Google-indexed content (Ahrefs, 1.4M-prompt study), but which pages it cites depends on how well structured and how widely referenced a brand is across third-party sources, not on rank position. A competitor with stronger third-party coverage, more Reddit presence, or cleaner structured data may win AI citations even if you outrank them in Google.

How many prompts do I need for reliable results?

Fifty is the minimum. Fewer than that and results are too volatile to distinguish a real pattern from run-to-run model variation. Cover multiple query types: category, comparison, use-case, and recommendation.

How often should I run my prompt set?

Monthly is the minimum for most categories. Fast-moving categories may warrant weekly runs. What you are looking for is a directional trend across three to four cycles. Any individual run contains noise. The trend is the signal.