How to Measure AI Citation Share for Competitor Keywords

By UpGeo · 2026-08-31

AI answer engine citation share for competitor brand keywords is measured by running a fixed set of competitor-related queries across ChatGPT, Perplexity, Google AI Overviews, Gemini, and Copilot, extracting every brand mention and source URL, and calculating: (your brand citations ÷ total brand citations) × 100. Track both mention-level citations (your brand is named in the generated answer) and source-level citations (your domain appears as a source or citation). These two numbers rarely match because an engine can mention a brand while citing a review site, competitor page, or another intermediary.

What counts as a citation?

AI answer engines don't treat citations like traditional blue links, so split measurement into two layers:

• Mention-level citation: brand name appears in the prose or list of the answer, such as "Acme, BrandX, and BrandY."
• Source-level citation: brand's root domain appears in footnotes, source cards, or linked citations, such as brandx.com.

For competitor keyword measurement, report both. A competitor can be mentioned repeatedly in prose without earning source citations, which is a different gap from having your domain listed as a source.

Step 1: Build the competitor brand keyword set

Track 50–200 keywords per competitor. Start with high-intent patterns that tend to pull brand citations:

• {competitor} alternatives
• {competitor} competitors
• {competitor} vs {your_brand}
• best {category} tools like {competitor}
• {competitor} pricing, {competitor} reviews, {competitor} features
• What to use instead of {competitor} for {use_case}

Group keywords into four intent buckets: alternative, versus, evaluation, and purchase. Citation rates move a lot between buckets. Alternative and versus queries typically produce 2–5× more brand citations than broad category queries.

Step 2: Run a fixed prompt set across engines

The measurement only works if every run follows the same setup: same prompts, same engine settings, and the same collection day each week.

• Perplexity API: set temperature=0 and return_citations=true.
• OpenAI Responses API: use web search, store=false, and capture url_citations.
• Gemini API: enable google_search_retrieval grounding and record groundingMetadata.
• Google AI Overviews and Microsoft Copilot: often require manual capture; log answer text, date, and visible source cards.

For alternative queries, don't ask the engine to include your brand. That makes the measurement circular. The exception is "{competitor} vs {your_brand}" queries, where your brand belongs in the prompt.

Step 3: Extract and normalize citations

For each response, record:

• Engine and date
• Query and competitor target
• Full answer text
• Brands mentioned in the answer
• Source domains cited by the engine

Normalize brand names to a canonical brand ID. For domains, roll up subdomains to root domain: www.brandx.com, blog.brandx.com, and brandx.com all count as BrandX. Keep publisher domains like g2.com, capterra.com, reddit.com, and forbes.com out of the brand-domain share. Count them as a separate category because they often mention several brands in a single answer.

Tracking table template

Week	Engine	Query	Competitor	Brands mentioned	Source domains	Your brand cited
W1	Perplexity	best alternatives to Acme	Acme	Acme, BrandX, BrandY	g2.com, brandx.com, acme.com	Yes: mention + source
W1	ChatGPT	Acme vs BrandX	Acme	Acme, BrandX	acme.com, brandx.com	Yes: mention + source
W1	AI Overviews	Acme pricing	Acme	Acme	acme.com	No

Step 4: Calculate citation share

Apply three formulas:

• Mention citation share: (your brand mentions ÷ all brand mentions) × 100
• Source citation share: (your domain citations ÷ all source citations) × 100
• Competitor citation gap: competitor share − your share

Metric	Formula	Example
Mention citation share	(your brand mentions ÷ total brand mentions) × 100	195 ÷ 1,300 = 15%
Source citation share	(your domain citations ÷ total source citations) × 100	72 ÷ 480 = 15%
Competitor citation gap	Competitor share − your share	35% − 15% = 20 pts

In a 100-keyword set run across 5 engines, you get 500 responses. If those responses contain 1,300 brand mentions and your brand appears 195 times, your mention citation share is 15%. If the primary competitor appears 455 times, their share is 35%, leaving a 20-point gap.

Leave responses with zero brand citations out of the denominator. A response that names no brands and cites no domains doesn't count toward brand citation share.

Step 5: Segment by engine and intent

Don't track a single blended number. Citation share changes by engine and query type, so create separate views for:

• Perplexity vs ChatGPT vs Google AI Overviews vs Gemini vs Copilot
• Alternative queries vs versus queries vs evaluation queries
• Mention-level vs source-level

This answers the operational questions: "Where are we being cited?" and "Which engine drops us first?" For example, an 18% share in Perplexity with 6% in Google AI Overviews points to different fixes than a uniform drop across engines.

Step 6: Set a baseline and monitor weekly

Run the same set weekly or biweekly. AI answer citation share can move 10–30% between snapshots because engines rerank sources, change models, or rotate citations. Treat a change below 5 percentage points as noise. Investigate if:

• Your share drops more than 5 points against the baseline.
• The primary competitor share grows by more than 10 points.
• Your source-level share falls while mention-level share stays flat.

When that happens, check whether AI crawlers can reach key pages. Use UpGeo’s AI crawlers list to verify access for GPTBot, PerplexityBot, Google-Extended, and others.

How to improve low citation share

Measurement is only useful if it leads to fixes. Start with these:

• Make your brand entity and category relationships explicit on comparison and category pages.
• Add structured product comparisons, pricing tables, and alternate-to-X sections.
• Publish clean machine-readable context for answer engines using an llms.txt file. You can create one quickly with the llms.txt generator.
• Build the fundamentals of GEO, not just link building: answer engines cite sources that are specific, current, and easy to attribute.