AI Search Competitor Monitoring Framework: Citation Share, Sentiment, Velocity

TL;DR

Pick 5-10 competitors, build a representative query basket, run those queries weekly across the AI engines you care about, and record who is cited, with what sentiment, and how fast new entrants appear. Convert results into citation share, sentiment, and velocity metrics. Review weekly, plan monthly, decide quarterly.

Why monitor AI search competitors

In classic SEO, you watch SERP ranks. In AI search, you watch citations: which sources the AI quotes, links, or attributes. Competitors who never ranked on Google can suddenly dominate AI Overviews because their content is more citable. Without a monitoring framework, the team finds out late — usually when traffic shifts.

Step 1. Select competitors

Pick 5-10 competitors using three criteria:

1. Direct competitors — same buyer, same product category.
2. Content competitors — ranking for or being cited on your priority queries, even if not direct competitors.
3. Authority benchmarks — high-authority sources the AI tends to quote (industry publications, vendor docs).

Freeze the list per quarter. Add new entrants only after they appear in two consecutive monthly reports.

Step 2. Build the query basket

A query basket is the fixed set of prompts you re-run every week.

• Top-of-funnel (~40%): "what is X", "X explained".
• Comparison (~25%): "X vs Y".
• How-to (~20%): "how to do X with Y".
• Buyer-intent (~15%): "best X for Y", "X pricing", "X review".

Target 30-60 queries per product line. Mix branded and unbranded; include long-tail queries that real prospects ask. Keep the basket stable for at least a quarter so trend lines mean something.

Step 3. Capture results

For each query on each engine, record:

• Cited sources (URL + display name).
• Citation position in the answer (top, middle, footer).
• Sentiment of the surrounding text toward each cited brand.
• Answer length and structure (is it a list, table, paragraph?).
• Timestamp and engine version (when available).

Manual capture works for small baskets; for larger ones, use an internal harness or a vendor tool that exposes the same prompts to each engine.

Step 4. Compute metrics

Citation share

citation_share(brand, engine, period) =

count(citations of brand in basket on engine in period) /

count(total citations across all brands in basket on engine in period)

Report per engine and aggregated. Track over time; the slope is more informative than the absolute number.

Sentiment

Classify the sentence(s) referencing each brand as positive, neutral, or negative. Compute net sentiment per brand per engine. Use a small classifier you can audit; do not rely on a black-box score.

Velocity

velocity(brand, period) =

citation_share(brand, period) - citation_share(brand, previous_period)

Velocity flags rising and declining sources before share leadership changes hands.

Content mix

For each cited URL, capture the section (definition, comparison, guide, case study, reference) and content type. Build a heatmap of which content types win in which engines.

Step 5. Action triggers

Define triggers in advance so the team reacts quickly:

Thresholds are starting points; tune them per workload.

Step 6. Reporting cadence

Always include the query basket diff in the monthly report so readers know exactly what changed in the inputs.

Dashboard schema

A minimal record per observation:

{
  "observation_id": "obs_2026-W18_perplexity_q14",
  "period": "2026-W18",
  "engine": "perplexity",
  "query_id": "q14",
  "query_text": "what is answer grounding",
  "answer_text": "...",
  "answer_length_tokens": 312,
  "citations": [
    {"brand": "geodocs", "url": "https://geodocs.dev/aeo/what-is-answer-grounding", "position": 1, "sentiment": "positive"},
    {"brand": "competitor_x", "url": "https://x.com/blog/grounding", "position": 2, "sentiment": "neutral"}
  ]
}

Aggregate this table into citation-share, sentiment, and velocity views.

Sample weekly report

Week 2026-W18 — Strategy product line

Engines: ChatGPT, Perplexity, Gemini, AI Overviews

Queries run: 48

Citation share (geodocs): 18% (+2 pp wow)

Top mover: competitor_y (+4 pp on Perplexity)

Triggers fired:

• competitor_y velocity > +5 pp on Perplexity → investigation queued

Notes:

• 3 queries newly cite a YouTube transcript page; investigate VideoObject schema

Common pitfalls

• Letting the query basket drift week to week.
• Conflating brand mentions with citations; only counted citations have URLs.
• Treating engine snapshots as ground truth; engines vary by date, locale, and account.
• Ignoring negative-sentiment citations — they still drive perception.
• Reporting raw counts instead of share; the basket size dominates raw counts.

FAQ

Q: How many queries should I monitor?

30-60 per product line is a good starting point. Smaller baskets are noisy; larger ones are expensive to label and rarely change conclusions.

Q: How do I measure sentiment fairly?

Use a small, auditable classifier or human labels with a clear rubric. Avoid black-box scores you cannot interrogate when leadership pushes back.

Q: Which AI engines should I track?

Start with the engines your buyers use. ChatGPT, Perplexity, Gemini, and AI Overviews are the common baseline; add Claude or Copilot when your audience is there (Perplexity Hub; OpenAI search).

Q: How often should I refresh the query basket?

Quarterly. More frequent changes break trend comparisons. Track adds and removes in a basket changelog.

Q: How does this differ from rank tracking?

Rank tracking watches blue links on a SERP. AI search monitoring watches who is cited inside an AI answer. The two are correlated but not the same; some pages that never rank are cited heavily, and vice versa.