AI Citation Mix Strategist

Research basis

This skill operationalizes Profound's research, Where do AI citations come from? (https://www.tryprofound.com/blog/where-do-ai-citations-come-from), published July 30, 2026.

Key takeaways from the study:

• Profound analyzed 11.84 billion citations across eight models, 29 industries, and 8,061 categories.
• About 57% of citations went to company-operated sites overall, but the mix varied sharply by model and industry.
• In 24 of 29 industries, the median Profound customer received more citations from brand sites than from earned or social sources.
• A universal prescription such as "do more PR" or "post on Reddit" ignores the category's actual citation market.

The study is an observational benchmark. Do not transfer its percentages to a brand forecast without matching industry, engine, language, geography, and time period.

Goal

Determine which source classes influence the target answer market, identify where the brand underperforms, and allocate effort to the smallest credible channel intervention.

Source taxonomy

Use granular classes for diagnosis:

• owned brand site
• competitor or adjacent company site
• earned editorial media
• institutional or primary authority
• PR wire
• social or user-generated content
• marketplace, directory, or review platform
• other or unknown

For executive reporting, roll these up into brand-operated, earned/institutional, and social/UGC.

Workflow

1. Define prompt clusters, engines, models or surfaces, language, geography, and date range.
2. Collect cited domains and URLs across a stable sample.
3. Normalize domains and classify every source.
4. Calculate citation share by source class, engine, and prompt cluster.
5. Separate citations to the target brand from citations that merely shape the surrounding answer.
6. Compare the brand's mix with relevant category benchmarks when available.
7. Identify addressable gaps:
   • weak owned evidence
   • missing third-party corroboration
   • absent institutional authority
   • relevant community gap
   • competitor-controlled source pressure
8. Map recurring citation neighbors and the role each source plays.
9. Recommend channel investment based on reach, evidence, feasibility, durability, and risk.
10. Define a longitudinal measurement plan.

Do not recommend manufactured mentions, fake reviews, undisclosed promotion, or irrelevant channel activity.

Decision rules

• Strengthen owned content when models already rely on company sites and the brand lacks useful facts or evidence.
• Pursue earned coverage when trusted editorial or institutional sources dominate the target cluster.
• Participate in communities only when those communities already contribute materially and participation can be authentic.
• Treat competitor citations as intelligence about missing facts and framing, not as permission to copy.
• Segment every recommendation by engine; aggregate figures can hide opposite platform behavior.

Output

Return:

1. Citation-market summary.
2. Source-mix breakdown by engine and cluster.
3. Citation-neighbor map with domain, class, role, frequency, and reachability.
4. Gap diagnosis distinguishing owned, earned, institutional, social, and competitive gaps.
5. Prioritized channel plan with expected impact, confidence, effort, owner, and evidence.
6. Measurement design with baseline, stable cohort, cadence, and review date.