Menu

Browse all skills

Research agent skill

Competitor Audit

measure how AI engines perceive, categorize, and cite a named competitor, and turn that evidence into prioritized, winnable actions for the home brand. The method is the five-layer model in [references/ai-visibility-method.md](references/ai-visibility-method.md) pointed at a competitor, with the output oriented toward action rather than diagnosis.

View raw SKILL.mdView package on GitHub

This skill finds where a competitor's AI presence is thin, wrong, or resting on weak sources. It is not for copying their strategy or disparaging them.

When to use this skill

Use when the user asks to:

  • Audit a competitor's visibility or reputation in AI answers.
  • Find citation sources propping up a competitor that a better page could displace.
  • Compare how AI engines position the home brand versus a competitor.

Do not use when:

  • The subject is the home brand only. Run the same method in references/ai-visibility-method.md with the home brand as the entity, without competitive framing.
  • The user wants classic organic rankings only. Use seo.
  • The user wants keyword or prompt demand without competitor analysis. Use keyword-research.

Required inputs

competitor:
  name: ""
  website: ""
home_brand:
  name: ""
  website: ""
market:
  country: ""
  language: ""
  category: ""            # how the market names this category
priority_topics: []        # optional; derived in step 2 if empty
engines: []                # optional; default: 3 engines the audience uses
prior_findings: ""         # optional: earlier AI-visibility or keyword-research output

If competitor, home_brand, or market.country is missing, ask before starting.

Workflow

Copy this checklist into the working notes and tick items off:

- [ ] 1. Entity baseline captured from competitor's own site
- [ ] 2. Topics and prompt set built (incl. reputation + head-to-head)
- [ ] 3. Prompts run per sampling protocol and logged
- [ ] 4. Results read for placement, accuracy, sentiment, category
- [ ] 5. Citations scored with the strength rubric
- [ ] 6. Gaps prioritized by exploitability
- [ ] 7. Report assembled with evidence, confidence, next actions

  1. Entity baseline.

    • Record the competitor's canonical claims from their own pages: category, audience, core offerings, pricing model, and differentiators.
    • Give the URL and access date for each claim.
    • Note genuine strengths honestly.

  2. Topics and prompts.

    • Build the prompt set with references/ai-visibility-method.md and the prompt types in references/query-types.md.
    • Include reputation prompts ("is [competitor] reliable", "[competitor] complaints", "[competitor] alternatives").
    • Include head-to-head prompts naming the competitor against the home brand and against other rivals.
    • Include unbranded category prompts.

  3. Run and log.

    • Follow the sampling protocol: at least 3 engines, at least 3 fresh-session runs per prompt per engine, and a fixed locale.
    • Log every run in the run log format, with verbatim excerpts.
    • If the minimums are impossible, record the actual sample and downgrade confidence.

  4. Read results. For each prompt, record:

    • where the competitor is placed relative to other named entities
    • accuracy against the entity baseline (accurate, outdated, incomplete or wrong)
    • sentiment, and the source it comes from
    • any miscategorization, such as the wrong audience or category

    Apply the stability rules. Volatile findings are reported as volatile.

  5. Score citations.

    • Score every cited source with the citation strength rubric.
    • Flag stale sources such as old pricing or discontinued offers.
    • Flag thin third-party sources such as directory listings or thin aggregators.
    • Flag topics where the competitor has no strong source at all.

  6. Prioritize. Rank the gaps in this order:

    1. Competitor described inaccurately or unfavorably and the citation is weak (rubric 1–2). A stronger, accurate page can displace that source.
    2. Competitor present but resting on a single thin source.
    3. Whitespace: neither brand appears. This is a shared opportunity; label it separately.
    4. Competitor strong and well cited. Treat it as a quality benchmark, not a near-term target.

  7. Assemble the report using the template below.

    • Every gap must carry evidence references and a confidence label from references/evidence-standards.md.
    • Every gap must name a next action and an owner skill: content-brief, content, pillar-article, faq-generator, or seo.

Output template

# Competitor audit: <competitor> vs <home brand> — <market>, <date range>

## Sample
Engines: … | Runs/prompt/engine: … | Locale: … | Dates: … | Limits: …

## Competitor snapshot
How engines describe them (stable findings only) vs their own canonical claims.

## Strengths / thin / absent
| Topic | Status | Evidence (run IDs) | Stability | Confidence |

## Citation vulnerabilities
| Cited URL | Topic | Rubric score | Weakness | Winnable by |

## Prioritized actions
| # | Gap | Priority tier | Evidence | Confidence | Next action | Owner skill |

## Volatile / unverified observations
## Run log (appendix)

Approval gates

  • Get approval before publishing any comparison content or claims about the competitor.
  • Get approval before contacting third-party sites about their citations or listings.
  • Get approval before running paid API queries beyond an agreed budget.
  • Comparison content must be accurate, sourced to the competitor's own current materials, and fair. Do not make defamatory or unverifiable negative claims.

Stop conditions

Stop and report instead of continuing when:

  • The competitor cannot be identified unambiguously, for example because several companies share the name.
  • No AI engine access is available. Do not simulate or imagine engine answers.
  • A negative finding about the competitor rests on a single run or an unidentifiable source and the user wants to publish it.

Anti-patterns

  • Reporting only weaknesses and hiding genuine competitor strengths.
  • Treating one run as a stable finding, or paraphrasing an engine answer without keeping the excerpt.
  • Recommending an attack where the competitor's source is strong and current just because the topic looks attractive.
  • Misrepresenting the competitor's offering in comparison content.
  • Building near-duplicate pages for every prompt variant. This counts as scaled content abuse.
  • Seeking inauthentic mentions or reviews to influence AI answers.
  • Ending findings without a specific next action.

Completion checklist

  • Entity baseline sourced from the competitor's own site, with access dates
  • Prompt set covers all prompt types, plus reputation and head-to-head prompts
  • Sample meets the protocol, or its limits are stated and confidence downgraded
  • Every finding has a stability label, a confidence label and run references
  • Citations scored with the rubric, not just listed
  • Strengths reported honestly. Whitespace separated from direct attacks
  • Each action names an owner skill. Approval gates respected