This is a bounded teaching exercise about AI-assisted keyword research. Use only the public context below. Do not browse, use tools, consult other conversations or use personal memory. Return hypotheses, not claimed search data. Do not invent search volume, difficulty, rankings, customer experience or performance.

Site context: SEO Informatica publishes practical guidance for service businesses and SEO practitioners. The intended research market is US English. We want useful reader tasks, evidence checks and page decisions, not one new article per phrase.

Public existing-page inventory, summarized for this exercise:
/what-is-ai-seo/ explains two meanings: using AI for SEO work and visibility in AI search.
/ai-seo-tools/ compares tool categories and selection criteria; it excludes writing assistants.
/ai-seo-companies/ helps readers evaluate providers and scope.
/ai-seo-statistics/ explains research findings and their limits.
/intent-routing-101/ explains matching reader tasks to page roles and avoiding overlap.
/answer-block-writing-framework/ explains answer structure and supporting proof.
/ai-visibility-benchmark/answer-source-format/ explains presenting sources with answers.
/ai-visibility-benchmark/monthly-ai-citation-tracking/ explains repeatable answer sampling and recording citations.
/chatgpt-search-readiness/ explains site readiness for ChatGPT Search.
/ai-visibility-cost/ covers the scope and cost of AI visibility work.
/ai-visibility-timeline/ covers how to interpret timelines for visibility work.

Seed phrases selected by the researcher, not a demand dataset:
ai keyword research
ai generated content seo
ai seo tools
ai seo companies
ai seo statistics
ai citation tracking

Task: produce exactly 30 distinct candidate search phrases across these seeds. Include all six seeds unchanged. For each phrase give a concise likely reader task, a provisional action (existing-page review, possible new article, or investigate), a relevant existing page if one fits, and the single most useful evidence check before accepting the action. You may recommend a new article only if the reader job is not adequately owned in the supplied inventory. Related variants can share one article. Do not call a phrase popular, trending, easy to rank or high intent without evidence. Do not make any new-page decision final.

Return only a Markdown table with columns: ID (K01-K30), candidate, likely reader task, provisional action, existing page, evidence check. No introduction or conclusion.