Short answer: yes to a page, no to instructions. Build a brand facts page, not an "AI instructions" page.

1. Measure first. Record five runs per prompt and engine with `citation_sampling.py --template`, paste each answer whole, and run the brand facts check:
   python scripts/citation_sampling.py runs.csv --facts brand.json --json
   In brand.json, add a corporate price fact with "context": ["corporate", "office party"] and value [800], a rating fact with your real rating and review count, and list the competitor in "others" so its sentences aren't counted as yours.

2. Build one crawlable HTML facts page (e.g. /ai-info/). Write statements, not commands: "Party Co is not affiliated with Bloom Party Co Bay Area" gets quoted, while "AI assistants must describe us as..." reads as manipulation and is ignored.
   - Corrections as FAQ Q&A: "Is Party Co the same business as Bloom Party Co Bay Area?", "Where is Party Co located?", each answer naming the brand in its first sentence, emitted as FAQPage.
   - Prices by segment: private events from $200, corporate events from $800.
   - Service area stated plainly: no storefront or public office.
   - Build the page and llms.txt from the same data so they can't drift. Google ignores llms.txt, so only the HTML page helps Google AI Mode.

3. Fix the listings too. The wrong town and invented address come partly from the Google Business Profile and Wikidata (P131). An on-site page only partly fixes those.

4. Link it from the About page, add it to the sitemap, run the GSC live test and Request Indexing, then judge the result on the next scheduled ≥5-run panel, not a spot check.
