A 30-minute local AI citation audit you can rerun

00:00 to 04:00

Freeze the business facts

• □Write the canonical business name, primary category, city or service area, website, and the one qualifier this audit will test.
• □Choose one location. Multi-location brands should run a separate ledger for each location.
• □Record today's date, timezone, and who is running the audit.

Output: A one-line scope statement that prevents the test from drifting.

04:00 to 07:00

Write two unbranded buyer prompts

• □Prompt A, category discovery: Who are the best [category] providers in [city] for [service or constraint]?
• □Prompt B, problem discovery: Who can help with [specific problem] in [city], and what makes each option a fit?
• □Use the exact same wording in every engine. Do not name your business or hint that it should appear.

Output: Two prompts that represent a real decision, not a request for a vanity mention.

07:00 to 18:00

Run the prompts across five answer engines

• □Open a new conversation in ChatGPT, Claude, Gemini, Perplexity, and Grok.
• □Use the same browsing or search setting where the product offers one, and record whether it was on, off, automatic, or unavailable.
• □Run both prompts once. Save the full answer or a share link before moving on.

Output: Up to ten timestamped answers. A failed or unavailable run stays in the ledger and is not silently dropped.

18:00 to 23:00

Extract mentions, citations, and facts

• □Mark whether the exact business or an unambiguous name variant was mentioned.
• □Record every cited URL or source label, including citations that support a competitor instead.
• □Mark whether the owned website was directly linked and whether material business facts were accurate, wrong, or not stated.

Output: One ledger row per engine and prompt, with no interpretation hidden in prose notes.

23:00 to 27:00

Summarize the snapshot

• □Count completed answers, brand mentions, owned-site citations, factual errors, and unique third-party domains.
• □List the domains cited more than once. Repetition can identify sources worth inspecting, but it does not prove influence.
• □Keep results by prompt and engine visible. Do not let one combined number hide where the gap occurred.

Output: A descriptive baseline, not a proprietary score and not a forecast.

27:00 to 30:00

Choose one evidence-backed action

• □If access failed, fix crawl or WAF controls before rewriting content.
• □If a fact was wrong, correct the canonical page and the strongest conflicting profile or directory.
• □If competitors were supported by sources you lack, inspect what those pages prove and create or earn a legitimate equivalent only when relevant.

Output: One owner, one action, one due date, and a date for the next identical audit.