Audit Whether AI Recommends Your Local Business

A local shortlist is short, and there is no second page to fall back to. This workflow reads the businesses named and the domains cited in your own monitored local answers, ranks the platforms by how often they appear, and stops there.

The GEO problem it solves

Local advice assumes one platform decides everything: complete the profile, gather reviews, push listings. AI answers cite review sites, directories, community threads, local press and first-party pages, and that mix moves by provider and by category.

Nothing here reads your listings, so this run cannot audit hours, addresses or structured data. It reads which businesses recur locally and which domains get cited, and returns a short ordered list of platforms to open by hand.

How the workflow breaks down

1. Find the prompts that carry local intent. list_topics returns every configured topic with its label, its monitored prompt count and a case-insensitive label search; list_tags does the same for tags. You need one of those IDs, because the topic and tag filters on get_answers take a UUID, not a label. If neither list holds a local group, drop the filter: get_answers returns the prompt text on every row, so you can pick local answers by reading them.

2. Pull the local baseline. get_local_insights returns businesses observed over the period with category, rating, review count, mention count, average position, website URL and a competitor flag, plus a category distribution. It caps the list at 20 and ties no business to a provider or prompt. Match yourself by name and by URL: a name variant reads as an absence.

3. Open the answers. get_answers on the chosen topic or tag, then get_answer_details on each answer kept, returns the provider, the names recorded as competitor mentions with their position, and every cited source with its domain and rank. Only this step reaches a provider.

4. Type the cited domains. get_source_domains filters on a domain string and returns the type held for it: institutional, commercial, media, blog, forum, social, educational or other. Its own ranking covers all monitoring with no local, topic, provider or date filter, so use it as a lookup, not as a local source map.

What to do with the output

Order the cited domains by how many sampled local answers cite each, highest first. Work the top three: check that your entry exists, that the details match, that the reviews are current. A domain cited by exactly one sampled answer goes in a log, not the backlog. If two tie, take the one where you can edit your own entry today.

Businesses that outrank you in the baseline are reading material, not tasks: the tool never links a business to the answer that named it.

Record before you change anything: the business list, the sampled answer IDs, the providers covered and the sample size. Without it a rerun compares nothing.

The Complete Prompt

I want to see which businesses and which platforms show up when AI assistants answer local questions in my category. If I monitor more than one brand, list them and ask me which one before pulling any data.

1. Run list_topics and list_tags. Show the labels with their prompt counts and ask which group holds my local prompts. If neither list contains one, say so and move on with no filter.
2. Run get_local_insights over 30 days. Report only the returned fields: name, category, rating, review count, mention count, average position, website URL, competitor flag, and the category distribution. Look for my business by name and by URL and say which match you used. If the list is empty, retry once over 90 days; if it is still empty, tell me the local surface holds no observations for this brand and end the run rather than continuing on inference.
3. Run get_answers, filtered on the topic or tag ID from step 1 if I gave you one, unfiltered otherwise. If unfiltered, read the prompt text on each row and keep only the rows with local intent. Tell me how many you kept. Under five, say the sample is too thin to rank, list what you found, and leave it unranked.
4. Open get_answer_details on every answer you kept. Record the provider, the names recorded as competitor mentions with their position, and every cited source domain with its rank. Do not infer hours, addresses, phone numbers or listing status: none of that is in the response.
5. Count how many sampled answers cite each domain. Run get_source_domains once per distinct domain using the domain filter, and report the type it returns instead of guessing. Give me a table sorted by that count, with the sample size.

Finish with the three most cited domains in order, and for each the one thing to verify by hand: that my entry exists, that it is accurate, that its reviews are recent. List anything cited once separately, labelled observed but not actionable. Do not tell me that fixing a listing will change a recommendation.