Method
The exact prompts, and the exact checks
Two agencies offering “AI visibility” are not offering the same thing. This page is the whole method: what we ask, how each answer is scored, and the ten technical checks behind it. Show it to a competitor's proposal and compare.
20 prompts
What we actually ask
The same set every month, frozen per account, so month two is comparable to month one. Placeholders are filled with the business's real category, city and service names.
Category and city · 8 prompts
The prompts a buyer types when they have a need and no brand in mind. These are the ones that matter, and the ones almost every business fails.
best {category} in {city}The most common commercial prompt in almost every local category.
who is the best {category} near {city}Tests whether the model can resolve the area at all.
top rated {category} in {city}Pulls on review data, which is why corroboration is a signal.
{category} recommendations in {city}Phrased as a request for a shortlist, not a search.
affordable {category} in {city}Price-led; models usually name whoever states prices plainly.
{category} open now in {city}Hours in structured data decide this one.
most trusted {category} in {city}Surfaces who has third-party mentions, not who has the best homepage.
{category} for {use_case} in {city}Your specific job, e.g. non-owner SR-22, emergency callout.
Question form · 4 prompts
Longer, more conversational prompts. These reward pages that answer a question in the first paragraph instead of making the reader scroll.
How much does {service} cost in {city}?Answerable only if the site states a price or a range.
What should I look for when choosing a {category} in {city}?A buying-guide prompt; the answer cites whoever published criteria.
Who do you recommend for {service} in {city}?Direct recommendation request — the hardest and most valuable.
Is {business} a good {category}?Brand-plus-judgement; shows what the model believes about you.
Spanish · 4 prompts
Run where the market speaks Spanish. A business can be named in English and absent in Spanish — usually because the Spanish pages are machine-translated or missing.
mejor {category} en {city}Direct equivalent of the primary category prompt.
¿quién hace {service} en {city}?Recommendation request in Spanish.
{category} de confianza en {city}Trust-led variant.
¿cuánto cuesta {service} en {city}?Price question; exposes untranslated pricing pages.
Brand · 4 prompts
Prompts naming the business. These are the control group: if a model cannot answer these, the problem is entity resolution rather than competition.
What is {business}?The most basic test of whether the entity resolves.
What services does {business} offer?Tests whether service data is readable.
Where is {business} located and what are its hours?Tests location and hours in structured data.
What do reviews say about {business}?Tests corroboration; often the most revealing answer of the twenty.
Scoring
How an answer is scored
Every answer to every prompt is scored 0 to 3. The score is what makes months comparable; the transcript is what makes it believable.
0
Absent
Not mentioned in the answer.
1
Named only
Listed as an option with no description.
2
Described
Named with a description of what it does.
3
Described and cited
Named, described, and supported by a citation or link.
10 checks
What decides whether a model can name you at all
Before competition matters, a model has to be able to resolve, read and trust the business. These ten checks are pass or fail, and each failed check maps to a specific fix.
Organization or LocalBusiness markup exists and validates
Pass: A validator returns zero errors, with name, address, phone and URL.
Fail: No markup at all, or markup with placeholder values a validator rejects.
sameAs links the site to Google Business Profile and social profiles
Pass: Every profile the business actually maintains is listed on the site.
Fail: No sameAs, so the model treats each profile as a separate business.
Name, address and phone are identical everywhere
Pass: One address and one phone number across site, GBP and directories.
Fail: An old number in the footer, or a suite number present in one place and missing in another.
areaServed names the cities actually served
Pass: A list of cities or a defined service radius.
Fail: A whole state, or a country — which tells a model nothing about who you serve.
Service pages carry Service and FAQPage markup
Pass: Each service page declares what it is and answers real questions in schema.
Fail: One generic Organization tag sitewide.
Pages answer the question in the first 40–60 words
Pass: The direct answer is in the opening paragraph, detail below it.
Fail: A hero sentence about passion, with the answer three scrolls down.
AI crawlers receive HTML, not an empty JavaScript shell
Pass: GPTBot, ClaudeBot and PerplexityBot user agents all return the page content.
Fail: A client-rendered page that serves bots an empty body.
Reviews are published, attributed and current
Pass: Recent reviews with text, on-site and on the profile.
Fail: No reviews, or five reviews all dated the same week in 2021.
Canonicals, robots and indexation are clean
Pass: Self-referencing canonicals, no stray noindex, sitemap submitted.
Fail: A staging noindex left live, or two URLs serving the same page.
Contact facts are machine-readable
Pass: Hours, service area, phone and email stated as text and in schema; llms.txt where useful.
Fail: A contact page containing an image of an address.
What happens if the audit finds nothing
It happens. Some businesses are already named in their category, described correctly and cited; the twenty prompts show it and there is nothing worth selling. In that case you get the report saying exactly that, and we do not pitch the sprint.
- The audit is free, and it is yours whether or not you hire us.
- No retainer, no onboarding fee, no discovery-call gate.
- If we cannot fix something — a platform with no editable head, no access to the profile — we say so before you pay.
- No claim of guaranteed recommendations. We do not guarantee AI recommendations. Nobody honestly can.