What an AI Answer Engine Needs to Know About Your Business

Ask ChatGPT, Gemini or Perplexity about a local business and you will get one of eight questions answered, whether the business likes the answer or not. Hours, location, price, what to expect, who it is for, how to book, what people say, and what makes it different. The engine builds those answers from whatever it can read, and if your site does not say it clearly, a review site, a directory or a competitor says it for you.

I have written schema and built entity pages for local businesses since 2018, and this year I built a chatbot for a wellness business that answers exactly these questions on the website, in social bios and on an in-store QR code. Building it forced the question this piece answers: what does the machine actually need from you? The list is shorter than the industry makes it sound.

The eight questions customers ask

Customers ask the same eight questions of a local business, and answer engines are trained on those questions because they are the ones people type. When I logged the intents for the wellness chatbot, everything a visitor asked fell into one of these:

1. Are you open right now, and when?
2. Where are you, and how do I get there?
3. What does it cost?
4. What happens when I come in, and what should I bring?
5. Is this right for someone like me?
6. How do I book, and do I need to?
7. What do other people say about you?
8. Why you and not the place down the road?

Google’s own guidance for local businesses has always centred on the first three, and the sixth is where the money is. Questions four, five and eight are the ones almost no business writes down, which is why an engine invents the answer from reviews.

Where the engine gets its answers

An answer engine reads three kinds of sources: your site, structured data on your site, and everyone else. In a GEO audit I ran this month on a UK retail brand, only 5 of 52 citations across a set of buying questions came from the brand’s own domain. The rest came from magazines, comparison blogs, a forum thread and a review platform. The brand had no voice on the question that drives its returns rate, because the page that answered it rendered no text to a crawler.

That is the pattern to expect. If a question has a clean, crawlable answer on your domain, the engine tends to use it and cite you. If it does not, the engine assembles an answer from third parties and cites them. So the job is not “optimising for AI”. It is making sure each of the eight questions has a plain answer on a page the machine can read.

The structured data that answers them

Schema.org markup gives the engine the eight answers in a form it does not have to guess at. This is the set I put on every local business site, with the question each one answers:

• LocalBusiness (or the specific subtype such as HairSalon, Dentist or HVACBusiness) with name, address, telephone, geo, url and image. Answers question two, and anchors everything else.
• openingHoursSpecification inside it, including special hours for holidays. Answers question one, which is the most-asked and the most often wrong.
• priceRange on the business, and Offer or Service entries with real prices where you publish them. Answers question three. Businesses that hide prices get a price guessed from a directory.
• FAQPage with the four questions the industry never writes down: what happens on a first visit, what to bring, who it suits, and how booking works. Answers four, five and six.
• potentialAction with a ReserveAction or the booking URL. Answers six for engines that act, not only answer.
• AggregateRating and Review only for reviews shown on the page, sourced from where they were left. Answers seven without inventing anything.
• sameAs links to the Google Business Profile, Facebook, Instagram and any directory listing that matches your name, address and phone exactly. This is what ties the entity together so the engine knows the three listings are one business.

Google’s Search Central documentation covers the LocalBusiness and FAQPage types, and the schema.org vocabulary itself is the source for the rest. Validate with Google’s Rich Results Test and the Schema Markup Validator before you trust it.

The page most businesses are missing

The page most businesses are missing is the plain-language “what to expect” page, and it is the one the engine wants most. Hours and address exist everywhere. The honest description of a first visit, in the business’s own words, exists almost nowhere. For the wellness chatbot, the single page that answered “what happens when I come in, what do I wear, how long does it take, what does it cost” carried more of the bot’s answers than the rest of the site combined.

Write it as if you were answering a nervous first-time customer on WhatsApp. Short paragraphs, real prices, no marketing adjectives. Mark it up as FAQPage. Link it from the homepage, the Google Business Profile and the booking page. It is one afternoon of work and it is the difference between the engine quoting you and the engine quoting a review site.

How to check what the engine says

Ask the engines the eight questions yourself, and record the answers before you change anything. Open a fresh session in ChatGPT with search on, in Perplexity, in Gemini and in Google AI Mode. Ask each question the way a customer would, with your business name and town. Note whether you are mentioned, what the engine says, and which URLs it cites. That is a baseline you can compare against in 90 days.

What you are looking for is wrong facts and missing citations. Wrong hours, an old address, a price from three years ago, a “not suitable for” claim you never made. Each one maps to one of the eight questions and to one fix above.

Common mistakes

What I see most often is a beautiful site with every answer locked inside an accordion, a tab or a JavaScript widget that never sends text to a crawler. The wellness business’s original help content behaved that way, and so did the UK brand’s help centre. Open your page, view the source, and search for a sentence you know is on the page. If it is not in the source, the engine cannot read it either.

The second mistake is schema that contradicts the page. Hours in the markup that differ from hours in the footer. A price range in the markup with no prices on the page. Engines treat contradictions as noise and fall back to third parties.

The third is blocking the crawlers by accident. Check robots.txt for GPTBot, ClaudeBot, PerplexityBot and Google-Extended, decide deliberately which you allow, and write the decision down.

Key takeaways

• Answer engines answer eight questions about a local business; make sure each has a plain answer on a crawlable page.
• LocalBusiness, openingHoursSpecification, Offer, FAQPage, potentialAction, AggregateRating and sameAs cover all eight.
• The “what to expect” page is the one most businesses are missing and the one the engine wants most.
• Record what the four engines say today, fix, and compare in 90 days.
• Contradictions between markup and page, and content hidden behind JavaScript, are the two failures I find on almost every site.

The eight JSON-LD templates I use, including the LocalBusiness and FAQPage ones, are in the free toolkit. If you want to know what the engines currently say about your business, ask me for an audit.