How AI Assistants Use Public Business Pages

Increasingly, people ask questions of assistants instead of typing keywords into a search box. "Is there a bookshop in Hartley open on Sunday?" "Who makes sourdough near me?" "What does this company do and who runs it?" The answers come as sentences, not lists of links, and they are assembled from public sources.

If you run a business, that raises a practical question: what do these systems find about you, and how do they decide what to say? This article explains, in plain terms, how assistants use public business pages, and what you can do to make the answers accurate.

How the information gets there

There is no single mechanism. Assistants differ in how they are built and what they can access. But the broad picture has familiar steps, and Google's description of how its search works is a useful frame: pages are found by automated crawlers, their content is analysed and stored in an index, and when a question arrives, relevant material is selected and served, shaped by factors such as location, language and device. Assistants built on search, or on similar collections of documents, follow a similar path of find, understand and select.

Some assistants also connect directly to data sources through open standards. The Model Context Protocol, for example, is described in its documentation as an open-source standard for connecting AI applications to external systems, such as data sources, tools and workflows, and is likened to a USB-C port for AI applications. That makes it easier for assistants to read structured data from places that offer it, rather than guessing from web pages.

Whatever the route, the quality of the answer depends on the quality of what is available.

What assistants tend to rely on

When asked about a business, an assistant is likely to draw on a mix of sources.

The business's own website. The about page, contact page, service pages and any structured data.

Directory profiles. Pages like those on this site, with fact boxes, categories and descriptions.

Official records, such as company registers, where they are publicly accessible.

Maps and local listings, for addresses, hours and categories.

Reviews and mentions, on other sites.

News and articles.

Social and professional profiles.

Not all sources are equal, and not all are fresh. An assistant may pick up an old address from an abandoned listing, or a description from a third-party page. That is why consistency across sources matters.

What makes a page easy to use correctly

Several qualities help.

Clear, self-contained statements

An assistant often extracts a sentence or two. If the sentence only makes sense in context, it may be misquoted. "We have been doing this since 2011" tells nobody who "we" are. "The Hartley Bookshop has traded on Mill Lane since 2011" can be lifted out safely.

Plain language

Jargon, slogans and vague claims are hard to quote and easy to misread. Plain descriptions of what you do, for whom and where are easier to repeat accurately.

Facts in predictable places

A fact box with labelled lines, a consistent order and short values is easy to read for people and for machines. Put the name, what you do, where you are, how to contact you and when you started in the same place on every profile.

Dates

Facts change. A date on a fact lets the reader, and the system, judge how much weight to give it. "Opening hours, checked March" is better than "Opening hours."

Consistency

If your website, directory profile and map listing all say the same thing, systems can be more confident. If they disagree, the answer may be a guess, or may reflect whichever source was crawled most recently.

Structured data

Markup that labels the facts in a standard way, matching what is visible, removes guesswork. Schema.org's Organization type includes properties such as name, legal name, address, founding date and links to other pages that identify the entity. Google notes that organisation markup has optional properties and encourages including those that apply, prioritising those that indicate real-world presence. If your site supports it, add markup, and keep it in step with the page.

Sources and verification

Marks showing that facts were checked, with their meaning explained, give readers, and potentially systems, reason to trust them.

What you cannot control

It is worth being realistic. You cannot decide what an assistant says. Different assistants have different sources, rules and quirks. Some answers will reflect opinions in reviews, or the way a competitor describes you, or a mistake on a page you do not own. And answers can change from one day to the next as systems update.

What you can do is improve the odds that the available material is accurate, consistent and clear.

Practical steps

1. Ask the questions yourself

Try a few assistants and search tools with questions a customer might ask: who are you, what do you do, where are you, when are you open, how do I contact you. Note the answers and, where possible, the sources they cite. Do it every few months.

2. Trace the errors

If an answer is wrong, find where it came from. Often it is an old listing, an out-of-date directory profile or a page you forgot. Correct the source. Where the source is not yours, ask politely for a correction.

3. Write a clear about page and summary

Keep a short, factual description of the business, around a hundred words, and use it on your site and profiles. Our article on writing an about summary shows how.

4. Keep your fact box accurate and dated

Check it every quarter, and after any change. Show the last-checked date.

5. Use consistent names and details everywhere

The same name, address format, phone number and description across your site, directory profiles and listings.

6. Add structured data where you can

Match it to the page, test it and maintain it.

7. Make key pages accessible and fast

Real text, clear headings, good contrast and quick loading. Pages that cannot be read or that load slowly are less likely to be used.

8. Maintain a master record

A single source of truth for your facts, from which you update everything.

9. Respond to reviews and questions

Where assistants draw on reviews, a thoughtful reply can add context. Do not offer incentives for reviews, and do not write fake ones.

10. Be honest

Do not stuff pages with keywords, hide text or exaggerate. Google's guidance on helpful content stresses people-first content, clear information about who is behind it and honest sourcing. Honest pages are better for readers and, over time, for any system that tries to serve them.

Be careful with claims

Assistants may repeat what you say. If your page claims "the best bakery in the region", that may be quoted as if it were a fact, or may be ignored as marketing. If it says "a family bakery on Mill Lane, open since 2011", it is easier to quote and easier to verify. Make claims that you can support, and put evidence beside them.

Privacy and boundaries

Public pages are public. Think before you publish personal details, such as a home address or a personal phone number, since they can be repeated. Provide business contact routes instead. If you find an assistant repeating information that you did not intend to be public, remove it from the source and, where possible, ask for the cached copies to be updated.

A worked example

A small bookshop asks three assistants: "Tell me about the Hartley Bookshop." One answer gives the right address and opening hours. A second says the shop closes at five, which was true two years ago. A third mentions a second branch that closed.

The owner traces the errors. The old closing time comes from a directory profile she forgot; the closed branch appears on an old listing. She updates the profile and marks the old listing closed, then checks her own site, her map listing and her fact box. She adds a short, clear summary at the top of her about page, with her name, place, services and opening hours, and she adds structured data matching the visible facts.

Three months later, she repeats the questions. Two answers are now correct, and one still mentions the closed branch, so she reports the remaining listing. She puts the check in her quarterly audit.

Common mistakes

1. Assuming an assistant will find the right answer unaided.
2. Ignoring old listings.
3. Vague, marketing-heavy descriptions.
4. Inconsistent facts.
5. No dates.
6. Markup that disagrees with the page.
7. Never checking.

A checklist

• Ask assistants about your business each quarter
• Trace and correct wrong answers at the source
• Clear about page and short summary
• Dated, accurate fact box
• Consistent facts everywhere
• Structured data matching the page
• Accessible, fast pages
• Master record
• Honest claims

The developers page here describes the free API and MCP server that let assistants read listing data directly, the categories and search pages show how profiles are found and the submit page lets a business create an accurate profile.

Frequently asked questions

Do I need to do anything special for assistants? Not special. Clear, accurate, consistent pages help people and machines alike.

How long do corrections take to appear? It varies. Systems revisit pages at their own pace.

Can I ask an assistant's provider to correct an answer? Some offer feedback routes. Fixing the source is usually more effective.