How AI Search Decides Which Businesses to Recommend, and What You Can Control

The short answer: AI search tools such as Google's AI Overviews and AI Mode, Microsoft Copilot and ChatGPT search build their answers from pages they can crawl, index and trust. Whether your business gets cited comes down to three things. Your pages have to be crawlable and quotable. Your answers have to be more specific than what everyone else publishes. And your business information has to agree everywhere it appears. No special file, schema type or writing style switches AI visibility on.

You may have heard this work called AEO (answer engine optimization) or GEO. Google's guide to optimizing for generative AI features, last updated in July 2026, describes both as still SEO. We agree, with one caveat: the bar for what counts as a useful page has moved, and most service-business websites were not built to clear it.

How AI answers choose their sources

Google's AI features work by retrieval. The model does not answer from memory alone. It runs searches, pulls pages from the index and writes a response grounded in what it found, with links back to those pages. Google calls one part of this process query fan-out: a single question is expanded into several related searches that run at the same time.

Take a property manager in Edmonton who asks whether an AI assistant can handle after-hours maintenance requests. Behind that question sit several others. What does it cost? Where does tenant information go? What happens when a request is urgent? Will it work with the maintenance software already in place? The pages that end up in the answer are the ones that deal with those follow-ups plainly. A service page that says "we build custom AI solutions" and stops there answers none of them.

This is why thin, generic pages lose ground in AI answers faster than they did in traditional results. When a system assembles one response from many sources, a page that repeats what fifty other pages already say gives it nothing to use.

What to ignore

A lot of AEO advice sells work you do not need. Google's guide addresses most of it directly:

• llms.txt and other "AI files." Google says it does not use them. Adding one will not hurt, and it will not help you in Google Search.
• Breaking content into tiny chunks. Google says its systems understand pages that cover several related topics. Write for a reader, use clear headings and leave it there.
• Special schema for AI. No structured data type is required for AI Overviews or AI Mode. Structured data is still worth doing properly for rich results, as long as it matches what is visible on the page.
• Rewriting pages for every phrasing of a question. These systems handle synonyms well. Producing near-duplicate pages for each variation can run into Google's scaled content abuse policy.
• Manufactured mentions. Paying for or seeding brand mentions across the web is not an effective shortcut.

If a proposal leads with any of these, ask what it expects to change and how you would measure it.

Five things a service business can control

1. Answer the questions people ask before they contact you

The most citable page on a service business's site is usually the one that answers pre-sale questions honestly: who the service suits and who it does not, what is included, what you need from the client, how pricing is scoped and where projects tend to get stuck. This is what Google means by non-commodity content. Nobody else can write your scoping rules, your delivery process or the trade-offs you have seen on real projects.

A practical test: list the ten questions your team answers most often on discovery calls. If a stranger could not find a direct answer to at least seven of them on your website, that gap is your content plan.

If your website has a chat assistant, its conversation history is the most honest version of that list, because it records how customers actually phrase their questions. Lumi, the assistant on our own site, flags question patterns and unanswered topics for our team to review, which shows us what the website still needs to explain. A well-scoped AI assistant with clear limits and a human handoff can do the same for your business.

2. Make sure your pages can be crawled, indexed and quoted

Google's baseline is simple. A page has to be indexed and eligible to show a snippet before it can support an AI feature. Three problems commonly block this without anyone noticing:

• Important content that only appears after JavaScript runs, or that sits inside images, PDFs or a chat widget instead of the page's HTML.
• Leftover noindex, nosnippet or restrictive max-snippet settings, often carried over from a staging site or an old plugin.
• Robots rules that block crawlers you actually want. OpenAI, for example, separates OAI-SearchBot, which powers ChatGPT search, from GPTBot, which collects training data. Blocking OAI-SearchBot keeps you out of ChatGPT search results. Blocking GPTBot does not.

3. Keep your business facts consistent

For local and service searches, Google's guide points to Google Business Profile as a source for AI responses. Your business name, service area, phone number, hours and service descriptions should match across your website, your Business Profile and the directories that matter in your industry. When those sources disagree, a system has to guess, and it may guess wrong or leave you out. If the same details live in a booking tool, a CRM and several listings, the lasting fix is usually to connect those systems so a change is made once, not to remember to update five places.

4. Treat old URLs as assets

When we rebuilt our own website this month, we checked the 20 article URLs from our previous site that search engines still knew about. Eighteen returned a 404 error. Rather than send them all to the homepage, we mapped each one to the current page covering the same topic with a permanent redirect, and left unknown paths returning a genuine 404. Google may treat redirects to unrelated pages as soft 404s, so the mapping only helps when every destination is a real equivalent.

If your site has been rebuilt in the last few years, run the same check. It is common to find years of earned links pointing at pages that no longer exist.

5. Show evidence people can check

Specific, verifiable details make a page more useful to a reader and safer for a system to quote: a dated example, a described process, a stated limitation. Project write-ups shared with client permission, photos of real work and plain statements of what a service does not include all do more than another paragraph of general claims.

How to measure AI visibility today

Measurement improved a great deal this year. You no longer need third-party estimates to get a starting point:

• Google Search Console. The Generative AI performance report became available to all sites on August 31, 2026. It shows impressions from AI Overviews and AI Mode by page, country, date and device. It does not include clicks yet, and smaller sites may see no data until they have enough impressions.
• Bing Webmaster Tools. The AI Performance report shows citations across Microsoft Copilot and Bing, the grounding queries behind them and your share of citations for each query. It is currently the more direct view of which questions your content is answering.
• Your enquiries. Impressions are an early signal. Qualified enquiries and booked calls are the outcome. Ask new leads how they found you and record the answer, because visits from AI assistants can be hard to see in analytics. The same discipline applies if you run Google or Meta campaigns: measure paid advertising by the enquiries it produces, not the clicks.

One caution. Search Console now has a setting that lets a site opt out of Google's generative AI features. Google says the setting is not a ranking signal for regular search, but opting out removes your site from AI Overviews and AI Mode. For most service businesses, that is visibility worth keeping.

A 30-minute check you can run this week

1. Search your three most valuable service queries in Google AI Mode, ChatGPT and Copilot. Note who gets cited and what their pages offer that yours do not.
2. Open your main service page and confirm the key answers are in plain text on the page, not only in images, PDFs or a chat window.
3. Check that page for noindex and nosnippet, and check robots.txt for blocked search crawlers.
4. Compare your business name, phone, hours and service area on your website and Google Business Profile.
5. List your ten most common pre-sale questions and mark which ones your site answers directly.
6. Open the AI reports in Search Console and Bing Webmaster Tools and record today's numbers as your baseline.

Where Lumin fits

Most of what decides whether AI search recommends a business happens away from the search box: how clearly the business answers questions, whether its systems keep information consistent, and whether it measures what turns into real work. That is the part we build.

• AI assistants that answer routine questions from an approved guide, capture enquiries and hand off to a person when judgment is needed, while showing you which questions your website still needs to answer.
• Connected workflows and custom applications that keep business details, bookings and enquiries in step across the tools your team already uses.
• Paid advertising support measured by qualified enquiries, so paid and organic demand can be compared on the same terms.

If you want a practical read on where your business stands, run the 30-minute check above and book a discovery call to talk through what you found.