A growing share of buying decisions now begins as a question to an assistant rather than a search. Someone asks for a good roofer in their city, or a furnished rental near a hospital, and gets back two or three names with a sentence about each. No results page, no ten blue links, no second page to scroll to.
That changes the math. Ranking fourth on Google still gets you seen. Being the fourth option an assistant considered gets you nothing, because it only named three.
Where the answers come from
It helps to know the plumbing. Assistants answering questions about real businesses generally do one of two things. They search the live web and summarize what they find, which for several major assistants means leaning on Bing's index. Or they draw on what was in their training data, which skews toward well-established, frequently-referenced businesses.
Both paths reward the same things: being indexed, being readable, and being described consistently in enough places that the model is confident about the facts.
Being readable is not the same as being online
This is where most businesses lose. A website built entirely as images, or as a page builder's nest of unlabeled containers, is technically online and practically illegible. When a system extracts meaning from that page, it gets very little.
Clean, semantic markup fixes it. Real headings in order, actual paragraphs, descriptive link text, alt text that describes images. The same structure that helps screen readers helps machines, for the same reason.
Structured data states the facts plainly
Schema markup is the part of a page that tells a machine exactly what it's looking at: this is a business, here's the name, address, phone, hours, service area, here are the services, here are the reviews and the average rating. Without it, a system has to infer all of that from prose and may get it wrong. With it, there's nothing to guess.
For local businesses this is the biggest technical win available, and it's invisible to human visitors.
Answer real questions in real sentences
Assistants quote text that answers a question directly. A page that says most business websites take four to eight weeks gets cited. A page that says timelines vary depending on your unique needs does not, because there's nothing in it to quote.
Write the answers you'd want repeated. Put them in plain sentences near a heading that asks the question. An FAQ section with genuine questions and specific answers is disproportionately useful here.
Consistency across the web
Models weigh corroboration. If your business is described one way on the website, differently on the Google profile, and with an old address on three directories, confidence drops. Same name, same address format, same phone, same description of what you do, everywhere.
Don't block the crawlers
Check your robots file. Plenty of websites quietly block GPTBot, ClaudeBot, PerplexityBot, or Google-Extended, sometimes because a security plugin added the rule and nobody noticed. Blocking them means opting out of every answer those systems produce. Unless there's a specific reason to stay out, let them in.
What you can't control
No one can promise placement in an AI answer. There's no submission form and no ranking dashboard. What's controllable is being indexed, being legible, being structured, being consistent, and having genuinely useful answers on the page. Businesses that do those things get recommended; businesses that don't are invisible to the question entirely.