Why your business is missing from AI answers

It is almost never one thing. Seven causes account for most of it, and they are not independent: the first three are structural, and they gate everything you might do about the rest. That is why the order matters more than the effort. Each cause below carries the check that confirms it. The checks are binary, they need no tools, and together they take about twenty minutes.

Your services are pictures

Check this one first, because it is the one owners least expect. The page looks finished to them. Treatment menus set as graphics. A price list exported as an image from a design file. A services grid built inside a page builder that renders only after a script has run.

Google’s own documentation on images describes how it works out what a picture is about: “Google uses alt text along with computer vision algorithms and the contents of the page to understand the subject matter of the image.” It adds that it draws on “the content of the page, including captions and image titles.” Read what that list is made of. Alt text, captions, the page’s own words. All of it is text sitting around the picture.

Be precise about what this does and does not prove. The documentation nowhere claims the words inside a JPEG are unreadable, and machine transcription of images is real and improving. What it describes is a system that learns what an image means from the writing near it. A price that exists only as pixels, on a page whose text never states that price, has nothing near it doing that job.

The check: open your services page, select the text with your cursor, and try to highlight a service name and its price. If you cannot select it, do not assume something else will read it.

Your listings contradict each other

An old suite number on a directory. A previous trading name on a professional association. A phone number from before the last move. Each is trivial alone. Together they teach every system reading you that facts about you are unreliable, and an unreliable fact is one a cautious model leaves out of the answer.

Google states the consequence plainly in its guidance on how local results are ranked: “Businesses with complete and accurate info are more likely to show up in local search results. If your business info isn’t accurate, your Business Profile might not show up for relevant searches in your area.”

The check: search your business name plus your phone number, then your business name plus your street. Read the first two pages for anything that disagrees with what you would say today.

Nobody else has ever written about you

A business that appears only on its own website has no corroboration. Nothing anywhere confirms that the claims on the site are true, or that the business is of any consequence in its town.

This is not a subtle signal, and Google names it in the same guidance, listing among the things prominence is based on “how many websites link to your business.” Independent mentions carry weight precisely because they are the hard ones to manufacture. A supplier’s page, a local paper, an association roster, a chamber listing, a real writeup by someone who is not you.

The check: search your business name in quotes, excluding your own domain. If everything on the first page is still you talking about you, this is your cause.

Your reviews have stars but no words

Here the honest version splits in two, and the split is worth knowing because most advice collapses it.

What Google documents as a ranking input is volume and score: “More reviews and positive ratings can help your business’s local ranking.” Not the wording. If ranking were the only concern, a five-star average built from two-word reviews would be doing its job.

The wording matters for a different reason. When an AI assistant describes you rather than ranks you, review text is the only place a specific service, a specific person, or a specific outcome is named in somebody else’s words. Stars can be averaged. They cannot be quoted. A model with forty five-star ratings and no sentences has been told that people were pleased and nothing whatever about what for.

The check: read your last twenty reviews and count how many name a specific service, person, or outcome. If it is under a third, you have a rating and not a description.

You are described in words nobody searches

Coined service names, internal shorthand, and brand vocabulary invented in a workshop. If a treatment is called something only you call it, no question anyone actually asks will match it. The name is not the problem. The name being the only word on the page is the problem.

The check: take your three most invented-sounding service names and search each one alone. If the results are all you, or all something unrelated, the term has no retrieval surface. Keep it as a name, and put the buyer’s word beside it.

Your site is slow, blocked, or partly invisible

Two separate failures live here, and they need separating because the second one is usually an accident.

The first is content that exists only after a script runs. When Vercel and MERJ examined crawler behaviour across their network, they found that the major AI crawlers fetch JavaScript files but do not execute them. Whatever your site assembles in the browser is not what those crawlers took away.

The second is a robots file that blocks more than the owner intended, and it is worth reading OpenAI’s crawler documentation before touching one, because the crawlers are not a single thing. OAI-SearchBot “is used to surface websites in search results in ChatGPT’s search features.” GPTBot “is used to crawl content that may be used in training our generative AI foundation models.” They are set independently, and OpenAI spells out both the option and the cost: “a webmaster can allow OAI-SearchBot in order to appear in search results while disallowing GPTBot,” but “sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers.”

That distinction is exactly what a plugin offering one switch labelled block AI throws away. An owner who wanted to protect their writing from training often finds they have also removed themselves from the answers. If you do change it, OpenAI notes it can take around twenty-four hours for the change to register, so do not judge the result the same afternoon.

The check: load your site with JavaScript disabled and see what survives. Then open your robots.txt and read it line by line, every user-agent, not just the first.

There is nothing on the site worth quoting

This is the one that separates present-but-thin from present-and-described. If every page is a brochure, there is nothing an AI assistant can lift with your name attached to it. Answers get quoted. Adjectives do not.

The check: pick a real question a customer asks you on the phone every week. Search your own site for a page that answers it in plain language. If there is not one, that is the gap, and it is the only cause on this list that no amount of technical work closes.

Fix them in this order

One to three first, because they are structural and they gate everything else. Then four and five, which are cheap and steady and mostly a matter of asking. Six when something is genuinely broken rather than as routine maintenance. Seven is the long game and the only one that compounds.

Two honest notes. Some of this sits outside your control: an AI assistant’s refresh cycle, a directory that will not correct itself, a competitor with fifteen years of press. And no order of operations guarantees a placement. Anyone promising one is selling a result they have no mechanism to deliver.

What the order buys you is narrower and worth more. It stops you spending a year on the seventh cause while the first three are still live, which is the ordinary way this goes wrong. That sequencing is most of what an audit is for.

Questions people ask